* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download 8_ecosystem-models-for-gcc
Global warming wikipedia , lookup
Fred Singer wikipedia , lookup
Public opinion on global warming wikipedia , lookup
Attribution of recent climate change wikipedia , lookup
Climate change and agriculture wikipedia , lookup
Climatic Research Unit documents wikipedia , lookup
Climate sensitivity wikipedia , lookup
Low-carbon economy wikipedia , lookup
Climate-friendly gardening wikipedia , lookup
Surveys of scientists' views on climate change wikipedia , lookup
Climate change and poverty wikipedia , lookup
Climate change, industry and society wikipedia , lookup
Years of Living Dangerously wikipedia , lookup
Effects of global warming on Australia wikipedia , lookup
Carbon Pollution Reduction Scheme wikipedia , lookup
Solar radiation management wikipedia , lookup
Citizens' Climate Lobby wikipedia , lookup
Numerical weather prediction wikipedia , lookup
Global Energy and Water Cycle Experiment wikipedia , lookup
Politics of global warming wikipedia , lookup
Climate change feedback wikipedia , lookup
IPCC Fourth Assessment Report wikipedia , lookup
Business action on climate change wikipedia , lookup
Validating models of ecosystem response to global change For the dificulty in geting full text, here I just give the full text of this great paper. Validating models of ecosystem response to global change., By: Rastetter, Edward B., Bioscience, 00063568, Mar1996, Vol. 46, Issue 3 How can we best assess models of long-term global change? Short-term test are unable to address the slow-responding mechanisms likely to dominate long-term responses to climate and carbon dioxide change Models are an essential component of any assessment of ecosystem response to changes in global climate and elevated atmospheric carbon dioxide concentration (Reynolds et al. 1996). The problem with these models is that their long-term predictions are impossible to test unambiguously except by allowing enough time for the full ecosystem response to develop. Unfortunately, when one must assess potentially devastating changes in the global environment, time becomes a luxury. Therefore, confidence in these models has to be built through the accumulation of fairly weak corroborating evidence rather than through a few crucial and unambiguous tests. The criteria employed to judge the value of these models are thus likely to differ greatly from those used to judge finer scale models, which are more amenable to the scientific tradition of hypothesis formulation and testing. A recent exchange in the literature has rekindled the debate over model testing in the earth sciences (Oreskes et al. 1994, Rykiel 1994). The essence of the debate is the so-called problem of induction, that is, the problem of extrapolating from the specific to the general (Popper 1959). Because of the problem of induction, no specific test or series of tests can establish the general validity of any model. Modern solutions to the problem of induction all rely on severe model tests. For example, Popper's (1979) falsification solution builds confidence in the validity of a hypothesis (model) through repeated attempts to falsify it with severe and crucial tests. Platt's (1964) method of strong inference also depends on an unequivocal test capable of eliminating one or more hypotheses from among several alternatives. The unequivocal tests required by modern scientific methods cannot be used to evaluate models of ecosystem response to changes in climate and carbon dioxide concentration (ERCC models). All such tests prove either irrelevant or ambiguous unless enough time is allowed for the full ecosystem response to develop. Because of the impossibility of rigorous tests, Weinberg (1972) would classify the questions addressed by ERCC models as "trans-scientific." That is, "though they are, epistemologically speaking, questions of fact and can be stated in the language of science, they are unanswerable by science; they transcend science" (Weinberg 1972, p. 209). The resolution of trans-scientific questions requires the synthesis of indirect scientific information from diverse sources. I believe that the major role of ERCC models is to provide this synthesis. Four categories of tests potentially could be used to evaluate ERCC models: tests against short-term data, space-for-time substitutions, reconstruction of past responses to climate change, and comparison with other models. I believe that these four categories cover all possible tests except the ultimate test of waiting the requisite time for the full response to global climate change to develop. I address each category in turn and illustrate why not one presents a means to construct severe and crucial tests. I then discuss the synthesis role of ERCC models and why they are vital to any assessment of long-term responses of ecosystems to changes in global climate and carbon dioxide concentration. Tests against short-term data Data derived from short-term (less than 10 years) experiments are often used to test models designed to simulate long-term (20-200years) ecosystem behavior. Two critical questions emerge as to the relevance of these tests: Is corroboration by short-term data