Methodology for Predicting Channel Migration (2004)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

77 CHAPTER 3 INTERPRETATION, APPRAISAL, APPLICATION OVERVIEW This chapter presents interpretation, appraisal, and applications of the methodologies for predicting the rate and extent of channel meander migration presented in the Handbook. While several topics are extracted from the Handbook, much of this information was not necessary for the Handbook and appears only in this Final Report. The step-by-step illustration of the manual overlay and prediction technique is taken from Section 7 of the Handbook. It is presented here to provide the reader of this Final Report a better understanding of the methodology and of the capabilities and limitations of the aerial photograph comparison techniques. Additional interpretation and appraisal of classification and screening, regression analysis, and frequency analysis, which are summarized in the findings of Chapter 2, are also included. Details of the data collection effort, concepts (and assumptions) applicable to measuring meander migration, and appraisal of the results of testing the GIS predictor and State DOT testing of the methodologies are presented in this chapter. Finally, the implementation plan for the Handbook is developed. DATA COLLECTION Existing data was collected from a variety of sources and researchers. The data set compiled encompasses rivers in 23 states across the United States. The primary data upon which much of this project is based comes from work conducted by Brice (72). The Brice data used for this project consists of 811 bends at 82 sites on 59 rivers. Eight additional data sets containing 692 bends at 59 sites on 30 rivers were acquired from other sources. The Brice bankline tracings are based on aerial photography flown primarily in the late- 1930s to early-1940s and the mid-1960s to early-1970s. Attempts were made to acquire historic aerial photography from various agencies taken in the 1930s and 1960s for the data sets that had no historic bankline comparisons. Aerial photography from the 1990s and topographic maps were acquired for all the data sets. The 1990s photos and topographic maps were either downloaded from sites on the Internet or ordered from various agencies. All the photography and maps for each site were overlain, registered, and geo-referenced, and are compiled into a data base for future use. Updated discharge data was also obtained for each of the sites used in this project. The data sets are composed of data that is both specific and general in nature. Data measured as part of this project is site or bend specific. The general data acquired from the various sources was obtained at gaging stations or is associated with river reaches that may or may not be in close proximity to the study reaches. Therefore, problems, restrictions and limitations associated with each data set are included in the following discussion.

78 Brice Data Set The data set collected by Brice was part of an FHWA project to develop a simple method for determining relative stability of streams based on stream type (49). The Brice data set consists of morphometric data as well as aerial photos, maps, and bankline tracings for 200 stream reaches in the United States. The original data set is located at the U.S. Army Engineers Waterways Experiment Station (WES) in Vicksburg, Mississippi. Copies of the bankline tracings for the 82 meandering stream sites used in this project were acquired from WES. Brice compiled the morphometric data for each of the meander sites for projects conducted in the 1970s (49, 72, 213). The data set was compiled from existing information, field investigations and surveys, gage records, maps, and aerial photos. Under a research project at Johns Hopkins University, the data set was inventoried and additional data was derived for 133 of the Brice sites by Cherry et al. (48). In 1999, field crews consisting of personnel from WES, the USGS, and the University of Nottingham, UK, visited the Brice sites and obtained survey and sediment data. The field crews obtained bed and bank samples, measured cross-sections, photographed each survey site, and estimated the percentage of vegetative cover on the tops of the banks. However, the survey and sampling efforts were confined primarily to straight channel reaches with stable banks at each of the Brice sites. Schumm Great Plains Rivers Data Set Schumm collected data on 90 Great Plains rivers and streams in the late-1950s and 1960s. Much of this information is compiled in USGS Professional Paper 352-B (214). Planform, cross-section, sediment, and discharge data were collected and compiled for each of the sites. The data for most of the sites were collected near bridge crossings in close proximity to gaging stations where possible. A total of 256 bends at 20 sites on 9 rivers were selected from the Schumm Great Plains Rivers data set. The rivers and streams of the Schumm data set used in this project are located in Kansas, Nebraska, Montana, and Wyoming. Complete banklines from the early-1970s were available in an atlas compiled by the USACE Kansas City District (215) for the Kansas, Smoky Hill, Saline, and Solomon Rivers in Kansas. Aerial photographs taken in the late-1960s or early- 1970s were obtained for the remaining sites. Schumm supplied the raw sediment data for the sites used in this project. Minnesota Rivers Data Set MacDonald et al. (216) compiled data and analyzed meander problems on several streams in Minnesota. The analysis was conducted using aerial photos and topographic maps. The data set consists of aerial photos from the 1930s through the 1980s and general morphometric, sediment, and discharge data for 16 streams. A total of 15 sites encompassing 240 bends on 13 of the 16 available streams were evaluated. The authors (216) supplied the negatives of the aerial photographs and maps and the sediment data used in their analysis. Bankfull channel width and slope were based on measurements taken on topographic maps and stream depth was synthesized using accepted methodologies. Bed sediment samples were collected and the sieve analysis results on all but two rivers were available (216).

79 Ayres Associates Data Set Ayres Associates compiled a complete data set for both the middle Sacramento River (211, 212) in California and the lower Alabama River in Alabama (217). Bankline tracings of 25 bends of the Sacramento River from the 1930s and 1960s were compiled from an atlas of meander migration of the river from 1896 and 1981 (218), aerial photography, historic surveys, and maps. A detailed data set on a large part of the Sacramento River system was compiled as a result of work for the USACE Sacramento District since 1986. The detailed data and bankline comparisons for 17 bends on the Alabama River were compiled as the result of a project conducted for the USACE Mobile District in 1986. Other Data Sets Additional data was obtained for the Des Moines, Genesee, Powder, and Carson Rivers. Data was compiled on 34 bends of the Des Moines River downstream of Red Rock Reservoir in Iowa from bank erosion studies conducted by Odgaard (219) and the USACE Rock Island District (220). Mussetter Engineering, Inc. (221) provided detailed data on a reach of the Genesee River that includes 10 bends downstream of Mount Morris Dam near Geneseo, New York. Banklines from the 1930s and 1960s and morphometric data for 59 bends at 6 sites on the Powder River between Moorehead and Broadus, Montana were obtained from Martinson and Meade (102) and Martinson (127). The Nevada Department of Transportation provided historic aerial photographs and morphometric data for 6 bends on the Carson River near Weeks, Nevada. Acquisition of Updated Aerial Photos, Maps, and Discharge Data Aerial photos from the 1990s for all of the sites were obtained from three different sources. The first and primary source of 1990s aerial photos is the TerraServer web site operated by Microsoft Corporation (see Table 5). TerraServer, in partnership with the USGS and Compaq, provides free public access to a vast data store of maps and aerial photographs of the United States. Aerial photos and topographic maps at a wide variety of resolutions for most of the meander sites were downloaded free of charge from the TerraServer Web site. For those sites where no photographic coverage was available from TerraServer, 1990s aerial photos were ordered from either the USGS Earth Resources Observation System (EROS) Data Center in Sioux Falls, South Dakota, or from the USDA Farm Service Agency (FSA) Aerial Photo Field Office (APFO) in Salt Lake City, Utah. Both agencies have Web sites with searchable catalogs of available aerial photography. Topographic maps, either in paper or electronic format, can be obtained from a variety of sources. Paper copies of topographic maps can be obtained from the USGS or any commercial map supplier. Digital maps (DRGs, DEMs) can be downloaded free from the EROS Web site or purchased from commercial suppliers as well. Most digital maps are geo-referenced and can be loaded directly into GIS-based software. Portions of geo-referenced topographic maps can be downloaded free from the TerraServer web site and pieced together to form a complete map of a given area or used to fill in gaps. Care should be taken when using digital maps and photos because the geo-referenced coordinates and dimensions are usually in metric units while the contours and spot elevations shown on the maps may be in English units.

80 Up-to-date discharge data was obtained, where possible, for gaging stations in close proximity to all the meander sites. Mean daily discharge and annual peak discharge data is available free from the Water Resources web site of the USGS. The web site contains a searchable data base broken down by state, county, basin, and gage number. In some cases, gage data may be available from city, county, or state-operated gages, which may or may not be available via the Internet. The operating agency may need to be contacted to obtain the data. MEASUREMENT PROCEDURES Historic and recent aerial photos and topographic maps were obtained for all of the meander sites used in this project. Measurements of channel and valley morphometry were then made using a common CAD package and a specially developed bend measurement tool in the form of an ArcView® extension. Measurements Using MicroStation® The maps and photos were scanned where necessary and then compiled and geo- referenced to each other in Bentley’s MicroStation® based either on known UTM coordinates or on common reference points. The Brice bankline tracings were digitized using a large digitizing tablet. In addition to the banklines, reference points on the tracings that could be identified with points on the topographic maps or recent aerial photos were digitized. The tracings were then overlain with the maps and recent photos of each reach in the appropriate MicroStation® files. Because of distortion in the Brice bankline tracings and unrectified aerial photos, MicroStation® Descartes™ was used to "warp" the images so that common reference points matched exactly. The Brice bankline tracings of each year at a given site were registered with the topographic maps and aerial photographs of the site and saved as individual layers. Once the historic and recent aerial photographs for meander sites were registered in MicroStation®, the banklines of the channel were digitized and saved as individual layers. Bankline Delineation The accurate delineation of a bankline on aerial photos is dependent primarily on the density of vegetation at the top of the bank. Where vegetation becomes increasingly dense along a bank, small sections of the top or edge of the bank may be visible such that a line can be drawn connecting the sections. Often the top of the bank may be completely obscured by vegetation and one may be required to locate the top of the bank by approximation. This can be done by assuming that the bankline does not extend beyond the middle of each tree growing at the edge of the stream and then drawing the bankline on the riverward side of the crown of the tree. Shadows and changes in shading can often delineate the crown of a tree. If the density of vegetation along a stream is such that an accurate delineation of the top of the bank cannot be made, then the use of the channel centerline may be necessary. Brice (72) delineated the banklines for two different years at each of his sites by projecting the aerial photographs of each year onto a piece of paper and tracing the banklines on the paper. The banklines were overlain using common reference points. The scales of the

81 photos were estimated based on known or approximate distances between reference points. However, distortion of the photographs was a significant problem in some cases, especially with regard to the alignment of reference points on the photos from the two different time periods. In addition, dense vegetation obscured the banklines at some sites such that the bankline positions had to be estimated. In a few instances, exact registration of the 1990s aerial photos and topographic maps with the Brice bankline tracings was not possible because most of the registration points no longer existed or were obscured by dense vegetation. In these cases, the 1990s photos were matched with the topographic maps and the banklines were overlain as closely as possible on the aerial photo using what registration points and physical features were available. The Schumm (214) and Kansas City District COE data set consists of sites on the Kansas, Saline, Solomon, and Smoky Hill Rivers. The historic channel positions for these rivers were taken from an atlas of historic channel migration (215). It is not known if the channels are based on the waters edge under low water or high water conditions or on top bank locations. The most recent course associated with 1970s river position obscures the older courses of the river and makes the older banklines unusable. Therefore, the 1970s banklines for these sites were the only banklines used. The atlas consists of the historic banklines for each of the rivers overlain on geographic maps that were easily registered with the topographic maps and the 1990s aerial photos. The historic (1979) banklines for the Des Moines River are based on aerial photography included as plates in the Rock Island District report on bank erosion (220). Historic banklines for the Minnesota rivers data set are based on negatives of a pair of historic aerial photos for each of the sites (216). Parker provided the negatives of the aerial photos. The historic banklines for the Sacramento River were taken from an atlas of historic banklines compiled by the California Department of Water Resources (DWR) for a salmon spawning gravel study (218). Aerial photos of the Sacramento River taken in 1978 provided the base on which the banklines were overlain in the atlas. Martinson and Meade (102) compiled the historic banklines for the Powder River on USGS 7.5-minute topographic quadrangle maps. Historical (1950s and 1960s or 1970s) and 1990s aerial photos were acquired from the USGS EROS Data Center or the USDA Aerial Photo Field Office for the remaining sites. Valley Width and Orientation The valley width and orientation for all sites were based on the evaluation of both topographic maps and aerial photos. Where possible, the valley margins were defined based on a significant change in contour elevations at the intersection between a well-defined valley wall and the edge of the floodplain/meander belt. The valley margins may also be defined by confining terraces or by structural features such as levees and roadway embankments, which would restrict the active migration of the river. Lines were drawn across the valley between the valley margins and perpendicular to the general valley direction at regular intervals at each site. The valley width was determined based on the average of the cross-valley line lengths. A line was fitted to the midpoint of each of the

