Temporal and Spatial Scaling: An Ecological Perspective
Over the past several years scientists have engaged in a wide array of investigations aimed at understanding the ecological consequences of climatic changes occurring over different temporal and spatial scales. Through these studies a great deal has been learned about the confounding methodological issues that arise when data characterizing climate impacts at one temporal or spatial scale are used to draw conclusions about potential impacts on a different scale. Such lessons are highly relevant to efforts to predict how disease pathogens and vectors will respond to climatic changes. In particular, insights from ecological studies can help us identify both the opportunities and the potential pitfalls of using studies of climate variability to predict disease impacts of long-term anthropogenic climate change.
BIOLOGICAL EFFECTS OF OBSERVED CLIMATE VARIABILITY
As discussed in Chapter 3, climate varies naturally on a wide range of temporal and spatial scales, and over the past century, the global climate has been gradually warming. Climate can also be manipulated under controlled experimental conditions to achieve variability at prescribed time scales. The following paragraphs provide an overview of the range of ecological phenomena that are observed to vary in response to these different forms of observed climate variability.
There are numerous ecological changes associated with spatial climate gradients. For instance, associated with latitudinal and altitudinal climate gradients are dramatic changes in soil fertility, species composition, growth rates, and
timing of reproductive cycles. The ecological consequences of regular diurnal and seasonal climate variability are also well studied and relatively predictable. Day/night and summer/winter temperature differences strongly constrain ecological processes and the population dynamics of disease vectors. Both excessively cold nights/winters and excessively warm days/summers can limit the geographic ranges and reproductive rates of insects as well as affect the rates of microbial activity in soil and water. Examples of plant and animal adaptations to diurnal and seasonal climate variability abound, including migration, hibernation, nocturnality, reproductive cycles, torpor, and leaf shedding. The constraining effects of diurnal and seasonal climate variability on the geographic ranges and the growth and reproductive rates of organisms are intertwined because the hot extreme comes in summer daytime and the cold extreme in winter nighttime.
Interannual to decadal scale climate variability results from coupled atmosphere-ocean processes, and possibly from sunspot cycles and other as yet unexplained driving forces. The ecological and health consequences of these variations can be considerable. For example, climatic variations associated with the ENSO, with multi-year droughts, and with multidecadal monsoon and hurricane cycles are known to correlate with vegetation productivity, bird-nesting success, abundance of insects, and numerous other ecological parameters (e.g., Levins et al., 1994; Tucker et al., 1991). On much longer time scales, the ecological changes caused by the “little ice age” and glacial-interglacial transitions and longer-term climate changes can be enormous, including huge range shifts or even extinction of species (Campbell and McAndrews, 1993; Davis and Zabinski, 1992; Webb, 1986).
A wide variety of ecological trends are statistically associated with the long-term warming trend that has been occurring over the past century. These include earlier arrival of spring (as marked by biological events such as egg laying and vegetation flowering); bird, mammal, and amphibian population declines; and range shifts in butterflies, birds, and marine invertebrates (Brown et al., 1999; Crick and Sparks, 1999; Grabherr et al., 1994; Myeni et al., 1997; Parmesan, 1996; Thomas and Lennon, 1999). Evidence that these changes are actually caused by trends in climate is compelling in some cases (e.g., observed trends in plant-flowering phenology) and at least suggestive in most of the other cases. At many locations, nighttime and winter air temperatures during the past 100 years have increased more than have daytime and summer temperatures. These reductions in the amplitudes of diurnal and seasonal cycles represent potential influences on ecological phenomena, independent of any effects of a change in mean temperature. Organisms are generally more sensitive to temperature extremes than they are to mean temperature, and these amplitude changes could generate geographic-range shifts and altered population densities of a variety of species. The duration of extreme temperature episodes can in some cases be as ecologically important as the extreme temperatures reached during such episodes.
Climate can also be altered by deliberate experimental manipulation to learn about ecological responses to these alterations. Climate manipulation experiments can endure for periods from less than a year to over a decade. They can be carried out either in small laboratory chambers or in field plots of varying sizes. In these experiments, soil and/or air temperature are manipulated with the use of overhead heat lamps or heating wires in the soil or with passive devices that serve as small open-top greenhouses (Shen and Harte, 2000). In some field experiments precipitation is also manipulated, and in a few cases temperature increase is combined with an increase in ambient carbon dioxide concentration to more accurately simulate the conditions of future atmospheric conditions. Experiments of these types can affect the relative growth rates of plant species, the population of invertebrates, the timing of plant reproductive cycles, biogeochemical cycling rates, and the quantity of carbon stored in organic form in soil (Harte and Shaw, 1995; Saleska et al., 1999).
CONFOUNDING INFLUENCES ON ECOLOGICAL FORECASTING
By investigating how a target parameter such as disease vector abundance responds to observed climate variability, it may be possible to make reliable deductions about how that target parameter will respond to future anthropogenic global climate change (AGCC), either by direct statistical extrapolation of the observed relationship or by a more complex analysis based on a mechanistic understanding of these relationships. Here we discuss the factors influencing the validity of the conclusions drawn from such investigations and the opportunities that this approach to prediction provides. Our purpose is to provide a framework for evaluating the suitability of the different empirical and theoretical approaches that have been proposed to predict the impacts of AGCC.
Mismatches in temporal scale impede efforts to predict. Ecological adaptations to glacial-interglacial and other long-term climate cycles occur on a time scale that is much longer than the time scales of concern for AGCC. Since climate is not the only environmental parameter that changes over these long time scales, ecological responses to AGCC are not necessarily predictable based on the insights gained from paleoclimatic investigations. For instance, factors such as soil quality constrain plant species composition and growth rates, and the mechanisms shaping soil quality operate over time scales that are relatively long compared to the time frame of AGCC. Thus, over sufficiently long time intervals, plant distributions are likely to correlate with climate change, but those correlations may not provide much insight into short-term responses of plants to AGCC.