sufficient to justify confidence in long-term projections made with the model? Is corroboration by short-term data necessary to justify confidence in long-term projections made with the model? The answer to both questions is no. The arguments for both aspects of relevance are based on an analogy to a well-corroborated photosynthesis model--the Farquhar model (Farquhar and von Caemmerer 1982). In each case, the analogy is developed by redefining short term and long term. By applying the Farquhar model at three different time scales (milliseconds, seconds to minutes, and weeks), I illustrate the problems inherent in testing a model at a scale other than that at which it was intended to be applied. In the Farquhar model, carbon assimilation is calculated as the minimum of the rate of ribulose bisphosphate (RuBP) carboxylation and the rate of RuBP regeneration. Variations on this model have dealt either with alternate equations to calculate the effect of quantum flux density (e.g., Harley et al. 1992, McMurtrie et al. 1992), ways to couple it to a stomatal model (e.g., Ball et al. 1987), or with additional constraints imposed by other factors, such as phosphorus concentration (e.g., Harley et al. 1992). Nevertheless, the Farquhar model is a well-tested, well-corroborated model of the responses of C3 plants(n1) to changes in carbon dioxide concentration and light on a temporal scale of seconds to minutes (e.g., Collatz et al. 1990, Farquhar and yon Caemmerer 1982, Harley et al. 1992). Sufficiency. Short-term tests are never sufficient to test long-term models, because slow-responding mechanisms that dominate long-term behavior can often be neglected in the short term. To illustrate this problem, I use the corroboration of the Farquhar model at the time scale of seconds to minutes as the analog for corroboration of ERCC models against data from experiments lasting less than ten years. I use three-week projections made with the Farquhar model as the analog to 20-to 200-year projections made with ERCC models. Is this strong corroboration on a seconds-to-minutes time scale (short term) sufficient to justify confidence in using the model to project responses of leaves to changes in carbon dioxide and light on a scale of weeks (long term)? Consider the results of a doubled carbon dioxide experiment conducted by Tissue and Oechel (1987). When leaves of Eriophorum vaginatum that had been grown at 340 ppmv carbon dioxide were suddenly subjected to 680 ppmv carbon dioxide, net photosynthesis increased just as the Farquhar model predicts (Figure 1a). However, after approximately three weeks of exposure to 680 ppmv carbon dioxide, net photosynthesis decreased back to its original rate. The Farquhar model cannot simulate this decrease in net photosynthesis because the mechanisms responsible for the acclimation to higher carbon dioxide, concentrations (e.g., a decrease in the maximal rate of carboxylation) are not represented. This application of the Farquhar model on a time scale of weeks is unfair--it was never intended to be applied at that scale. However, it does illustrate the point: Corroboration of a model against short-term data (seconds to minutes in this case) does not justify confidence in longer term projections with the model (weeks). The potential always exists that long-term behavior is affected by slow-responding mechanisms whose effects can be neglected in the short term. Short-term data are never sufficient for testing how well these slow-responding mechanisms are represented in the model (Reynolds and Leadley 1992). Thus, even if the Farquhar model were coupled to an acclimation model, the short-term (minutes) data would not be sufficient to test how adequately the acclimation mechanisms were represented in the model. Necessity. Short-term tests are not necessary to test long-term models because the fast-responding mechanisms that dominate short-term behavior are often not important to long-term behavior. To illustrate this case, let us use manipulations at a time scale of milliseconds as the analog for less-than-10-year experiments and the seconds-to-minutes projections made with the Farquhar model as the analog to the 20to 200-year ERCC model projections. Emerson and Arnold (1932) found that the rate of photosynthesis for Chlorella pyrenoidosa cultures exposed to a bright light that flashed once every 20 msec was no slower than that of cultures exposed to continuous bright light. By adjusting the relative lengths of the light flashes and the interspersed dark periods, they found that they could change the efficiency of photosynthesis per unit light quanta. With a 4-millisecond flash followed by a 16-millisecond dark period the efficiency increased fourfold over that of continuous light (Figure 1b). The Farquhar model is unable to simulate this increase in photosynthetic efficiency because, again, its structure does not incorporate the short-term mechanisms controlling the rapid charge and discharge of chlorophyll and the short delay as energy is transferred between the light and dark photosynthetic reactions. However, does this falsification of the Farquhar model with short-term data (milliseconds) preclude its use to model the long-term responses