82 cross-valley lines to define the valley orientation line. Cross-valley lines and the valley orientation lines were drawn to reflect a significant change in valley direction. The valley orientation is measured counter-clockwise from a zero angle defined to be due east. Slope and Sinuosity The ideal method for measuring valley slope (Sv) requires at least two well-defined sequential contours crossing the valley floor perpendicular to the valley direction. The straight- line distance between these two contours generally defines the valley floor slope. However, contours rarely cross the valley floor in a perfectly straight line perpendicular to the valley direction, nor do they cross the valley floor without significant breaks or multiple changes in direction. Therefore, estimates of valley slope require either actual measurements of the valley slope or the identification of contour crossings of the channel. Valley slope can be estimated by dividing the change in elevation between contours by the measured straight-line distance between contour crossings of the channel along a line defining the general channel orientation (Figure 26A). However, rivers rarely follow the direction of the valley exactly, but instead often wander across the valley floor. In these cases, the valley distance should be measured along a composite line that defines the trends in the river’s orientation. For example, Figure 26B shows the valley slope determination based on the distance between contours as measured along a compound line defined by multiple channel trends. Channel slope (SC) can be obtained from surveys or by measurements taken from topographic maps. Channel slopes are obtained from maps by dividing the elevation change between successive contour crossings by the distance measured along the centerline of the river between the contour crossings. This provides the slope of the river for the date of the aerial photography used to make the map. If channel slope data is not available for other periods, the slope can be estimated by using the position of the contour crossings from the map projected onto the position of the channel for the period in question and then measuring the channel length between the crossings for that period. Channel sinuosity (P) is the ratio of the channel length (CL) to the valley length (VL) over a given reach. Some researchers prefer to measure sinuosity by measuring the straight-line distance from crossing to crossing between each bend. However, this can be problematic, especially where the channel has an extremely large number of highly contorted and compressed bends that wander back and forth across the valley floor. Therefore, for the purposes of this project, sinuosity is obtained by measuring the length of the channel orientation line (as shown in Figure 26) between two points and then measuring the channel centerline length between the same points. The channel orientation line defines the direction along which the channel is flowing across the valley floor and the length measured between the two pre-defined points is the valley length of the channel used in the sinuosity calculations.

83 Figure 26. Simple (A) and compound (B) channel orientations used in valley slope determinations. Measuring Meander Migration Before predictive tools for channel migration can be developed, one must be able to measure and describe channel migration. A standard approach for use in analyzing data sets must be developed and this approach should be adhered to for all subsequent measurements. The initial or existing meander bend should be represented by a starting point (upstream end), an ending point (downstream end), a location of the center of bend radius (bend centroid), an orientation with respect to a baseline (e.g., down valley direction), and an outside bank radius (RC). As shown in Figure 27, it can be assumed that the bend starts and ends at the flow crossing. Valley Margins Valley Orientation Line Cross-Valley Width Line Contour Crossing Contour Crossing Channel Orientation Line Valley Margins Valley Orientation Line Cross-Valley Width Line Contour Crossing Contour Crossing Channel Orientation Line (A) Simple (B) Compound

84 Extension (Across-Valley Migration) Translation (Down-Valley Migration) Expansion (Increasing Radius) Rotation All Four Modes of movement Flow Upstream End (crossing) Downstream End (crossing) Rc Channel Centerline Down-Valley Across-Valley Initial Bendway Figure 27. Measuring meander migration. Bend migration can be reasonably described by four modes of movement. Extension is across-valley migration and is easily measured at the bend centroid. Similarly, translation is down-valley migration and is also measured at the bend centroid. Expansion (or contraction) increases (or decreases) bend radius. Rotation is a change in the orientation of the meander bend with respect to the valley alignment.

85 A change in any of these four modes of movement results in a change in the location of the outer bankline. Combinations of these modes of movement would result in a wide variety of meander bend shapes through time. To apply this approach one must identify a valley orientation, locate the bend centroid, and measure the bend radius and orientation of the bend with respect to the valley. If this is performed for consecutive aerial photos, rates of change in each of the modes of movement can be computed. This type of geometric information is needed to graphically depict channel migration of individual bends. Predicting four modes of movement is a significant task for every bend of interest (Figure 27). However, actual bend migration is even more complex. For example, one part of the bend may be expanding faster or translating down-valley faster than another resulting in changes in bend symmetry. As a concession to practicality one must limit the number of modes of movement to the fewest possible. In the methodology developed, extension and translation are considered directly (as a vector sum). Expansion (a change in RC) is included as it could have a major impact on the location of the outer bank and because rates of migration appear to be correlated to RC/W (bend radius of curvature/width). If movement in these three modes can be predicted, the primary threats to a bridge, highway, or other facility will be established. Rotation is considered only indirectly as a component of the combined movement in the other three modes relative to adjacent bends. GIS Measurement Tool ArcView is a GIS and mapping software package developed by Environmental Systems Research Institute Inc. (ESRI). An ArcView extension, the meander bend Data Logger, is a GIS- based, menu-driven, circle-template methodology that was developed to streamline the measurement and analysis of bend migration data and aid in predicting channel migration. The Data Logger provides users with a quick and easy way to gather and archive river planform data. The physical banklines are represented by one or more ArcView themes. A theme is a set of geographic features in a view. A view is an interactive map that allows the user to display, explore, query and analyze geographic data in ArcView. The bend delineation points for each bend and each historical record are archived in individual themes to provide a graphical record of the user’s interpretation of each bend. For each river bend and each historical record, the Data Logger records various river characteristics (bend radius, bend center, river widths, bend wavelength, etc.). This data is organized by river reach and recorded in a table identified by the reach name. Figure 28 shows some of the measurements made using Data Logger. Data Logger requires the following actions to be performed at each bend for each historical record: 1. Locate registration points along the outer bank on a river bend. 2. Inscribe an arc of a circle using the registration points to describe the bend radius (Rc) and orientation. 3. Estimate the channel widths (W) at the bend apex and at the upstream and downstream crossings (ends of the bends). 4. Estimate the meander wavelength (λ) and bend amplitude (A).

86 Figure 28. Some of the bend measurements made using the Data Logger. Radius of Curvature The radius of curvature (RC) of the outside bankline is defined by setting 5 to 7 registration points along the bankline from the beginning of the bend to the end of the bend (Figure 28). Once the points are set, a circle is inscribed on the bend that best fits the registration points and best describes the bend. The centroid of the circle represents the centroid of the bend and the radius of the circle defines the radius of curvature of the outer bank. Meander Bend Orientation A line that extends from the bend centroid to a point on the outer bank arc midway between the upstream and downstream end points defines the bend orientation (Figure 28). Like the valley orientation angle, the bend orientation angle is measured counterclockwise from a zero angle defined to be due east. Meander Wavelength and Amplitude The meander wavelength (λ) is defined by identifying the upstream and downstream crossings, setting a point at the centerline of the river at the crossings, and then drawing a line between the points. The wavelength is twice the length of this line. The bend amplitude (A) is defined by a line drawn perpendicular to the wavelength line between the wavelength line and the outer bank at the bend apex. The bend apex is the farthest extension of the outer bank relative to the bend centroid. Best-Fit Circle Bend Centroid Bankline Registration Points Crossing Rc ½λ A

87 Channel Width Channel widths are measured from top bank to top bank at the crossings and at the widest point in the bend. The channel width of the crossing in the data set is the average of the channel width at both crossings. The widest point of the channel generally occurs near the bend apex. All the width measurements are made and recorded using the GIS measurement tool. SCREENING AND CLASSIFICATION PROCEDURES One of the objectives of this project was the development of a quantitative screening procedure to identify stable meandering reaches [e.g., Brice’s (49) equal width vs. random width comparisons]. In addition, a classification system for river/meander types was developed to support stratification of the data base for use in the quantitative multiple regression analyses. Screening Rivers are often categorized as either straight, meandering, braided, or anabranching. The range of channel types for meandering, braided, and anabranching channels is illustrated in Figure 10. The degree and character of sinuosity portions of Figure 10 are directly related to the objectives of this project, whereas the braided and anabranched portions are not. Therefore, the first step in screening is to identify and remove the braided and anabranched channels from the data base. Once a channel has been identified as meandering, it should be possible to identify the stability of meanders by their width characteristics (e.g., equiwidth vs. variable width). For the purposes of this project, a meandering river was considered sufficiently stable where it does not pose a threat to bridges or highway structures. Brice (49) attempted to discriminate qualitatively between very stable and less stable channels. He discovered that channels that do not vary significantly in width were relatively stable, whereas channels that were wider at bends were more active. Brice demonstrated this by plotting sinuosity against an erosion index (erosion rate in channel widths per year times percent of reach eroded times 100). High sinuosity equal-width streams were the most stable, whereas other equal-width streams of lower sinuosity were less stable, and wide bend streams had the highest erosion rates. This simple stratification of meanders will be of value to the bridge engineer as a screening procedure, allowing preliminary identification of meanders that are very stable. Brice's conclusions were validated by the expanded data base assembled for this project (see Chapter 2, Regression Analysis). Meander Classification Channel classification systems provide engineers with useful information on typical characteristics associated with a given river type and establish a common language as a basis for communication. Although classifications are useful for clarity of communication and as an index of the numerous types of channels that exist, it is the characteristics of an individual channel that are important in defining channel processes and response. Classification systems, alone, are of little value for deriving process significance or predicting channel response (see for example, (222)). When quantitative information about a river is available, classifications are only the first step in evaluating channel stability and predicting channel change.

88 A classification procedure modified from the channel pattern classification originally developed by Brice (72) (Figure 10) was developed for this project for both screening and classification (Figure 11). This approach was based on the evaluation of a number of classification schemes (72, 126, 201, 202, 203, 204, 205, 206, 207), and is the most applicable to project objectives. The following paragraphs provide a survey and appraisal of other classification procedures considered. Assuming a graded stream, one that is neither progressively aggrading or degrading, the type of sediment transported by the river has a major influence on channel shape, pattern, and gradient. Table 6 summarizes a classification of alluvial channels based on the relative proportions of sand and silt-clay transported by a stream. Based on studies of rivers on the great plains of the United States and the riverine plain of Australia, it was determined that suspended- load streams that transported very little bedload were narrow, deep, gentle, and sinuous whereas bed-load streams were wide, shallow, steep, and relatively straight. This classification relates channel characteristics to type of sediment load. During experimental studies it was further determined that valley gradient exerted a major influence on channel patterns. Table 6. Classification of Alluvial Channels (126). Channel Stability Mode of Sediment Transport and Type of Channel Bedload (percentage of total load) Stable (graded stream) Depositing (excess load) Eroding (deficiency of load) Suspended Load <3 Stable suspended- load channel. Width/depth ratio <10; sinuosity usually >2.0; gradient, relatively gentle Depositing suspended load channel. Major deposition on banks cause narrowing of channel; initial streambed deposition minor. Eroding suspended- load channel. Streambed erosion predominant; initial channel widening minor. Mixed Load 3-11 Stable mixed-load channel. Width/depth ratio >10 <40; sinuosity usually <2.0 >1.3; gradient, moderate Depositing mixed- load channel. Initial major deposition on banks followed by streambed deposition. Eroding mixed-load channel. Initial streambed erosion followed by channel widening. Bed Load >11 Stable bed-load channel. Width/depth ratio >40; sinuosity usually <1.3; gradient, relatively steep Depositing bed-load channel. Streambed deposition and island formation. Eroding bed-load channel. Little streambed erosion; channel widening predominant.

89 Figure 29 suggests that the range of channels from straight through braided forms a continuum, but experimental work and field studies have indicated that within the continuum, river-pattern thresholds can be identified where the pattern changes between straight, meandering, and braided. The pattern changes take place at critical values of stream power, gradient, and sediment load (202). In addition to the channel patterns shown in Figure 29, there are five basic bed-load channel patterns (Figure 30A) that have been recognized during experimental studies of channel patterns. These five basic bed-load channel patterns can be extended to mixed-load and suspended-load channels to produce 13 patterns (Figure 30). Patterns 1-5 are bed-load channel patterns (Figure 30A), patterns 6-10 are mixed-load channel patterns (Figure 30B), and patterns 11-13 are suspended-load channel patterns (Figure 30C). For each channel type, pattern changes can be related to increasing valley slope, stream power, and sediment load. The different bed-load channel patterns (Figure 30A) can be described as follows: Pattern 1: straight, essentially equal-width channel, with migrating sand waves; Pattern 2: alternate-bar channel with migrating side or alternate bars and a slightly sinuous thalweg, Pattern 3: low-sinuosity meandering channel with large alternate bars that develop chutes; and Pattern 4: transitional meandering-thalweg braided channel. The large alternate bars or point bars have been dissected by chutes, but a meandering thalweg can be identified. Pattern 5 is a bar-braided channel. Figure 29. Channel classification and relative stability as hydraulic factors are varied (203).