Attempts to deduce the potential impacts of AGCC based on the impacts of diurnal, seasonal, interannual, or decadal climate variability suffer from related problems, since in these cases the time frame is inappropriately short compared
to that characterizing AGCC. Many organisms' traits are genetically adapted to short-term diurnal and seasonal climate cycles and are not likely to provide reliable indication of responses to AGCC. For instance, ecological responses to the ENSO cycle may not always be indicative of responses on the time scale of AGCC. Changes in ecosystem productivity that are observed to accompany the ENSO cycle are, in part, constrained by the species composition of these ecosystems; and major transitions in species composition occur over century to millennial time scales, not in response to individual El Niño/La Niña events. Finally, extrapolation across these different time scales can be confounded by the fact that anthropogenic stresses on ecosystems such as land-use changes, resource exploitation, and population growth are likely to change more significantly over the course of decades to centuries than over seasonal to interannual time scales.
Correlation in time does not imply causation. The problem of establishing causation besets efforts to deduce future responses to AGCC from observations of ecological responses to the past hundred years of climate change. The correlations between ecological and climatic time series could be coincidental if, for example, third-party factors such as land-use changes are simultaneously driving both climate and ecosystem change. Only manipulated climate change experiments provide an unambiguous way to distinguish causality from correlation; but such experiments are intrinsically limited to temporal and spatial scales that may be too small to provide reliable predictions.
Correlation in space does not imply causality at AGCC time scales. By examining the spatial correlation between climate variations and ecosystem parameters, deductions can be made about how ecosystems will respond to changes in climate over time. The validity of those deductions, however, depends on two premises. The first is that the changes in ecological parameters along spatial gradients are actually driven by the associated climate gradients. The second, called the “space-for-time” assumption, is that the mechanisms of ecological change over time (on decadal to century time scales characteristic of AGCC) are sufficiently similar to the mechanisms that create ecological gradients resulting from spatial climate variability. Because the latter operate over much longer time scales (typically many centuries to millennia) than the former, and since climate is not the only environmental parameter that varies along spatial climate gradients, the space-for-time assumption is not necessarily valid, although it does appear to hold true in some cases. For example, experimental studies have shown that variations in plant phenology (timing of the reproductive cycle) along an elevational gradient in montane meadow habitat provide a remarkably accurate prediction of the phenological response of plants to manipulated climate warming (Price and Waser, 1998). On the other hand, the responses of a number of other ecological variables (including soil organic matter, plant productivity,
Ecological and climatic phenomena are dependent on spatial scale. Because of the relatively small size of experimental field plots and laboratory chambers used for climate manipulation experiments, observed ecological responses may be quite different from those that would occur in large ecosystems. For example, studies of the influence of climate change on soil fertility and plant productivity on experimentally warmed plots of area about 10 m2 could not possibly capture the influence on plants and soil of changing populations of large herbivores.
Linear extrapolation is often misleading in ecology. Ecological phenomena often depend on climate parameters in highly nonlinear ways, and thus the differences in magnitude between natural climate variability cycles and AGCC can confound efforts at prediction. The mismatch in time scales necessitates the application of reliable interpolation procedures to predict the effects of a relatively small change in temperature (AGCC) based on understanding of the effects of a larger change (e.g., a diurnal or seasonal cycle). For example, many insects in temperate or cold climates are active during daytime and go into nighttime cold-induced torpor. The dependence of insect nighttime activity rates on air temperature is not well characterized, but is thought to be a threshold relationship. Thus, it is unlikely that a simple linear interpolation of activity between day and night temperatures would yield reliable predictions of activity rates at the new nighttime temperatures resulting from AGCC (which would likely be intermediate between current day and night temperatures).
and species composition) to manipulated climate change were not well predicted by their patterns of change along spatial climate gradients (Dunne, 2000).
For all of the reasons discussed above, ecological studies point to both pitfalls and possibilities in the use of observed climate variability data to forecast the effects of AGCC. Each type of dataset possesses shortcomings that reduce the opportunity to draw unambiguous conclusions. Were a combination of such datasets (e.g., from interannual variability, spatial gradients, and responses to experimentally manipulated climate) all to point to a single consistent conclusion about ecosystem response to climate variations, this would greatly enhance confidence in the predictions about responses to AGCC. Even when results from different kinds of measurement yield diverging predictions (because of differing spatial and temporal scales), the use of models that explicitly include scale-dependent mechanisms may allow reconciliation of the differing conclusions and provide the insight needed to draw defensible predictions.
The types of scaling difficulties faced in the study of ecological changes are highly relevant to the study of infectious diseases, especially in cases where the transmission cycle is closely associated with ecological changes. For instance,
in the urban dengue system breeding occurs primarily in man-made containers, an “environment” that isn't expected to change significantly in response to climatic changes. Under these circumstances, responses to ENSO events or to artificially manipulated climate could provide reliable insight into the potential consequences of AGCC. In contrast, in the trypanosomiasis/tsetse fly system of sub-Saharan Africa, vector dynamics are critically influenced by natural ecosystems that are subject to uncertain change in the face of AGCC, and in this example there may be scaling problems inherent in using observed climate variability responses to anticipate AGCC responses. Malaria may present an “intermediate” example, since the environments that foster its transmission range from quasi-stable ecosystems (e.g., rice-growing regions) to more complex situations where vector dynamics are intimately tied to the natural environment.