for which it was designed (seconds to minutes)? Again, applying the Farquhar model on this fine time scale is unfair, but it illustrates that falsification of a model against short-term data does not mean that the model is unreliable for making longer term projections. The Farquhar model can still be used reliably to predict photosynthetic responses to changes in carbon dioxide and light on time scales of seconds to minutes. Clearly corroboration of a model with short-term data is neither sufficient nor necessary for making reliable longer term projections. This conclusion was obvious for the Farquhar model, because in each case I had access to both the short-term and long-term data. In the case of ERCC models, the long-term data are not available, but the conclusion nevertheless holds. Data are relevant for testing a model only if the time scale of the mechanisms represented in the model and the time scale of the data are the same. It is likely that ERCC models have both short-term (less than 10 years) and long-term (20-200years) mechanisms. Shortterm experimental data currently available are applicable only for testing the short-term mechanisms in ERCC models. The long-term mechanisms (e.g., those involving slow turnover of woody tissues and soil) cannot be tested in this way. Unfortunately, the long-term mechanisms are likely to dominate the long-term responses of ecosystems to changes in global climate and atmospheric carbon dioxide concentration. To place this conclusion in the context of ERCC models, consider a model (Marine Biological Laboratory's General Ecosystem Model, MBL-GEM; Rastetter et al. 1991) calibrated and tested using data from a nineyear experimental manipulation of temperature, nutrients, and light (Chapin et al. 1995).(n2) I applied this model to make 50-year projections, assuming that the 9-year experimental data were sufficient to constrain longer term mechanisms represented in the model. Simulations were run to examine the responses of arctic moist tundra to the combined effects of a 5 Celsius increase in temperature and a 10% decrease in soil moisture (Rastetter et al. in press). Both treatments were expected to increase the rate of decomposition in this waterlogged ecosystem (Oechel et al. 1993). Thus the initial effect in the simulation was a release of soil carbon as carbon dioxide. However, increased decomposition also releases mineral nitrogen, which becomes available to vegetation and stimulates productivity in this strongly nitrogen-limited ecosystem (Chapin et al. 1995, Shaver and Chapin 1991, Shaver et al. 1992). The simulations indicate that for approximately 20 years the rate of nitrogen uptake by vegetation is approximately equal to the mineralization rate of nitrogen from soils. Thus, nitrogen lost from the soils is accumulated in vegetation, and little of the mineralized soil nitrogen is lost from the ecosystem. However, after approximately 20 years, the vegetation becomes dense enough to intercept all usable light. Further increases in vegetation are therefore limited by the availability of light needed for photosynthesis. Nitrogen can then no longer be used by vegetation at the same rate that it is mineralized in the soil, and the loss rate of nitrogen from the ecosystem increases markedly (Figure 2). The ramifications of this so-called nitrogen breakthrough for terrestrial ecosystems downslope, aquatic ecosystems downstream, and nitrogen-based greenhouse gases in the atmosphere could be important in the context of global change. However, can short-term experiments validate the result? Clearly tests can be made on the short-term responses of the model like the initial increase in decomposition and nitrogen mineralization and the increase in productivity and rate of nitrogen uptake by vegetation (e.g., Chapin et al. 1995).(n3) However, the key to the predicted result is the decoupling of soil nitrogen release and vegetation nitrogen uptake, which, according to the model, does not happen for approximately 20 years. Unanticipated slow-responding mechanisms could easily change the predicted result. No short-term data could uncover the omission of such mechanisms from the model. Even if all the important mechanisms are in the model, the short-term data are not sufficient to constrain the parameters controlling long-term responses, because errors in these parameters would have little effect on short-term predictions. Therefore, the result, although important and deserving of due consideration when assessing responses of tundra to global climate change, cannot be corroborated with short-term data. Space-for-time substitutions Space-for-time substitutions are among the most powerful means available for studying succession (i.e., the progressive change in plant communities at a site following a major disturbance). Studies employing this method have, for example, examined ecosystem characteristics of a series of plots in old agricultural fields abandoned 75, 42, 34, 22, 11, 3, 2, and i year(s) ago (e.g., Oosting 1942); plots from which a glacier retreated 180, 120, 100, 70, 35, 31, 29, 23, and 11 years ago (e.g., Crocker and Major 1955); or lava flows that are 47, 137, 300, 400, and 3000 years old (e.g., Drake