90 As compared to the bed-load channel pattern, the five-mixed load patterns (Figure 30B) are relatively narrower and deeper, and there is greater bank stability. The higher degree of bank stability permits the formation of narrow, deep straight channels (Pattern 6), and alternate bars stabilize because of the finer sediments, to form slightly sinuous channels (Pattern 7). Pattern 8 is a truly meandering channel, wide on the bends, relatively narrow at the crossings, and subject to chute cutoffs. Pattern 9 maintains the sinuosity of a meandering channel, but due to the greater sediment transport the presence of bars gives it a composite sinuous-braided appearance. Pattern 10 is an island-braided channel that is relatively more stable than that of bedload channel 5. Suspended-load channels (Figure 30C) are narrow and deep. Suspended-load Pattern 11 is a straight, narrow, deep channel. With only small quantities of bed load, this type of channel will have the highest sinuosity of all (Patterns 12 and 13). The linkage between sediment-load characteristics and planform provided by Table 6 and Figure 30 were important in interpreting the results of the data analysis for this project. This linkage was considered in developing the recommended classification of Figure 11. However, to use sediment load type as a primary classification approach could require a substantial field sampling program for every river/bend to be analyzed by the user of the prediction methodology. The pictorial approach of Figure 11 was considered more directly applicable to State DOT needs. It requires only a map, photograph, or visual inspection to apply. Application of this procedure to 58 Brice sites indicates that all sites would fit into one of the categories, without apparent anomalies, and the classification results are replicable. The preceding classification applies to adjustable alluvial rivers, with sediment loads primarily of sand, silt and clay, which would be considered regime channels by Montgomery and Buffington (205) who considered the full range of channels from high mountain bedrock channels to those described previously in Figures 29 and 30. This classification (Figure 31) starts at the drainage divide and moves down through bedrock and colluvial depressions or chutes to the point where one can recognize fluvial channels. Five distinct reach morphologies are identified: cascade, step-pool, plane-bed, pool-riffle, and dune- ripple (regime). Most of these reaches will be confined by valley walls and terraces in contrast to the alluvial regime channels. Table 7 summarizes the important characteristics of each channel type. Since reach morphologies are considered in profile, rather than in planform, this classification would not contribute to an analysis focused on meander mode or sinuosity. Rosgen (206, 207) developed a comprehensive system for classifying natural rivers. This system divides streams into seven major types on the basis of degree of entrenchment, gradient, width/depth ratio, and sinuosity. Within each major category there are six subcategories depending on the dominant type of bed/bank materials. The classification system shows a distinct bias toward streams that are relatively small and steep. For example, of the stream types categorized based on dominant bed material, seven are braided, 30 are entrenched, in the sense that overbank floods are confined by valley walls or terraces, and four are narrow, sinuous mountain meander-type channels. The basic framework of Rosgen's method is set out in Figures 32 and 33. While this classification is comprehensive in its scope, just as with the Montgomery and Buffington approach of Figure 31 and Table 7, it does not support classifying rivers based on meander mode or sinuosity.

91 Increasing Valley Slope Increasing Stream Power ⇒ Increasing Sediment Load Figure 30. The range of alluvial channel patterns. (A) Bed-load channel patterns, (B) Mixed-load channel patterns, (C) Suspended-load channel patterns (204).

92 Figure 31. Idealized long profile from hillslopes and unchanneled hollows downslope through the channel network showing the general distribution of alluvial channel types (from (205)). Summary Of the approaches reviewed, the classification system of Figure 11 was adopted as the most applicable for the purposes of this project. As shown in Figure 11, nine screening and classification categories can be used to represent the full range of meandering rivers encountered in the field. As noted above, equiwidth rivers, such as A, B1, and G1, can be screened as stable. One class, the "wandering" river shown as F, should be screened as potentially so unstable and unpredictable that further evaluation would not be likely to produce a meaningful result (in terms of predicting meander migration). All other meandering rivers can be classed as one of the remaining five categories, B2, C, D, E, G2, and analyzed by the photogrammetric comparison techniques presented in the Handbook. This approach incorporates the following characteristics: • Simple and directly applicable to the meander process with minimal training and/or explanation • Combines a classification system with a screening procedure to identify stable or highly unstable patterns • Pictorial in format, requiring only a map, aerial photograph, or visual inspection to apply • Provides a reasonable range of classes with which to segment the meander data base • Does not require field data (e.g., sediment sampling) to apply • Test case shows that most meandering river types or meander modes of interest to this project will fall in these categories • Test case shows that results are replicable when applied to the same data set by different evaluators

93

94 Figure 32. Key to classification of rivers in Rosgen's method (modified from Rosgen (206) by Thorne (200)).

95 Figure 33. Longitudinal, cross-sectional, and planform views of major stream types in Rosgen's method (modified from Rosgen (206) by Thorne (200)). AERIAL PHOTO COMPARISON METHODOLOGY In Chapter 2 (The Handbook), three methodologies for predicting meander migration are summarized: manual overlay techniques, computer supported techniques, and GIS-based measurement and extrapolation techniques. This section outlines the steps necessary to conduct a detailed meander migration analysis using aerial photo comparison techniques. The manual overlay of historic maps and aerial photos is used to illustrate the general comparative approach and the application of the acquired data and information to predict the position of a meander bend in the future. The computer assisted technique differs from this general approach only in the use of common computer software to assist in the historic assessment and prediction of bend migration. Similarly, the GIS-based technique applies the same general approach but uses the Data Logger and Channel Migration Predictor ArcView extensions supplied with the Handbook to conduct the comparison and prediction steps. Sources of error and limitations of the map and aerial photo comparison techniques are also appraised in this section. Manual Overlay and Prediction The following steps illustrate a simple overlay comparison of historic banklines and the process of predicting the potential future position of a bend based on past channel migration characteristics. STEP 1 - The first step in conducting a meander migration analysis using an overlay technique is to obtain aerial photographs and maps for the study area. As summarized in Chapter 2, the Handbook provides a detailed listing of sources and Appendix A of the Handbook provides general instructions on downloading digital aerial photographs and topographic maps from Microsoft's TerraServer Web site.

96 STEP 2 - The maps and photos must be enlarged or reduced to a common working scale. The scale of the most recent map or photo should be used since it will be the basis for making and comparing historical meander pattern changes and predicting the position of a given bend in the future. STEP 3 - After defining a working scale, the photos and maps are registered to a common base map or photo by identifying several features or points that are common to each photo/map being compared. The registration points do not need to be common to all the maps and photos, only to the subsequent map or photo to which it is being compared, since comparisons can be performed in pairs. For example, Figures 34 and 35 show the 1937 and 1966 aerial photos, respectively, for a reach of the White River in Indiana. Four registration points have been identified on the 1937 photo that are also on the 1966 aerial photo. Two registration points are road intersections and two are isolated vegetation (trees or large shrubs). Registration points that bracket the site on both sides of the stream and at both ends of the reach are most useful because they reduce the amount of potential error within the bracketed area. Intermediate points between the end points are helpful in accurately registering the middle sections of the reach. More than five or six registration points can make registration difficult because of the difficulty in aligning all the registration points among the various aerial photos and maps used. However, there will be instances where there will be very few identifiable registration points common to both photos, and these sites may have the potential for significant error. STEP 4 - After identifying the registration points, banklines and registration points for each year are traced from the aerial photo onto a transparent overlay. The method for identifying and tracing the banklines is described in detail in Chapter 5 and Appendix B of the Handbook. Registration points are included on the overlay so that they can be easily plotted onto other aerial photos or maps for comparative purposes. The traced banklines and registration points of the White River for 1937 are plotted on the 1966 aerial photo in Figure 36 for comparative purposes. Figure 34. Aerial photograph of a site on the White River in Indiana showing four registration points (circles designated a through d) common to the 1966 aerial photo.

97 Figure 35. Aerial photo of a site on the White River in Indiana showing the four registration points common to the 1937 aerial photo in Figure 34. Levee Flow Figure 36. The 1966 aerial photo of the White River in Indiana with the 1937 bankline tracing and registration points. Since most meander bends are not simple loops, the loop classification of Brice (72) can be used to characterize the shape of each bend that is to be analyzed (Figure 37). Meander bends seldom form single symmetrical loops, but instead are comprised of one or more arcs combined to form either symmetrical or asymmetrical loops. Brice (72) derived the classification scheme for meander loops from a study of the meandering patterns of 125 alluvial streams. The scheme consists of four main categories of loops (simple and compound symmetrical and asymmetrical) comprising 16 form types. Although compound loops are regarded as aberrant forms of indefinite radius and length, the meandering patterns can be divided into simple loops whose properties can be described, measured, and analyzed. The radius of curvature of most bends can be defined by fitting one or more circles or arcs to the bend centerline or outer bankline of a meander loop.

98 STEP 5 - Once the banklines for each of the historic aerial photos have been traced, circles are best-fit to the outer bank of each bend to define the average bankline arc, the radius of curvature (RC) of the bend, and the bend centroid position (Figure 38). The number of circles required to define the bend is based on the loop classification described above and shown in Figure 37. A detailed description of the method used to fit a circle to the outer bankline of a meander bend is provided in Appendix B of the Handbook. The radius of curvature and centroid position of the circle used to describe the bend will be used to make comparison with the bend measurements of previous and subsequent years. These measurements can then be used to determine migration rates and direction and estimate future bend migration characteristics. Figure 39 compares the best-fit circles and bend centroids for each bend traced from the aerial photographs for 1937 and 1966. The vector arrow at each bend shows the direction and magnitude of movement of the bend centroid between 1937 and 1966. For each bend, this vector may be resolved into cross- and down-valley components to determine the rates of meander migration. The change in radius of curvature of each bend is defined by the difference between the magnitudes of the vectors for 1937 and 1966. Figure 37. Meander loop evolution and classification scheme proposed by Brice (72). Flow is left to right.

99 1937 RC Figure 38. Circles that define the average outer banklines from the 1937 aerial photo of the White River site in Indiana. Also shown are the bend centroids and the radius of curvature (RC) for one of the bends. 1937 1966 Figure 39. Depiction of the bends from 1937 (dotted line) and 1966 (dashed line) outer banklines as defined by best-fit circles. The movement of the bend centroids (arrows) defines migration of the bends.

100 STEP 6 - The position of the bend at a selected date in the future can be predicted by simple extrapolation if it is assumed that the bend will continue to move at the same rate and in approximately the same direction as it has in the past. To estimate the position of a bend centroid in 1998, for example, the distance the centroid would be expected to move during the 32 years between 1966 and 1998 can be determined by multiplying the annual rate of movement for the 1937 to 1966 period by 32. This distance is plotted along a line starting at the 1966 centroid point and extending in the direction defined by the 1937 to 1966 migration vector. The radius of curvature of the bend in 1998 can be defined by determining the rate of change of the bend radius from 1937 to 1966 relative to the 1966 radius and multiplying this value by number of years from 1966 to 1998. A circle with that radius, centered on the predicted location of the centroid, is plotted on the tracing to indicate the expected location and radius of the bend in 1998. Figure 40 shows the expected outer bank circles for each of the bends of the White River in 1998, based on simple extrapolation of the rates and directions of change during 1937-66. Banklines for the 1998 channel can then be constructed on the tracing by joining the outer bank circles through interpolation, with the 1937 and 1966 banklines used to indicate the reach-scale configuration of the channel. Figure 41 shows the banklines observed in 1937 and estimated for 1998, overlain on the 1966 aerial photo. Inspection of the estimated banklines reveals that Bend 1 would encroach into the levee to the north by 1998 while growth of Bend 5 would likely cutoff Bends 6 and 7. 1937 1966 1998 (est.) Figure 40. Depiction of the bends from the 1937 (dotted line) and 1966 (dashed line) outer banklines, as defined by best-fit circles, and the predicted location and radius of the 1998 outer bankline circle (solid line).