and Mueller-Dombois 1993). In this way, space (location of the plot) is substituted for time (time since abandonment or glacial retreat or age of lava flow). Fastie (1995) has pointed out the weaknesses of this space-for-time substitution, but the approach has nonetheless been powerful in the development of theories of succession. Responses of ecosystems to changes in global climate and atmospheric carbon dioxide concentration also can be viewed as a type of succession. Therefore, it seems reasonable to expect that a similar approach could address climate change issues and, in particular, test ERCC models. If the climate at one location is expected to change so that it resembles the climate at another location, then the ecosystem characteristics (e.g., species composition and biogeochemistry) at the original location might be expected to change so that they resemble the present-day characteristics of the second ecosystem. Of course, this hypothesis must itself be tested. However, for the present purposes, I assume that the hypothesis is valid. There are two problems with space-for-time substitutions in the context of testing ERCC models. First, it is difficult to find regions with carbon dioxide levels as high as those projected for the next century. If the direct response to carbon dioxide is an important aspect of ecosystem response to changes in climate and carbon dioxide, then the space-for-time substitution is difficult to apply. Second, unlike the case for succession, applying space-for-time substitutions to changes in climate involves a comparison with ecosystems that have come into balance with the local climate. Consider the Amazon basin. With a doubling of carbon dioxide, it is projected that temperatures in this region could increase by as much as 4 Celsius within 60-80 years, that soil moisture could decrease by more than 2 cm, and that rainfall (and cloudiness) could change substantially (Bretherton et al. 1990). Although these changes are large, the present-day range of climatic conditions within the Amazon are of approximately the same magnitude. We have used this range in present-day climatic conditions to constrain and test the equilibrium predictions of MBL-GEM (McKane et al. 1995). However, unlike the abandoned fields or the retreating glacier, there are no places within the Amazon or elsewhere where the climate has changed by this magnitude within the last 20200 years. The ecosystems across the climatic gradient have had hundreds or thousands of years to adjust to the respective climates, and the available data can therefore only be used to test the long-term, equilibrium predictions of the model. Even if the underlying assumption that ecosystems eventually come to resemble one another if placed in similar climates is valid, there is no information in this space-for-time substitution to indicate how the ecosystem characteristics change through time. The data only indicate the final state. Responses of terrestrial ecosystems to climate change involve slow changes in species composition, the growth of long-lived tree species, and the slow turnover of soil organic matter. These responses can therefore take centuries. If the goal is to model the 20-200 year responses, then the space-for-time substitution does not yield the necessary information. There could be many ways by which the ecosystem might arrive at a new equilibrium state under a new climate. Thus, the second problem with space-for-time substitutions is exactly the opposite of that of the short-term tests. Therefore, the same arguments can be applied in reverse to the relevance of very-long-term (i.e., equilibrium) tests of long-term models. That is, there easily could be unanticipated mechanisms acting on a 20- to 200-year time scale that would not be obvious from an examination of equilibrium data. No equilibrium data could uncover the omission of such mechanisms from the model. Reconstruction of the past There have been significant changes in climate and carbon dioxide concentration in the recent past (last 200 years; e.g., Boden et al. 1994, Fritts and Lough 1985, Graumlich and Brubaker 1986). On first inspection, these changes present the ideal opportunity for testing ERCC models. Not only would such tests involve responses to many of the same factors that are of concern for the future, but the responses would be at the appropriate time scale. However, other problems with the data preclude their use for severe and crucial tests of ERCC models. The most obvious problem is the potential for nonlinear responses to changes in climate and carbon dioxide. Because of these nonlinearities, there is no guarantee that tests against the range of climate and carbon dioxide changes in the past will apply to the range expected in the future. Short-term tests can be used to gain confidence that the nonlinearities in fast-responding processes are well represented, but the slow-responding processes remain problematic. A second problem is that the magnitude of the response of ecosystems to changes in climate and carbon dioxide thus far is too small to be discernible given the quality of available data (Mooney et al. 1991). That is, there is a signal to noise problem. This problem is exacerbated by the inherently stochastic nature of many ecosystem processes, such as seed dispersal and tree mortality. Models that incorporate these stochastic processes have correspondingly broad confidence ranges in their predictions (e.g., Shugart and West 1980). A consideration of the atmospheric carbon budget illustrates the magnitude of the signal-to-noise problem. Each year approximately 7.0 +/- 1.1 x 1015g carbon is released to the atmosphere through fossil fuel burning, cement manufacturing, and changes in land use (Watson et al. 1990). Of this amount, 3.4 +/- 0.2 x 1015g accumulates in the atmosphere, 2.0 +/- 0.8 x 1015g is taken up by ocean water, and 1.6 +/- 1.4 x 1015 g is removed from the atmosphere by some unknown process. The 1.6 x 1015g carbon that is removed by an unknown process is often called the missing carbon (Watson et al. 1990) and is often thought to be stored in terrestrial ecosystems as a result of an imbalance between net primary productivity and soil respiration (e.g., Kauppi et al. 1992). If the missing carbon is entering terrestrial ecosystems, it represents a storage rate equal to only approximately 3% of the net primary production of all terrestrial ecosystems (Melillo et al. 1993). Similarly, the cumulative missing carbon since 1850 is only 70-95 x 1015g carbon (Houghton 1993, 1995), or approximately 4% of the total carbon in soils and vegetation around the world (Schlesinger 1991). If this rate of carbon accumulation is distributed among terrestrial ecosystems in proportion to their respective rates of net primary production, then this rate is far too small to be detected as either the difference between net primary production and soil respiration or as a cumulative increase in ecosystem carbon stocks. Even if all the storage were restricted to temperate and boreal forests (Francey et al. 1995), it would equal only approximately 15% of net primary production in these ecosystems and would be difficult to detect. The detection problem is exacerbated by past disturbances (e.g., fire or logging) in many ecosystems. Past disturbances make it difficult to discern how much of the present-day carbon accumulation rate is attributable to recovery from disturbance and how much is attributable to changes in climate and carbon dioxide concentration (e.g., Rastetter and Houghton 1992). Even if the present-day rate of carbon accumulation associated with changing climate and carbon dioxide could be measured, it would not be enough to test a model designed to simulate long-term carbon accumulation. To test the model, information would also be needed to initialize it for preindustrial conditions. Any error in this initial estimate would propagate through the simulation and corrupt predictions throughout. The problems with estimating present-day carbon stocks and flux rates are amplified for estimates of these quantities in the past. I know of no reliable estimates of preindustrial organic matter stocks in soils and vegetation. To get the timing of ecosystem response correct, information would also be needed at several times during the reconstruction, but such records of changes in ecosystem structure and function through time are poor (Mooney et al. 1991). In addition, time-series data on climate and carbon dioxide concentration would be needed to drive the model. The carbon dioxide data are readily available from Mauna Loa measurements (Keeling and Whorf 1994) and the Sieple (Neftel et al. 1994) and Greenland (Wahlen et al. 1991) ice-core records and are of high quality. Temperature records, on the other hand, are only available for some locations and often do not extend back more than a few decades or are reported from fairly broad regions (e.g., Karl et al. 1994, Wilson and Hansen 1994). Temperature records can also be inferred from tree rings, but the correlation coefficients (r2) for even high-quality tree ring temperature reconstructions are typically less than 50% (e.g., Graumlich and Brubaker 1986). Clear-day irradiance is readily calculated from latitude and time of year (e.g., McMurtrie 1993), but climatic changes in cloudiness can greatly influence both photosynthetically active and total radiation. I know of no way to reliably reconstruct cloudiness. Soil moisture models require not only precipitation data, which might not be available in all areas, but radiation and temperature data as well, with the associated difficulties just mentioned. Thus there are major problems with the input data needed to model ecosystem responses to changes in climate over the last few hundred years. Any errors in these data would propagate through the model and corrupt the resulting reconstruction. In the case of reconstructed responses to past changes in climate and carbon dioxide concentration, there could easily be either unanticipated mechanisms or poorly constrained representations of known mechanisms acting at a 20- to 200-year time scale. Available data could not uncover these problems in the model because the responses thus far have been too small to measure. Even if these responses could be measured, however, discrepancies between model predictions and the measurements could easily be attributed to errors in the initialization