101 Flow 1 2 3 4 5 6 7 Figure 41. Aerial photo of the White River in 1966 showing the actual 1937 banklines (white) and the predicted 1998 bankline positions (black). In Figure 42 the banklines predicted for 1998 by extrapolation of trends of change between 1937 and 1966 are superimposed on an aerial photograph taken in 1998. Two of the registration points used for this comparison are different because two of the original registration points from the previous aerial photos are no longer present on the 1998 aerial photo. Comparison of the actual and estimated banklines in Figure 42 illustrates that meander migration can be predicted relatively accurately using this simple approach. For example, the positions of Bends 3 and 4 and the cutoff at Bend 5 are accurately predicted. Errors in the predicted banklines can be accounted for by: (1) an artificial cutoff that affected Bends 1 and 2; (2) the natural cutoff at Bend 5 that led to Bends 6 and 7 being abandoned; and (3) construction of bank protection at Bends 3 and 5 during the period 1966-95. The artificial cutoff at Bend 1 may have been in response to the serious threat posed by bend migration toward the nearby levee. That cutoff caused Bend 2 to distort in a way that could not have been predicted from its previous behavior. Outer bank migration at Bends 3 and 5 appears to have been curtailed by bank revetments. The migration of Bends 3, 4, and 5, the cutoff of Bend 5, and the abandonment of Bends 6 and 7 were predicted with sufficient accuracy to meet the objectives of this study. It is likely that the positions of Bends 1 and 2, as well as the banklines in the revetted portions of Bends 3 and 4, would have been as predicted except for these engineering interventions. The case study of the White River used a single period (1937 to 1966) to predict the position of the banklines in 1998. To improve the reliability and accuracy of predictions it is desirable to use multiple pairs of aerial photographs to generate more than one period of analysis. By evaluating multiple periods, meander migration analysis can detect trends of change in the rate and direction of bend migration as well as time-averaged values.

102 1 2 3 4 5 6 7 Levee Figure 42. Aerial photograph of the White River site in Indiana in 1998 comparing the predicted bankline positions with the actual banklines. Sources of Error and Limitations Map and Aerial Photo Errors and Limitations The principal errors associated with aerial photos, and ultimately with maps, are systematic errors. These are errors that follow some mathematical or physical law. A correction can be calculated and the systematic error can be eliminated if the conditions causing the error are measured and properly modeled. The major sources of these errors are: • Film distortions due to shrinkage, expansion, and lack of flatness • Failure of fiducial axes to intersect at the principal point • Lens distortion • Atmospheric refraction distortions • Earth curvature distortion A detailed description of the causes of these sources of error for aerial photos was considered beyond the scope of the Handbook, but can be found in most textbooks on photogrammetry (e.g., Wolf and Dewitt (223)). Depending on the precision and accuracy requirements of a given project, corrections can be applied to eliminate the effects of these systematic errors. The primary sources of map error are associated with the vertical and horizontal accuracy and the age of the map. Most federal maps are required to meet rigorous standards for accuracy. The National Map Accuracy Standards, which were issued in 1941, apply to all Federal agencies that produce maps. Guidance is provided for horizontal and vertical accuracy and methods for testing map accuracy are outlined.

103 Although the Federal standards for accuracy may seem reasonable, the true accuracy of topographic maps may be insufficient or problematic when using the comparison techniques for defining and predicting meander migration. For example, if the potential horizontal error of the topographic map used in the comparison is a significant percentage of the actual channel width, then there can be substantial error between the mapped bankline position and the true bankline position for the same time period and between time periods. The map error may also be problematic if comparisons are made with historic survey data. Comparison of newer maps and aerial photos with older maps may also pose a problem since the older maps may have been compiled before the use of aerial photos. These maps are based on physical ground surveys, field notes, surveyor descriptions, and sketches made in the field. Therefore, the accuracy of historic maps decreases with increasing age. Maps that are geo-referenced may not match the positions of geo-referenced aerial photos. This can occur if geo-referenced digital maps are obtained from sources other than the USGS and used in conjunction with geo-referenced digital aerial photographs compiled by the USGS or other agencies. There may also be problems associated with the use of different horizontal datums. However, transformations from one datum to another have become commonplace with the increasing use of GIS. There are several potential limitations to the use of aerial photos and maps in the comparison techniques and in the evaluation of meander migration. Scale can be a significant limitation. There are potential problems associated with major scale differences between maps and photos and changes in scale on a given photograph because of distortion across the photo. In some cases, high altitude aerial photographs (scales of 1:40,000 or 1:60,000) may be all that are available for a particular site. A comparison of bankline positions from high altitude photos with those from topographic maps may be difficult because of the significant scale difference. In some cases, an enlargement of the aerial photo may provide an image of sufficient resolution or clarity to be used in the comparison. However, this is also dependent on the relative size of the channel with regard to the scale at which it is being evaluated. Often, the physical enlargement of high altitude photos using a copier or a flatbed scanner in conjunction with photo editing software yields images with poor resolution. Even though an aerial photo can be scanned at a high resolution, the quality of the resolution and the amount of visible detail is greatly dependent on the original image quality and clarity. However, the quality of an enlarged scanned image degrades rapidly after the image has been enlarged to more than 2 or 3 times its original size. The resolution of an aerial photograph also generally decreases with age because of changes in technology over time. In contrast, digital images such as those found on the TerraServer Web site can be downloaded at various scales and resolutions. Brightness and contrast also play a role in the usability of an aerial photograph. Photos can be too dark or too light and the contrast may be so coarse that the banklines of a channel may be difficult or impossible to identify. The time of the year or day on which the aerial photo is acquired is also important and can limit the usability of a photo. Long shadows, dense vegetation, and cloud coverage may partially or totally obscure the bankline or the entire channel. Aerial photos that were flown at midday and during winter months with little cloud

104 coverage are optimal. Photos flown during early spring, prior to leaf-out, may be useful, but spring floods may obscure the tops of the banks. Photos flown during summer months can be used if the density of the bankline vegetation is sparse enough to allow the user to adequately define the bankline. Otherwise, bankline positions will need to be estimated based on the locations of the crowns of the trees growing at the bankline. In this case the accuracy of the measurements is questionable. The age of an aerial photo or map may also limit their usefulness. Old maps and photos may not have the same identifiable geographic features or landmarks that are found on newer maps and photos. In these cases, identifiable landmarks that can be used as registration points may not be present on the older maps and photos. In addition, the township, range, and section lines found on newer topographic maps whose intersections could be used as registration points, may not be in the same location or may not be available on older maps. Measurement Error As with any methodology that requires the physical measurement of a quantity, the accuracy and precision of the measurements conducted under the comparison technique described in the Handbook can limit the usefulness of the acquired data. Obviously those measurements made visually using a ruler or engineering scale will be less accurate than those made using a computer. Also, repetitive measurements should be made the same way each time. Scale plays an important role in measurement error as well. Large-scale images (e.g., 1:10,000) show ground features at a larger, more detailed size and small-scale images (e.g., 1:50,000) show ground features at a smaller, less detailed size. Thus, using identical measurement techniques, measurements made on a large-scale maps and photos generally will be more accurate than those made on smaller scale maps and photos. Meander bends are rarely perfectly round with smooth banklines. They often are oddly shaped and their banklines are irregular (see Figure 37). As a result, fitting a circle to the channel centerline or outer bank can be very difficult. As a rule, the circle should be fit to the bend centerline or outer bankline between the crossings or at the point where the bend begins to straighten or where there is a major inflection in the channel. As much of the circle as possible should intersect the bankline or centerline and the amount of area outside the circle should match the amount of area inside the circle as closely as possible. The radius of curvature of a bend centerline or bankline can be significantly different depending on how the circle is fit to the bend, especially on smaller channels. Limitations of Overlay Techniques Overlay techniques require the availability of adequate maps and aerial photos that cover a sufficient period of time to be useful. The identification and delineation of a sufficient number of registration points common to each map and photo are also a fundamental requirement. All the registration points do not need to be found on all the maps and photos, but an adequate number of registration points used on each map or photo should match those on the previous or following map or photo. The registration points should bracket the area of interest (this would

105 require at least 4 common registration points) and should not change significantly in size over time. Even when a sufficient number of registration points are available, photo distortions or inaccuracies in mapping may not allow for an accurate registration of the images. In these cases, one must decide whether "close" is good enough or if the image should be abandoned. Excessive or very limited movement of the channel, cutoffs, and bank erosion countermeasures will also limit the usefulness of the comparison techniques. An analysis of the rate and extent of historical movement may be useless if excessive meander migration is a problem (as with meander Class F in Figure 11). Depending on the scale of the overlays, the amount of migration may be so small as to be undetectable or the overlays may be at such a small scale that the movement is not measurable. Countermeasures to halt bank erosion or protect a physical feature within the floodplain can also have an impact on the usefulness of the overlays. These features should be identified prior to developing the overlays. Anomalous changes in the bend or bankline configuration or a major reduction in migration rates may suggest that bank protection is present, especially in areas where the bankline is not completely visible or on images with poor resolution. Geologic features, such as clay plugs or rock outcrops, in the floodplain can also limit the usefulness of the overlays because they can have a significant influence on migration patterns. Bends can become distorted as they impinge on these features and localized bankline erosion rates may decrease significantly as these erosion resistant features become exposed in the bank. Where the channel encounters a geologic control or man-made feature, the channel may intersect the feature at a sharp or abrupt angle and migrate more rapidly down valley along the feature or become highly distorted. An example of this might be where a channel encounters geologic controls, bank protection, or levees that run parallel to the valley direction. In some cases the channel may encounter a very localized outcrop or hardpoint in the bank creating an irregular bankline or causing the bend to deform around it. In these cases, determining the radius of the outside bankline of a bend may be very difficult. Since any evaluation of meander migration requires an assessment of, among other things, changes in bend radius, judgment must be used in determining the radius of the bend, and possibly the bankline, by defining it with a best-fit circle of known radius. Where the channel makes a sharp or abrupt turn, mud flats or bars may develop along the outer bank in the upstream half of the bend, and the delineation of the outer bankline on a photo or map may be difficult. In this case, there are two methods of defining an approximate outer bankline radius. The first method is to identify the radius of the inner bankline by inscribing a best-fit circle on it and then determining the average channel width at the crossings in the reach. Then, the user can add 1.5 to 2 average channel widths to the inner bankline circle to define the outer bankline radius. Once this is accomplished, the user will need to evaluate how well the estimated outer bankline fits relative to the actual channel position, to similar bends that may be located in the reach, and to other features along the channel at the bend.

106 The second method requires the use of the edge of water at the outer bankline of the channel on the photo or map. This should provide a relatively close approximation of the outer bankline radius of curvature. Although both of these methods can contain significant error, they may provide the only reasonable approximation of the outer bankline and radius of curvature necessary to make a prediction of future bankline position. In reaches where geologic controls are exposed predominantly in the bed of the channel, migration rates may dramatically increase because the channel bed is not adjustable, which may cause the channel to migrate rapidly across the feature. A fundamental assumption of the overlay techniques based on aerial photo or map comparison is that a time period sufficient to "average out" such anomalies will be available, making the historic meander rates a reasonable key to the future. GIS PREDICTOR RESULTS The photo comparison methodology was tested by applying the ArcView-based measurement (Data Logger) and extrapolation (Channel Migration Predictor) tools to 50 sites. The Brice data set was used because banklines for three points in time (two from the original Brice data and one from recent aerial photography that was obtained as part of this project) were required to test the accuracy of the predictions. The first two bankline locations were used to predict the recent bankline location for comparison with the bankline on the recent photography. Sites that were classified as C sites (single phase meandering channels with point bars and wider bends) were selected because this classification included the greatest number of active, freely meandering bends with three time periods of coverage. Of the 50 sites, seven were excluded based on the same rationale that should be used in deciding whether the method should be applied to any specific bend. In several cases, the time interval between the first two banklines was 10 to 12 years whereas the second time interval (from the second bankline observation to the recent aerial photograph) was 25 to 33 years. Since the methodology is essentially an extrapolation of past movement, the extrapolation should not be significantly longer than the observed time period. Other sites were excluded due to close proximity to a major tributary, sand/gravel mining, and natural bend cutoffs. In one case, the bend was partially revetted and both upstream and downstream bends were completely revetted between the time of the second bankline observation and the recent aerial photo. The results of this evaluation are, therefore, representative of 43 relatively freely meandering bends. The average time intervals for the remaining 43 bends were 27 years between the first and second bankline observations and 26 years between the second and third bankline observations. Using the ArcView measurement tool, the bend radius and center location were measured for each of the three bankline locations. Typically, the first bankline was from the 1930s, the second from the 1960s and the third from the 1990s. The prediction tool was then used to predict the radius and center location for the third time period. Since the movement of the bankline is the item of interest, the evaluation focused on the accuracy of predicting bankline location. Four parameters were measured and compared for each of the 43 bends: (1) bend radius; (2) magnitude of maximum bankline movement; (3) direction of maximum bankline movement; and (4) maximum difference between the predicted and observed bankline at any

107 point along the bend. These four parameters represent different types of error in the prediction. The method may accurately predict direction of maximum migration, but could over- or under- predict the amount of movement. If both direction and maximum migration are predicted well, but the radius is not accurate, then the error in radius would result in some error in the bankline prediction at other locations along the bend. These measurements are illustrated in Figure 43, which depicts a bend on the Tombigbee River near Amory, Mississippi. The banklines are from the 1937 and 1969 Brice data set and the 1996 aerial photograph. The 32 year time period (1937-1969) is used to predict the movement over the subsequent 27 years (1969-1996). The 1937 channel is shaded and flow is from left to right. The two arrows show the magnitude and direction of the maximum bend movement between 1969 and 1996 for the predicted and actual bank location. The bend radius of the actual 1996 outer bank location was measured but is not shown in this figure to improve clarity. In addition to the predicted 1996 circle, the predicted upstream and downstream 1996 banklines were sketched based on past movement. Finally, the maximum difference between any point along the actual and predicted 1996 banklines was measured. In this case, the maximum difference occurs at the bend apex, although along the downstream limb of the bend similar amounts of difference occurs. For this site, the direction of bank movement was very accurately predicted. The radius was also well predicted, but the amount of bankline migration was under- predicted by approximately 50 percent. Overall, the prediction appears to be very reasonable and would alert a structure owner to potential problems with channel migration in this vicinity. There is one other noteworthy feature illustrated in Figure 43, the extreme change in channel width that occurs between bankline observations. There is a slight increase in width between 1937 and 1969 and, at the apex, approximately a 70 percent increase in width between 1969 and 1996. Nearly 40 percent of the sites experienced channel width changes of greater than 30 percent (increase or decrease) between the second bankline and the recent aerial photograph (the average change in channel width was 32 percent). On average, there was a 10 percent increase in channel width. If sites that experienced a 50 percent or greater change in width had been eliminated from the test of predictor results, an additional 10 sites would have had to be eliminated. As shown in Figure 43, the direction of maximum bank migration was measured for the predicted and actual banklines. The difference between the predicted and actual bankline migration directions is shown in Figure 44 as a bar chart and as a cumulative percentage. Fifteen of the 43 bends (35 percent) showed predicted and observed maximum bank migration within 10 degrees. More than half (60 percent) of the maximum bank erosion was within 20 degrees of the observed direction. Three of the bends had bank erosion direction greater than 40 degrees. It should be noted that at this level of error (45 degrees or more), one would be essentially predicting greater down-valley movement (translation) when the bank is actually eroding more across-valley (extension) or vice versa. A 90 degree difference would be predicting one mode of movement when the bank is actually migrating in the other mode. None of the 43 test bends showed more than 70 degrees difference between the actual and predicted direction of maximum bank erosion. In two cases the migration was actually in the up-valley direction and this direction was predicted by the methodology.

108 1937 1969 1996 – Actual 1996 – Predicted 100 m 300 ft Figure 43. Channel migration comparison for the Tombigbee River near Amory, MS. Figure 45 shows a comparison of predicted versus observed bend radius for the 43 test bends. In addition to the data, perfect agreement, ± 25, ± 50 and + 100 percent error lines are included. Fifty percent of predicted bend radii (22 of the 43 bends) were within 15 percent of the actual bend radii, and 65 percent (28 of 43) of the predicted bend radii were within 25 percent of the actual bend radii. On average, the predicted bend radii were approximately 5 percent smaller than actually observed. This could relate to the fact that, on average, channel width increased by 10 percent between the second and third bankline observations. Figure 45 also shows predicted and observed maximum bank migration magnitudes. Each of the channel bends were actively migrating. In one extreme case, a bend migrated down valley nearly 2,950 feet (900 m) in 24 years. The predicted down valley migration for this bend was nearly 3,610 feet (1,100 m). On average, the maximum predicted bank erosion was 22 percent greater than observed. This is probably due to the fact that maximum possible underestimation of bank migration is 100 percent (zero migration predicted versus some

109 measurable amount). However, it is possible to overestimate bank migration by more than 100 percent (490 feet (150 m) of predicted migration versus 213 feet (65 m) of observed migration is an overestimation of 130 percent). In the cases were the amount of migration was significantly overestimated, the prediction still appeared quite reasonable. Seventy percent of the predicted bank migration amounts were within 50 percent of observed amounts and 42 percent of the bank migration amounts were within 25 percent of the observations. 0 0 111 6 8 11 15 35% 60% 79% 93% 100%100%100% 98% 95% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 0-10 10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 Predicted versus Actual Maximum Bank Migration Direction (degrees) Pe rc en t i n Ca te go ry 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cu m ul at iv e P er ce nt Percent in Category (N=43) Cumulative Percent Figure 44. Predicted versus actual bank migration direction. One way of interpreting Figure 45 is to compare the range of scatter to that of sediment transport calculations or measurements. This is a valid comparison because bank migration is related to erosion and sediment transport. In the test data set, the maximum range of scatter for bank migration is approximately 0.8 of a log scale and the bulk of the data fall within one-half a log scale. Sediment transport measurements and calculations can easily scatter over a full log scale (and often two log scales) for a given hydraulic condition (see 224, 225). This has led to the tongue-in-cheek observation that the "best" sediment transport estimate is zero, because the prediction is then "only" 100 percent off. Certainly there is a desire for prediction with greater accuracy, but given the complexity and variability of this process, the range of scatter in the test data is expected and the predictions are reasonable.

110 10 100 1000 10 100 1000 Actual Bend Radius or Maximum Bank Migration (m) P re di ct ed B en d R ad iu s or M ax im um B an k M ig ra tio n (m ) Bend Radius Maximum Bank Migration Perfect Agreement +/- 25 % +/- 50 % +100 % Figure 45. Predicted versus actual bend radius and bank migration magnitude. Another way of assessing the accuracy of the predicted bank migration amounts is to compare the errors in bank migration to changes in channel width. A large absolute error in predicted channel migration is more likely for a large channel than a small channel. Figures 46 and 47 show errors in channel bank migration relative to the channel width as both incremental and cumulative percentages. The channel width used for normalization was the width for the intermediate time period. For comparison, the change in channel width between the second and third time periods is also shown. These figures illustrate that error in migration (defined as difference between the length of the arrows shown in Figure 43) is most frequently within 20 percent of channel width. These errors are distributed similarly to the change in channel width. In other words, errors in predicting bankline location are comparable to the changes expected in channel width. Given that the predictions are, on average, for a 26 year time period the amount of error in migration (or change in width) is on the order of one percent of the channel width per year.

111 Figures 46 and 47 also show the maximum difference in bankline at any location along the bend. This is the most extreme measure of error for the methodology. Even if the maximum bank erosion amount and direction were predicted accurately, some point along the actual bankline could deviate from the prediction. Although these errors are greater than width variability, they are still similarly distributed. In summary, the evaluation of 43 active, freely meandering bends indicates that: (1) bank erosion direction is predicted within 0 to 30 degrees in nearly 80 percent of the cases; (2) for nearly 60 percent of the cases bank migration magnitude is predicted within an accuracy of one percent of channel width per year over the time period covered by the prediction; and (3) this level of accuracy is comparable to the variability of channel width. A qualitative assessment of the procedure indicates that the majority of the predictions were reasonable and compared well with the actual channel migration. 0% 5% 10% 15% 20% 25% 30% 0-10 10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100 >100 Error or Difference as Percent of Width P er ce nt in C at eg or y Error in Migration Change in Channel Width Maximum Error in Channel Migration Figure 46. Error in bank migration as a percent of channel width (incremental percent). REGRESSION ANALYSIS The rates of bend expansion, extension, and translation were computed for each location and each time period. The bank line data are generally from the 1930s, 1960s, and 1990s. Using the first and second, second and third, and first and third time periods resulted an average intervals of 27, 26, and 56 years, respectively. The data were grouped using the Modified Brice Classification. The A, B1, B2, and C classes included 89, 249, 408, and 915 data points in the Brice data set, respectively. Standard single variable and multi-variable regression techniques were applied in an attempt to obtain regression relationships for predicting change in component variables that describe meander migration.

112 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0-10 10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100 >100 Percentage of Channel Width Cu m ul at iv e Pe rc en t Error in Migration Maximum Error In Migration Change in Channel Width Figure 47. Error in bank migration relative to channel width (cumulative percent). Expansion One mode of meander migration is radius expansion. Bends can either expand or contract (negative expansion). Figure 48 shows the ratio of bend radius of curvature at the end of a time period to radius of curvature at the beginning of the time period plotted versus initial radius of curvature over width (Rci/Wi, bend tightness). Although there are expanding and contracting bends throughout the range of Rci/Wi, Figure 48 shows that tighter bends tend to expand (Rcn/Rci >1) and longer bends tend to contract (Rcn/Rci <1) by reducing their radius. The data set is dominated by a cluster of data points with 1<Rci/Wi<3 centered on Rcn/Rci = 1. Rcn/Rci equal to one indicates that the bend did not change its radius of curvature over the time interval. An equation that appears to describe expansion is: c c R bW R1a dt dR ⎟⎠ ⎞⎜⎝ ⎛ −= This equation accounts for the trend that, as the radius changes, the value of Rc/W changes and, therefore, the rate of expansion should change. When integrated, the solution to this equation is: )atexp( bW R1 bW R RR i ci i ci ci cn ⎟⎟⎠ ⎞ ⎜⎜⎝ ⎛ −+ =

113 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Rci/Wi R cn /R ci Time 1-3 (52 yr. avg.) Time 1-2 (27 yr. avg.) Time 2-3 (25 yr. avg.) T = 25 years T = 52 years Figure 48. Change in Radius of Curvature versus Radius of Curvature/Width (C Sites). The coefficients "a" and "b" were fit to the data of Figure 48. At t=0 Rcn = Rci and at t = ∞ the ultimate radius is equal to bWi. When a bend has an Rci/Wi = "b", then the bend does not change its radius. For a bend that starts at an Rci/Wi ≠ "b", as time progresses, Rci/WI approaches "b" and the rate of change approaches zero. Note that the value of "b" has physical meaning. The coefficient "b" is the ratio of bend radius to bend width that is most stable, at least in terms of expansion. This does not, however, mean that the bend is not migrating through extension or translation. The best fit results come from using different values of "a" for Rci/Wi less than "b" and for Rci/Wi greater and "b." From the C Site data shown in Figure 48, "b" = 2.2, a+ = -0.033 and a- = -0.0019. Plots of this equation for 25 and 52 years are shown in Figure 48. The equation produces an R2 = 0.70 for the future radius of curvature, Rcn. While this may appear to be a good fit, most of the correlation is due to the fact that the existing radius (Rci) is a fairly good estimate of the future radius (R2 = 0.65). A more meaningful way to compute R2 for this equation is to see how well the ratio of Rcn to Rci is predicted. Although the existing radius may be a good estimate of the future radius, this assumption gives the equation Rcn/Rci = 1.0 radius, the equation yields an R2 = 0.23, indicating that while there is a trend (which is evident in Figure 48), there is significant scatter around the equation. As shown in Figure 48, the data set is dominated by a cluster of data points with 1<Rci/Wi<3 centered on Rcn/Rci = 1. Visually, it appears that there should be a steeper slope for Rci/Wi<2. In an attempt to put greater weight on the low values of Rci/Wi, the data were grouped by time interval and by Rci/Wi (Figure 49) and the equation was fit with the grouped data. The grouped data show the trend of the data much more clearly and put equal weight on data over the entire range. The data follow the expected trends of greater change outside an Rci/WI = 3 and

114 greater change for longer time periods. The R2 for this data is 0.96, although this is really only a measure of how well the equation fits the mean trend of the data. The equation coefficients for Figure 49 are "b" = 2.6, a+ = -0.060 and a- = -0.0025 as compared with "b" = 2.2, a+ = -0.033 and a- = -0.0019 from the original data. The data show a trend for expanding radius for tight bends, contracting radius for long bends, and relative stability in radius for bends with R/W equal to 3. Attempts to improve the predicted radius by including discharge, unit discharge, slope, stream power, unit-stream power, grain size and percent silt-clay did not yield increased R2. 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 0 1 2 3 4 5 6 7 8 9 10 11 12 Rci/Wi R cn /R ci 10-15 yr 18-32 yr 33-45 yr 50-61 yr 12-yr 25-yr 40-yr 55-yr Figure 49. Change in Radius of Curvature versus Radius of Curvature/Width (C Sites grouped data). Translation and Extension The two other modes of meander migration are translation and extension. These modes may also be positive or negative depending on the direction of movement, but they tend to be positive. Statistically significant relationships for extension and translation were also not forthcoming. Figure 50 shows the C site data from Figure 17 comparing bank migration (the vector sum of expansion, extension and translation) to channel width. There is a very weak correlation between bank migration and width, although the data scatter over 2.5 log scales. Although the "best fit" line has a slope of 0.47, the data appear to follow a steeper trend. Multiple regression did not improve the value of R2. The variables included in the multiple regression were channel width, channel slope, average annual unit discharge, average peak unit discharge, stream power, bed material size, bank material percent silt-clay and radius of curvature. The primary variable appeared to be width, although at a very low correlation.