data and the data used to drive the model rather than to an inherent error in the model structure or parametrization. The best that can be done with the available data is to put an upper limit on the magnitude of responses in the reconstruction. Little can be said about the timing or direction of these responses. Comparison with other models Because of the uncertainty associated with global circulation models, it has become commonplace to publish side-by-side climate predictions from several global circulation models (e.g., Mitchell et al. 1990). The implication is that the uncertainty is reduced if a prediction from one global circulation model agrees with the predictions from other models. It is only a small logical step from this implied decrease in uncertainty to an implication of corroboration among models. Of course, there is the problem of independence to resolve: If the models are based on the same underlying concepts, would they not be expected to make similar predictions? The flip side of the issue is more difficult: If two models do not agree, which of the two is false, or are both? Most empiricists and many modelers are rightly uncomfortable with this logic of testing models against models. It clearly should not be considered a severe and crucial test. However, there is some merit in the logic. Consider two cases: If a tentative model is compared with an established model about which there is a high degree of confidence, then surely some of that confidence is transferred if the two models' predictions agree. In this case, the established model is simply acting as a synthesis of the data with which it had already been compared and from which the high degree of confidence is derived. If the tentative model agrees with the established model, then it should also agree with the data against which the established model was corroborated. Within the realm of global climate and carbon dioxide change, it is more common to compare models about which there is equal skepticism (e.g., global circulation models). Each model in such a comparison is likely to have been developed from an underlying set of established or hypothesized principles (e.g., mass balance, resource limitation, competitive interactions). Agreement among models in this case means that the results of one model do not conflict with the principles underlying the other models. For example, one model might be based on biogeochemical principles of mass balance and the interactions among carbon and nitrogen cycles (e.g., MBL-GEM; Rastetter et al. 1991). Another model might be based on the principles of competition among individual trees for light and soil resources (e.g., Shugart and West 1980). If the two models agree on the accumulation rate of carbon in the vegetation, then the biogeochemical model has not conflicted with the constraints of tree-to-tree interactions in the competition model, and the competition model has not conflicted with the constraints of mass balance and carbon-nitrogen interactions in the biogeochemical model. The degree of confidence that emerges from such a comparison depends on how well the underlying principles in each model have been established and how independent these principles are among the models. That confidence should build as the predictions are confirmed by more and more models with widely different underlying principles. For example, confidence that carbon is likely to accumulate in terrestrial ecosystems under projected changes in climate and carbon dioxide concentration should be greater if predicted by ten different models than if predicted by only one or two models based on similar underlying principles. The important property of this consensus-building approach is that the models be independent and based on widely different underlying principles. It makes little sense to compare a model with another that is based on the same underlying principles, because the model is essentially being compared with itself. A weakness of the consensus-building approach is that it is not clear what a lack of consensus means. For example, if the biogeochemically based model and the competition-based model disagree on the rate of carbon accumulation, neither can be said to be falsified. There is no way of telling from an analysis of the models whether the long-term carbon dynamics of the ecosystem are likely to be governed by the biogeochemical constraints or the constraints imposed by tree-to-tree interactions. Similarly, if six out of seven models agree on a particular prediction, there is no way to tell whether the principles underlying the one nonconforming model are the ones that actually govern the long-term dynamics of the ecosystem. However, such a comparison of models and an analysis of their underlying differences should suggest productive lines for further inquiry. In particular, the comparisons can identify mechanisms that were overlooked or omitted from the model but nevertheless have been found to be important in other models (Oreskes et al. 1994). Only a few formal comparisons have been made among ERCC models (e.g., Ryan et al. in press, VEMAP members 1995). For example, Ryan et al. (in press) compared seven models based on their predictions