115 C Sites y = 0.40x0.47 R2 = 0.12 0.1 1 10 100 10 100 1000 10000 Channel Width at Apex (ft) A bs ol ut e M ov em en t ( ft /y r) C Sites Power (C Sites) Figure 50. Bank Erosion Rate versus Channel Width. Channel width appears to be an important variable in predicting all the modes of migration. Nanson and Hickin (91) attempted to use R/W to predict erosion rates. Cherry et al. (48) showed a weak correlation between average bank erosion rates and width. However, as illustrated in Figure 51, channel width is a river property that varies considerably over time. Figure 51 shows the C site data, where for each bend the channel widths from the 1990s are plotted on the Y-axis versus the 1960s data on the X-axis (as well at the 1960s data plotted versus the 1930s data). If a relationship for predicting future meander migration is developed using channel width as a variable, then the large variability in channel width (from one time period to the next) will only confound the prediction. As shown in Figure 51, width values scatter as much as an entire log scale and generally over half a log scale. For example, some channels that started as 200 feet (61 m) wide were less than 100 feet (30 m) wide 30 years later and other 200 feet (61 m) wide channels were greater than 800 feet (244 m) wide after 30 years. Change in channel width has a direct impact on bank location, but it is not truly meander migration as defined for this project. Therefore, any attempt at predicting meander migration, whether empirically, through physical modeling, or by photo comparison, is subject to a large degree of uncertainty based on the fact that width varies considerably over time.

116 (W1965=200,W1998=844) (W1938=180,W1968=86) 10 100 1000 10 100 1000 Channel Width at Beginning of Time Period (ft) C ha nn el W id th a t E nd o f T im e Pe rio d (f t) Perfect Agreement +/- 50 Percent C ha nn el W id th a t E nd o f T im e Pe rio d (f t) Figure 51. Change in Channel Width at C sites. EVALUATION AND TESTING OF THE METHODOLOGY The project scope included two tasks to evaluate and test the methodology. Task 6 involved internal testing by the Research Team and Task 8 required providing the methodology to at least five State DOTs for their independent assessment and report on the results. The methodologies and the Handbook were revised following each task based on results and recommendations from individuals and agencies involved. This section describes the Beta test approach, the results reported, and revisions made based on recommendations from the evaluators. Overview A Beta test of the methodology for evaluating and predicting meander migration using aerial photo and map comparison techniques described in the Handbook was conducted in two phases: September and October 2001 (internal testing) and August – October 2002 (State Beta testing). The Task 6 internal test was conducted by three different evaluators that included an undergraduate geology student from the University of Nottingham, UK, a water resources project engineer (P.E.) employed by Ayres Associates, Inc., and a graduate degreed civil engineer (P.E.) from Mussetter Engineering, Inc. (MEI). Dr. Colin Thorne (UK) and Dr. Robert Mussetter, P.E, also reviewed the Handbook for consistency and accuracy. The Task 8 State Beta test participants are shown in Table 8.

117 Table 8. Task 8 State Beta Test Status Participants. State Contact Alabama Tom Flournoy, Bridge Hydraulic Engineer Alaska Mark Miles, State Hydraulic Engineer California Bill Lindsey, Structures Hydraulics (Kevin Flora) Maryland Andy Kosicki, Bridge Hydraulic Engineer (Stan Davis) Nevada Amir Soltani, Chief Hydraulic Engineer (Ron Schilling) Austin, TX Mike Kelly, Watershed Protection Georgia Tom Scruggs, Geotechnical Engineer (Jason Duley) Wyoming Bill Bailey, Hydraulic Engineer All evaluators were provided with maps and aerial photos and basic instructions on three techniques to be tested. The first technique requested that the evaluators conduct a meander migration evaluation and make a prediction of future channel position for a site on the Sacramento River using historic survey maps and paper copies of aerial photographs. The second technique requested that the evaluators use digital aerial photography to assess historic meander migration and make predictions on the future channel position for a site on the Minnesota River using Microsoft PowerPoint or some other graphics software. The third technique requested that the evaluators use the ArcView-based Channel Migration Predictor tool developed for this project to predict the future channel position of the Sacramento and Minnesota River sites. Once the Beta test was completed the results of each evaluator were examined and any problems or discrepancies were discussed with each evaluator. The results from each evaluator were compared with the results from the other evaluators to define any errors or inconsistencies in the methodology or determine if clarification in the techniques needed to be made. Each evaluator was requested to provide written comments on the usability of the methodology as part of the Beta test. Evaluation Procedure The evaluators tested the methodology set forth in the Handbook using the three techniques described therein: (1) the simple overlay technique; (2) the computer assisted technique; and (3) the Channel Migration Predictor technique. Simple Overlay Technique The site that was evaluated using the simple overlay technique consisted of three bends on the Sacramento River at Sidds Landing, California (Figure 52). The river is bound on the west side (right bank) by a levee that is unprotected along much of the reach. The three bends in this reach of the river are actively migrating downstream and pose a significant threat to the levee. Each evaluator was given a set of paper copies of historic maps and aerial photographs to conduct the evaluation of this site. The maps consist of a hydrographic and topographic survey of the river and meanderbelt conducted in 1937 by the U.S. Army Corps of Engineers and drawn at a scale of 1 inch = 400 feet. Paper copies of 1972 aerial photographs of the site, compiled by California Department of Water Resources (CDWR) at an optimal scale of 1 inch = 400 feet,

118 were made available to the evaluators. An enlarged black and white print (optimal scale = 1:27,000) of a NAPP (National Aerial Photography Program) aerial photo taken of the site in 1998 was obtained from the USDA Farm Service Agency Aerial Photo Field Office. 1 2 3 Figure 52. Beta test site on the Sacramento River at Sidds Landing, California. Flow is left (north) to right (south). The first step in the evaluation of this technique required that the evaluators register all the maps and aerial photographs together. Registration points common to all three sets of maps/photos needed to be identified first and then the maps and photos needed to be registered to a base map or photo by enlarging or reducing the maps and photos. The base map or photo to which the other maps and photos were registered was selected by the evaluator. The evaluator was then required to determine the approximate scale of the base. Once the maps and photos were registered to each other and to a common scale, the evaluator was required to delineate the banklines of the river for each year within the given reach. The banklines and registration points were traced onto transparent Mylar. Delineation of the banklines is somewhat subjective and often requires sound judgment based on observed conditions and knowledge of features common to rivers. In some cases, delineation of the bankline can be difficult especially where there is no well-defined scarp or vegetation line. In this case, the evaluators required some assistance in defining the banklines, particularly where the outer bank was gradually sloped or very irregular and where there were large, active point bars present. After the banklines for each data set were traced onto transparent Mylar sheets, they were overlain together and compared with regard to historic channel migration. This allowed the evaluators to see how the bends migrated and to assess outer bank retreat and inner bank growth over time. The next step in this technique required the evaluators to inscribe a circle along the outer bank of each bend for each year (Figure 53). The position of the bend centroid (center point of the inscribed circle) was defined on the tracings and the radius of the circle (radius of curvature) at each bend for all years was noted (Figure 54).

119 YEAR 1 YEAR 2 YEAR 3 Figure 53. Banklines and circles inscribed on outer bankline positions for a hypothetical channel at 3 different years. Rc1 Rc2 Rc3 YEAR 1 YEAR 2 YEAR 3 θA θBDA DB Figure 54. Diagram defining the outer bank radius of curvature in Years 1, 2, and 3, and the amount (DA and DB) and direction (θA and θB) of migration of the bend centroid during time periods A and B for a hypothetical bend.

120 Once the circles defining each bend for all years had been delineated, the evaluator used the changes in the radius of curvature and bend centroid position to predict the position and radius of the bend for some year in the future. In the case of the Sacramento River site, the position of the centroids and the radii of the three bends were predicted for the year 2028. Computer Assisted (PowerPoint) Overlay Technique The second method of predicting meander migration allows the evaluator to use a relatively common graphical editing software package, in this case Microsoft’s PowerPoint, to conduct the meander migration assessment and prediction. The steps used in the methodology are the same as those used in the simple overlay technique described in the previous section. The site used for this technique is located on the Minnesota River at Judson, Minnesota (Figure 55). The site consists of five relatively unconfined bends actively migrating across and down valley. The evaluators were provided with electonically scanned aerial photographs from 1950 and 1968. The evaluators were required to download, via the Internet, the most up-to-date image (1991) from Microsoft's TerraServer Web site. The scanned images are of unrectified aerial photos whereas the TerraServer image is rectified and georeferenced. The images were compiled in PowerPoint and then common registration points were located, banklines delineated, and circles inscribed along the outer bank of each bend for each year. Once this was done, the banklines and circles for each year were overlain using the common registration points. The unrectified images could be manipulated so that all the registration points closely matched those of the TerraServer Image using the sizing attributes of PowerPoint. 1 2 3 4 5 Figure 55. Beta test site on the Minnesota River at Judson, Minnesota. Flow is left (west) to right (east).

121 The banklines were overlain in a separate file from the bend circles for later comparison and assessment. Once the bend circles were overlain and the bend centroids delineated, a scale was placed on the overlays based on the georeferenced Terraserver photo, which has a known scale. After placing a scale bar on the bend circle overlay, the image was printed out so that the evaluator could use the bar scale to determine the bend radii, the migration distances between centroids, and the angle of migration for each bend in each year. From this information, the evaluators were able to determine the radius, angle of migration, and migration distance for each bend at some point in the future, in this case the year 2021. ArcView-Based Bend Measurement and Migration Predictor Technique This method required the evaluators to use the ArcView-based Bend Measurement tool and the Channel Migration Predictor developed for this project to conduct the analysis of the meander bends for both the Sacramento River and the Minnesota River sites. In both cases, the evaluators were given electronic files with the georeferenced banklines for all years for both sites. The evaluators were required to fit circles to the bends using the method described in the Handbook. This method required the user to place regularly spaced points around a bend to define the bend shape. The program then fits a circle to the bend points and provides the user with the radius and centroid location. This is done for all bends in all years. Once this is accomplished, the Channel Migration Predictor uses the data to delineate a circle that defines the given bend for a user-defined year in the future. Testing Results Results of Beta testing of the meander prediction methodologies are compiled in Appendix C as follows: • Table C1 – Simple Overlay: Sacramento River Site • Table C2 – ArcView: Sacramento River Site • Table C3 – Computer Assisted: Minnesota River Site • Table C4 – ArcView: Minnesota River Site • Table C5 – Summary of Beta Test Results Simple Overlay – Sacramento River Site Although there were differences in the magnitude of the predicted migration distances and the radii of the bends for the year 2028 among the evaluators (Table C1), the evaluators made similar predictions of the direction and extent of migration. The differences in the predicted bend radius of curvature and migration distance among the evaluators was attributed to differences in judgment in delineating the banklines and fitting a circle to the outer bank, both of which have a major influence on the radius and centroid position of the predicted bend. However, each evaluator did come to the same conclusion with regard to the relative speed of migration and the general change in the radius of each bend as well as the threat to the west (right) bank levee posed by continued meander migration.

122 Computer Assisted – Minnesota River Site Table C3 provides a comparison of the results of this technique for each evaluator. Again, there are difference among the evaluators in measuring migration distance and radii for each bend, which can be explained by differences in judgement in delineating the banklines and fitting the outer bank circles. However, an examination of the results indicates that the differences in measurement among the evaluators are generally smaller than with the simple overlay technique, suggesting that the use of a graphical editing software package may be more accurate because the evaluators are able to more accurately delineate the banklines. The greater accuracy in delineating the banklines may be attributed to the ability to acquire images with greater resolution, the evaluators ability to freely zoom in on the images, and the fact that the images can be printed at a usable working scale. ArcView Based Measurement and Prediction Tables C2 and C4 summarize the results of the ArcView analysis of the Sacramento River and Minnesota River sites. Differences in measurement and prediction indicated that describing a bend by fitting points to the outer bankline in the Channel Migration Predictor is also somewhat subjective. However, the differences in the results among the evaluators are considerably smaller than in the previous two methods. The smaller discrepancies among the evaluators is attributed to the fact that each evaluator used the same banklines and that redefining the fitted circle was relatively easy if the user was not satisfied with the fit. Statistical Analysis of Test Results The Beta test evaluation provided useful comments and recommendations on the Handbook and on the methodology. Since the evaluation included numerous measurements and predictions of bend properties, the results compiled in Appendix C were also used to evaluate whether consistent measurements were made by the various Beta testers and whether consistent measurements were made using the various measurement techniques. Figure 56 shows a comparison of the average bend radius measurements and predictions for each of the bends in the Beta tests. The simple overlay and computer assisted results are plotted versus the ArcView results. The data show that the two "manual" techniques produce average bend radius measurements and predictions typically within 20 percent of the ArcView tools. There are twice as many measurements as predictions because the methodology requires two measurements of bend radius to make one prediction of bend radius. The two sites provide a range of measured bend radius from approximately 500 ft to over 2,000 ft (152 m to 610 m). Similar results are produced for migration distance when the simple overlay and computer assisted results are compared with the ArcView results (Figure 57). As with the bend radius, the two manual techniques produce average migration distance measurements and predictions typically within 20 percent of the ArcView tools. The two sites provide a wide range of migration distances from less than 100 feet (30 m) (Minnesota River) to over 2,000 feet (610 m) (Sacramento River). This shows that consistent results are obtained using the three techniques for a range of bend sizes and rates of movement.

123 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 ARCVIEW Radius (ft) S im pl e O ve rl ay o r C om pu te r As si st ed R ad iu s (ft ) ARCVIEW vs. Simple Overlay (measurements) ARCVIEW vs. Simple Overlay (predictions - 2028) ARCVIEW vs. Computer Assisted (measurements) ARCVIEW vs. Computer Assisted (predictions - 2028) + 2 0 % - 20 % Per fect Ag reem ent Closed symbols are Sacramento River Bends Open symbols are Minnesota River Bends Figure 56. Comparison of bend radius measurements. 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 ARCVIEW Migration Distance (ft) Si m pl e O ve rl ay o r C om pu te r A ss is te d M ig ra tio n D is ta nc e (ft ) ARCVIEW vs. Simple Overlay (measurements) ARCVIEW vs. Simple Overlay (predictions - 2028) ARCVIEW vs. Computer Assisted (measurements) ARCVIEW vs. Computer Assisted (predictions - 2028) + 2 0 % - 20 % Per fect Ag reem ent Closed symbols are Sacramento River Bends Open symbols are Minnesota River Bends Figure 57. Comparison of migration distance measurements.

124 Figures 56 and 57 indicate that generally consistent results (within 20 percent) are to be expected between the three techniques for measuring and predicting bend radius and movement. These results are encouraging considering the widely varying backgrounds and experience of the Beta testers. The results do, however, show that the manual techniques (simple overlay and computer assisted) tend to predict slightly smaller bend radii and rates of movement. Figure 58 shows measurements and predictions of bend radius and movement for the two sites. The line of perfect agreement and the best-fit lines are shown for bend radius and migration distance. This plot shows that the simple overlay and computer assisted techniques produce measurements of 11 or 12 percent less than the ArcView. This is probably related to a difference in manually fitting a circle versus the least-squares technique used in the ArcView tool. In the least-squares algorithm, the user digitizes points along the bankline and the algorithm fits a circle to the data. In the manual techniques, there is probably a tendency to inscribe a circle within the bankline, which would tend to yield smaller radii. With either technique, it is important to be consistent. This is discussed in the Handbook with recommendations given to the user to measure an average bankline rather than to inscribe a bankline. Radius y = 0.88x R2 = 0.95 Migration Distance y = 0.89x R2 = 0.96 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000 ARCVIEW (ft) Si m pl e O ve rla y or C om pu te r A ss is te d (ft ) Radius Migration Distance Perfect Agreement Linear (Radius) Linear (Migration Distance) Figure 58. Comparison of ArcView and manual techniques for radius and migration. Another use of the Beta test data is determining the consistency of bend measurements. For the two Beta test sites, the variability of the measurements was also reviewed. Tables C1 through C4 show all of the bend measurements for the in-house and state evaluations. For each bend, the average value and standard deviation were calculated for radius and migration distance. These results are summarized in Table C5. If all the measurements of radius and migration were identical, the standard deviation of the measurements would be zero. Since there is variability in the measurements, the magnitude of the standard deviations of the measurements is a measure of consistency.

125 As shown in the tables, the ArcView measurements produced slightly lower standard deviations than the manual techniques, so ArcView produces more consistent results. Also, measurements of bend radius had much lower standard deviations than migration distance, so measurements of radius are more consistent than migration distance. In summary, using ArcView, the average standard deviation for bend radius is 16 percent of the average measured value (68 percent of the measured values are within 16 percent of the mean) and the average standard deviation for migration distance is 34 percent of the average measured value. Using the simple overlay and computer assisted techniques, the average standard deviation for bend radius is 19 percent of the average measured value and the average standard deviation for migration distance is 36 percent of the average measured value. The Beta test provided useful comments and suggestions on the Handbook and methodology. Between five and nine independent measurements of bend radius and migration distance were performed for each of the bends in the Beta test data set. These data showed reasonable consistency in the measurements and between the methods. Summary of State DOT Comments Table 8 provides contact information on the Beta test agencies. Two State DOTs (Alabama and California) provided comments on the Handbook, recommendations on the methodology, and returned their working documents (overlays) and digital files for evaluation. Maryland and Wyoming provided comments on the Handbook and methodology, and returned all files for evaluation except their ArcView files. Alaska provided comments and a detailed errata sheet for the Handbook, but did not return results or files from testing the methodology. The City of Austin, Texas provided general comments on the Handbook and methodologies. GaDOT was unable to complete the evaluation as the materials did not arrive in time for a summer intern to undertake the evaluation. General comments on the Handbook and methodologies are summarized below. ALABAMA Handbook • Alabama has quite a few streams that meander across the State and this will be a very good and useful tool. • The handbook was very well laid out and organized. It clearly stated the problem and reason for this project. The background chapters covered the different aspects of stream meandering and aerial photography where (anyone) could follow along. • The procedure for the methodology was presented in a fashion that I like, step by step, and the explanations and illustrations were very good. I did have some problems especially in the introduction chapter, some sentences just didn't flow smoothly. If these sentences were broken up and pared down or just rewritten, the material would flow and read more smoothly. • The work problems associated with the handbook were well prepared and thought out.

126 • PowerPoint and ArcView (were) a little frustrating at times. Manual Overlay Technique • The handbook outlined the manual overlay technique thoroughly using a step by step procedure. • The biggest problem was getting the photos printed out. • The steps outlining the manual overlay technique were easy to follow. Computer Assisted Technique • The process made it a lot easier once (you) figured out how to use the software tools. ArcView Predictor • The Data Logger and Channel Migration Predictor provided an even easier and faster way of predicting channel migration. • I did not fully understand or comprehend how to import georeference photos into ArcView or other historical photos. • There was also confusion on my part as to where the best location was to take certain measurements (upstream and downstream). ALASKA Handbook • The manual is in very good shape. It will be useful for the practitioner. • Most of my comments can be classified in the clarification or nitpicking categories. CALIFORNIA Handbook • Thanks for letting me be a part of the Beta test for this project. I found it to be insightful and had some promise. • I would like to get the final copy of the Handbook and programs when they are finalized. Chapters 1 - 4 were clear, informative and straightforward.

127 Chapter 5 – The data that needs to be measured with the data logger is presented without much rationale for why or how the data is used. Chapter 6 – The description of how to fit a circle to an odd shaped bend could be amplified more. Chapter 7 - The methodology seemed straightforward when I read it through, but questions/problems arose when I tried the test cases. Manual Overlay Technique • Printing out the aerials on large 30 x 40 paper was a pain. • I was not able to get the common registration points to overlay exactly making me wonder about the accuracy of my observations and work. • Calculations of the predicted Radius of Curvature and Migration Angle left me confused. Issues regarding the use of judgment and reasonableness should be clarified further. • More guidance should be given for when to use Eqn. 7.4 rather than using the Period B angle for Period C. • Not discussed is Channel Width. After drawing the prediction circles, I was unclear on how to connect the new banklines together and just eyeballed a channel width to draw in the inner bank. This could be clarified. • I was confused about how to deal with man-made structures. Computer Assisted Technique • Using the electronic files was easier than the manual method due to the size of the mapping used in the manual method. • I was comfortable working with PowerPoint and the Drawing tools, but I think the Handbook should provide more basic instructions for others who may not be as familiar with these tools. • The Handbook should have explained about how to get the size of the circles in PowerPoint. Also, instructions for how to create a new size of circle using the shape attributes and how to place the predicted circles at set distances and angles from the prior circle should be in the book. ArcView Predictor • Overall, I felt most confident using this method, because at least some of the subjectivity was removed by having the circles automatically fit the bend points.

128 • Measuring the wavelength and Amplitude was confusing. The start and end points for the wavelength and amplitude should be clearly defined. • Where is the Apex of the Bend? At the center or the widest point? This should be clearly defined and illustrated. • The Frequency Characteristics Section was a little confusing. MARYLAND Handbook • We would like to compliment you for undertaking this effort that we see as an important step in creating design procedures based on stream morphology. • We will not be able to test the procedures at one of our sites as all of them fall into the category of "confined" sites. We believe that such sites cannot be analyzed using the current procedures. • The methodology developed in this handbook is the first of this kind to quantitatively predict stream bend migration. • The text is well organized and concise. The contents are explained well and should be easily understood by practicing engineers. • It is a very simple and workable method. However, the applications may be limited to stream bends of an unconfined nature with isotropic soil conditions. Where natural hard points or man-made structures confine the stream flows, the extrapolation of the historic records of bend migration into the future may not prove to be reliable. • With regard to use of equations for prediction, a non-linear equation like Eq. 7.3 may serve better than a linear equation like Eq. 7.4 because non-linear extrapolation is based on three data points instead of two data points for linear extrapolation. Computer Assisted Technique • The (PowerPoint) process requires a change in scale in three different steps during the process. These changes are likely to create a scale factor problem in the developed prediction model. For this reason, we suggest using the CADD method instead of the PowerPoint method. • Using CADD will serve to reduce the steps in this procedure and also provide better answers for the prediction. The user can import a digital map into CADD with a given scale, trace the stream alignments and perform various calculations for parameters such as the center of the circle and the radius of curvature. From this CADD file the user can then import the traced stream alignments into the bend prediction program, perform the analysis, and determine the

129 resulting prediction of the future bend location. During this procedure, only two steps are involved and no change of scale takes place. NEVADA Handbook • I read the Handbook for clarity and found that it was easy to read and understand. I thought that tabs would be helpful for going back through the Handbook to do the exercises. • There is only one area I would expand on and that was Frequency Characteristics of Bend Migration. This section was a little confusing to me. Manual Overlay Technique • The "Manual Overlay Technique" was very easy. The only thing I would add is to reiterate the importance of having the photos/map at a common scale to one another. • I would like to reiterate that the Manual Overlay technique was so easy I don't see the need for the other options. Computer Assisted Technique • When I used the computer assisted overlay technique, I found this method to be time consuming and tedious. I would never use this method unless I was going to use it for a presentation or have some nice pictures for documentation purposes. ArcView Technique • While using the ArcView technique, I experienced several error messages bringing the data into ArcView. I've never used ArcView until now and I felt that the Handbook should give step-by-step instructions using ArcView. Because of time and my inexperience using ArcView, I didn't proceed any further. TEXAS (City of Austin) Handbook • In general Chapters 7 and 8 contained examples that were easy to follow, once the map justification was complete. After completing one exercise, the procedure was easy to replicate.