of long-term responses of pine plantations (Pinus radiata and Pinus sylvestris) to increased temperature and carbon dioxide concentration. The models were classified into two groups based on their predictions. The physiologically based models predicted a decrease in productivity with increased temperature because of increased water stress and consequent stomatal closure. In contrast, the nutrient-oriented models predicted increased productivity with increased temperature because of increased rates of nutrient mineralization and consequent increases in nutrient availability. Because none of the seven models incorporated both the physiological aspects of stomatal control and the recycling of nutrients from soil organic matter, there was no way to evaluate which of these two mechanisms would dominate the long-term responses of the forests. Ryan et al. (in press) call for "experiments ... designed to resolve this issue" and models linking the pertinent aspects of both classes of model to "help explore the trade-offs." Inasmuch as the exercise reported by Ryan et al. (in press) identified a potentially important omission in both sets of models, this comparison constituted an important test. However, the test could not falsify or validate any of the models. Synthetic value of ERCC models The true value of ERCC models is in their synthesis of available information. An analogy to Weinberg's (1972) discussion of engineering as trans-science illustrates that value. Weinberg (1972) used large engineering projects, such as the design of a major dam, as an example of a trans-scientific activity, because the design "typically involves decisions made on the basis of incomplete data." An assessment of ecosystem response to global change has much in common with a large engineering project. In both cases, tests on a full-scale prototype are impractical, long-term performance of the system must be evaluated, and decisions must be made within a rigid time schedule based on "whatever scientific data are at hand" (Weinberg 1972). The scientific data at hand encompass a great deal of knowledge about the components of the systems being analyzed. Those components are often small enough and respond rapidly enough that their independent performance can be evaluated using traditional scientific tests. The scientific data at hand also include observations on similar systems in other situations, past and present. The assessment must synthesize all this information so that well-informed decisions can be made. The role of ERCC models is to provide this synthesis, a role that is often overlooked. A great deal is already known about how ecosystems function, at least qualitatively. For example, the metabolic responses of plants and soil microorganisms to temperature are known; the response of photosynthesis to light and carbon dioxide concentration for most species is known; that plants compete for light and nutrients is known. As skeptics, scientists tend to concentrate on the conjectural content of ERCC models and do not give enough credit to their synthetic content. Synthesis is the most important aspect of ERCC models. Because of this synthesis, ERCC models surpass any other method of projecting responses to changes in climate and carbon dioxide concentration. For example, projections could be based directly on the results of short-term experiments. However, this method would be unreliable for longer term projections for exactly the same reasons that the short-term experiments cannot be used as severe tests of ERCC models. The same is true of projections based on space-for-time substitutions and assessments based on past responses to climate and carbon dioxide change. However, a model can be constrained to be simultaneously consistent with all these empirical sources of relevant evidence. That is, it can be used to build confidence through the synthesis of many lines of empirical evidence, each of which might not be compelling in and of itself. The synthesis within an ERCC model is more than just a compilation of independent information. Because the model places the available information within the same context, the synthesis provides a test of the "internal consistency of the system [model]" (Popper 1959). That is, do the pieces logically fit together? For example, can the increased rates of growth predicted by greenhouse carbon dioxide experiments be maintained by the rates of nutrient supply predicted by experiments on the turnover of soil organic matter? The formal structure of a model is conducive to the resolution of this type of question. Because of the complexity of interactions among the many components of an ecosystem, it would be difficult to develop a self-consistent synthesis of the available information without a model. Passing this test of internal consistency validates the model as a synthetic tool, if not as a predictive tool. Conclusion I have examined four ways to evaluate ERCC models and illustrated why none of them can provide severe and crucial tests of those models. Short-term empirical tests are unable to address the slow-responding mechanisms likely to dominate long-term responses to climate and carbon dioxide change. Space-for-time substitutions can