130 Manual Overlay Technique • It would be helpful to have more detailed instructions on justifying the three maps. It took a good deal of trial and error to produce maps that were of comparable scale. Computer Assisted Technique • It would be helpful to include more complete instructions on using PowerPoint for those not completely proficient. ArcView Technique • It would be helpful in using the data logger to have more precise instructions on the sequence of measuring and archiving. • It would be helpful to have more detailed directions for delineation of banklines, image use and georeferencing in ArcView. • It would be helpful to have more detailed directions on what to look for after themes are added on channel migration predictor module. • Overall, I think the data logger and channel migration predictor hold great promise for using aerial photos to predict channel migration. WYOMING Handbook • I found the paper (Handbook) to provide an excellent description of the meandering process and excellent background regarding past meander studies by other researchers. This lays a good foundation on which to study meandering patterns and erosion. The methods appear to be practical enough to employ on many highway projects. • The Handbook presented a clear and organized background on stream morphology and meandering. The explanation for applying this methodology is also well written. • These methods for predicting channel migration would be beneficial to our department. Each method proves useful based on project objective and resource constraints. Thanks. Manual Overlay Technique • The explanations for applying the methodology for the manual overlay technique were very clear. Explanations were sufficient enough that review of examples wasn't necessary. The examples were clear and provided guidance into the application of equations, and calculation of variable.

131 • The one exception was that there was some confusion when using the rate of change of migration angle equation. Computer Assisted Technique • MicroStation (CADD) was implemented for the computer assisted methodology. ArcView Technique • The Channel Migration Predictor was very well explained and worked with few problems Summary and Planned Response to State Evaluation Appendix C provides a summary of results from both internal testing (Task 6) and state evaluation (Task 8). While a range of results is apparent, this is not unexpected with an empirical approach requiring subjective judgments. The results are influenced by the background and experience of the evaluator and the care with which the measurements are done. They are also a function of the experience of the evaluator with manipulating map or aerial photography scales, determining registration points, and recognizing geomorphic features on aerial photography. As with any new skill, making reliable meander migration predictions from aerial photography requires practice and the skill can be improved with training (see Applications and Implementation). In general, it can be concluded that the Beta test results indicate: • The Handbook is well organized, well written, and generally easy to follow. The step-by- step approach on examples was well received. • The Handbook provides useful methodologies that are easy enough to apply in practice to be used by DOTs on a regular basis to support design, rehabilitation, and maintenance decisions. • The complexity of the computer assisted technique did cause problems for those not proficient in PowerPoint. Similarly, reviewers not familiar with GIS/ArcView had difficulty with that approach. • All reviewers seemed to be comfortable with the basic manual overlay technique, which provides a good fall-back approach for any analysis. In fact, this fundamental approach may be the preferred methodology for a DOT with only a few sites to analyze or where meander predictions are required only infrequently for specific projects. As a result of the Task 8 Beta test, the following modifications/revisions were made to the Handbook: • Minor editorial changes and revisions. • Review the Introduction for readability and make appropriate changes.

132 • Rewrite section on Frequency Analysis and provide more explanation. • Clarify where and how to make critical measurements (e.g., amplitude, bend apex, channel width, wave length). • Clarify the data requirements for the ArcView data logger. • Provide cross references to Appendix B of the Handbook in all discussions describing how to fit a circle to a bend. • Provide cross references to the expanded frequency analysis section where issues of judgment and reasonableness of a prediction are discussed. • Amplify the discussion of the prediction of migration angle. • Provide a section in Chapter 7 of the Handbook that references the possible use of CADD for analysis (see Maryland SHA comments on Computer Assisted Technique). Several reviewers suggested that the Handbook should have more basic instruction in PowerPoint (California) and ArcView (Nevada). It is not the purpose of the Handbook to provide an introduction to or instruction in specific software packages such as PowerPoint or ArcView. There are numerous manuals, texts, and help files for this purpose. There are also other drawing tools that could be used for the computer assisted technique (e.g., Corel Draw) and CADD packages that will accomplish many of the same functions (e.g., MicroStation Descarte, AutoCAD). The Handbook assumes a certain level of familiarity and skill on the part of the user or access to staff (e.g., a GIS section) who can assist. The manual overlay technique provides a good fall-back approach for those not familiar with the more advanced approaches (see Nevada DOT comment). APPLICATIONS AND IMPLEMENTATION The Handbook Approximately 84 percent of the 575,000 bridges in the National Bridge Inventory (NBI) are built over streams. A large proportion of these bridges span alluvial streams that are continually adjusting their beds and banks. Many, especially those on more active streams, will experience problems with scour, bank erosion, and channel migration during their useful life (201). The magnitude of these problems is demonstrated by the estimated average annual flood damage repair costs of approximately $50 million for bridges on the Federal aid system. Highway bridge failures caused by scour and stream instability account for most of the bridge failures in this country. A 1973 study for the Federal Highway Administration (226) indicated that about $75 million were expended annually up to 1973 to repair roads and bridges that were damaged by floods. Extrapolating the cost to 2003 makes this annual expenditure to roads and bridges on the order of $300 to $500 million. This cost does not include the additional indirect costs to highway users for fuel and operating costs resulting from temporary closure and detours and to the public for costs associated with higher tariffs, freight rates, additional labor costs and time. The indirect costs associated with a bridge failure have been estimated to exceed the direct cost of bridge repair by a factor of five (227).

133 Rhodes and Trent (227) document that $1.2 billion was expended for the restoration of flood damaged highway facilities during the 1980s. The damages, costs, and lost time resulting from bridge scour and stream instability during the 1992-1993 floods in the upper Mississippi River basin were extremely large. From available information on 23 bridge failures during the 1993 upper Mississippi River floods, 16 bridge failures were attributable to lateral channel migration or abutment failure, in which lateral migration may have been a contributing factor. An earlier study of 373 bridge failures in 1973 (226) indicated that 72 percent of the failures involved abutment damage. A more extensive study in 1978 (213) showed about 50 percent of the failures were from abutment problems, in which lateral channel migration may have been a contributing factor. Although it is difficult to be precise regarding the actual cost to repair damage to the nation's highway system from problems related to channel migration, the number is obviously very large. In addition, the costs cited above do not include the extra costs that result from over design of bridge foundations (deeper foundation depths, unnecessary or over designed countermeasures) that result from our inability to predict of stream instability and channel migration. This lack of knowledge often results in overly conservative design. A practical methodology to predict the rate and extent of channel migration could help reduce the cost of design, repair, rehabilitation and countermeasures for lateral channel instability. A screening procedure to identify stable meandering stream reaches would ensure that engineering and inspection resources are not allocated to locations where there is little probability of a problem developing. The limitations of the comparison technique for predicting channel migration are related primarily to the quality and availability of the aerial photography. There are no inherent limitations with the GIS measurement and extrapolation tool developed for this project. Training time will be required for technicians or engineers to become familiar with the tool, and the DOT will need the hardware and software (in this case ArcView) to implement the procedure. Indications are that most DOTs currently have access to or will soon acquire the necessary GIS capability. The Archive Data Base The archive data base on CD ROM includes all meander site data acquired for this study. With the archive data set produced by this project, future researchers will have a readily accessible data base in a very useable format for a variety of studies. These studies could include additional empirical analyses and more complex regressions based on the archive data. The Brice data alone, which is part of the archive data set, is an invaluable resource for future researchers, particularly as it includes the field measurements compiled by the U.S. Army Corps of Engineers Waterways Experiment Station for their study of stable channel design. Additional data could be added to supplement or complement the data base. As deterministic modeling code improves over the next decade, this archived data will facilitate calibration and verification of physical-process models of river meandering, providing additional tools for the highway hydraulic engineer beyond the empirical techniques of this research.

134 Implementation The Audience The target audience for the results of this research are hydraulic engineers and maintenance and inspection personnel in state, federal, and local agencies with a river-related responsibility. These would include in rough order or priority: • State Highway Agencies • Federal Highway Administration • City/County Bridge Engineers • State Departments of Natural Resources • U.S. Army Corps of Engineers • U.S. Bureau of Reclamation • Federal Emergency Management Agency • Natural Resource Conservation Service (SCS) • Consultants to the agencies, above • Academic researchers in river engineering and geomorphology Impediments to Implementation A serious impediment to successful implementation of results of this research will be difficulties involved in reaching a diverse audience scattered among numerous agencies and institutions; however, this can be countered by a well-planned technology transfer program. Because of the complexity of the meander migration process, the major challenge was to "package" the results in a form and format that can be used by a diverse audience with varying levels of technical sophistication. The Handbook, as a stand-alone document, provides a qualitative screening procedure and a range of photo comparison quantitative techniques from the relatively simple manual overlay approach to more sophisticated GIS-based measurement and prediction tools. With the guidance and examples contained in the Handbook, there should be something in these research results that will be of interest and assistance to almost every level of the primary target audience including: bridge inspectors, highway engineers, and practitioners in river engineering and geomorphology. Leadership in Application Because of its broad-based mission to provide guidance to the state highway agencies, the Federal Highway Administration must take a leading role in disseminating the results of this research. Through the National Highway Institute and its training courses, FHWA has the program in place to reach a diverse and decentralized target audience. The Transportation Research Board through its annual meetings and committee activities, and publications such as the Transportation Research Record, as well as periodic bridge conferences can also play a leading role in disseminating the results of this research to the target

135 audience. The numerous committees of the American Association of State Highway and Transportation Officials (AASHTO) can also assist in this regard. Finally, professional societies such as the American Society of Civil Engineers (ASCE) host conferences and publish peer reviewed journals through which the latest advances in engineering research and applications reach a wide audience, including many state, federal, and local hydraulic engineers. In this regard, the preliminary results of this research have already been presented by Research Team members at two ASCE conferences and the First International Conference on Scour of Foundations. An abstract has been submitted to the Sixteenth Hydrotechnical Conference, sponsored by the Canadian Society of Civil Engineering (CSCE) scheduled for the Fall of 2003. Activities for Implementation The activities necessary for successful implementation of the results of this research relate primarily to technology transfer activities as discussed above. FHWA/NHI have implemented the following: • The latest edition of Hydraulic Engineering Circular (HEC) 20, "Stream Stability at Highway Structures" (201) introduces the basic concepts of meander migration prediction using comparative aerial photography. • NHI Course #135046, "Stream Stability and Scour at Highway Bridges" includes a 90- minute demonstration workshop (Lesson 19) on the manual overlay meander prediction technique from the Handbook. Given that the next revisions to this training course may be three to five years in the future, FHWA requested and TRB approved adding a workshop on the Handbook methodology during the 2002 update of this course. FHWA/NHI should implement the following: • Include the results of this research in the next edition of HEC-20, "Stream Stability at Highway Structures." • Include the results of this research in the next edition (or supplement the current edition) of Hydraulic Design Series (HDS) 6, "River Engineering for Highway Encroachments" (224). • Add an instructional module on the Handbook procedures during the next revision of NHI course #135046, "Stream Stability and Scour at Highway Bridges." • Add a lesson on the Handbook procedures during the next revision of NHI course #135010, "River Engineering for Highway Encroachments."

136 Criteria for Success The best criteria for judging the success of this implementation plan will be acceptance of the methodology and techniques that resulted from this research by state highway agency engineers and others with responsibility for design, maintenance, rehabilitation, or inspection of highway facilities. Progress can be gaged by peer reviews of technical presentations and publications and by the reaction of state DOT personnel during presentation of results at NHI courses. A supplemental critique sheet could be used during NHI courses to provide feedback on the utility of the methodology and suggestions for improvement. The desirable consequences of this project, when implemented, will be more efficient design, maintenance, and inspection of highway facilities considering channel migration impacts, and more effective use of countermeasures against lateral channel instability. The ultimate result will be a reduction in the number of bridge failures and reduction in damage to highway facilities attributable to channel migration.