only address the equilibrium responses to climate change and cannot address responses to changes in carbon dioxide concentration. Reconstruction of responses to past climate change are ambiguous because of the uncertainties in the data needed both to drive the model and to test it. Comparisons among models can be used to evaluate the relative importance of various processes in determining the long-term responses to global climate and carbon dioxide change, but they can never substitute for tests against realworld data. The unfortunate conclusion from this analysis is that ERCC models cannot be rigorously tested. The whole topic of ecosystem responses to global climate and carbon dioxide change is therefore transscientific. There are two potential ways the topic might be removed from the trans-scientific realm. First, methodological or technical breakthroughs could improve the precision of data needed to drive and test ERCC models through a comparison with past responses to climate and carbon dioxide. I believe that the order of magnitude improvement that would be needed is unlikely. Second, responses to continued changes in climate and carbon dioxide in the future are likely to be strong enough to make these responses detectable even with currently obtainable precision. However, by the time the changes are detected in many ecosystems, it may be too late for effective mitigation efforts. Therefore, the focus of both experimental and monitoring efforts should be on ecosystems that are expected to respond rapidly. Identification of these ecosystems should be a high priority and could be aided by the analysis of ERCC models. Although the prospect for testing ERCC models is grim, these models are likely to remain a vital part of any evaluation of the responses to changes in global climate and carbon dioxide because the alternatives are worse. It would be foolhardy to plunge blindly into the future without some attempt to evaluate the global consequences of human activities. Evaluations should not be based solely on data from short-term experiments, space-for-time analogs, or reconstructions of the past. Such evaluations would suffer from the same shortcomings that limit the use of these data for testing long-term model predictions. The advantage that a model-based evaluation has over all of these empirically based extrapolations is that it can be used to synthesize all of the empirical evidence. The synthesis in the model provides a means of testing the self-consistency of the interpretations of empirical evidence. Therefore, the model is a vital part of any evaluation of the responses of ecosystems to global climate and carbon dioxide change. Acknowledgments This article is derived from work supported by the National Science Foundation, the US Environmental Protection Agency, and the National Aeronautics and Space Administration. I would also like to thank Gus Shaver, Goren Agren, David Ford, Terry Chapin, Chuck Hopkinson, Dave McGuire, Bob McKane, Jerry Melillo, John Hobbie, Dave Fernandes, and Mat Williams for their helpful comments. (n1) Plants using the most common photosynthetic pathway, in which carbon dioxide is incorporated to form the three-carbon compound 3-phosphoglyceric acid. (n2) R. B. McKane, E. B. Rastetter, G. R. Shaver, K. J. Nadelhoffer, A. E. Giblin, and J. A. Laundre, 1995, submitted manuscript. (n3) See footnote 2. GRAPH: Figure 1. Short-term tests are neither sufficient nor necessary to validate long-term model predictions. (a) The model of Farquhar and von Caemmerer (1982) applied to the data of Tissue and Oechel (1987). Although the model predicts changes in net photosynthesis minutes after carbon dioxide levels are increased from 340 to 680 ppmv, this prediction is not sufficient to validate the model for predicting responses weeks in the future. (b) The Farquhar and yon Caemmerer (1982) model applied to the data of Emerson and Arnold (1932). Although the model is unable to predict increased rates of photosynthesis per unit quantum flux (efficiency) as the ratio of light to dark periods decreases in a rapidly flashing light (20millisecond cycle), agreement at this fine time scale is not necessary to validate the model for use at the time scale of seconds to minutes. GRAPH: Figure 2. Simulated 50-year response of the nitrogen cycle in arctic moist tundra to a 5 Celsius increase in temperature and a 10% decrease in soil moisture based on simulations reported in Rastetter et al. (in press). The rates of nitrogen mineralization and nitrogen uptake by vegetation are approximately equal for 20 years after the change in temperature and moisture. However, after 20 years the two processes become decoupled and the loss of nitrogen from the ecosystem increases. There are no short-term data that can validate this long-term prediction. References cited Ball JT, Woodrow LE, Berry JA. 1987. A model predicting stomatal conductance and its contribution to the control of photosynthesis under different environmental conditions. Pages 221-224 in Biggins I, ed. Progress in photosynthesis research. Vol. 5: Proceedings of the VII International Photosynthesis Congress. Dordrecht (the Netherlands): Martinus Nijhoff Publishers. Boden TA, Kaiser DP, Sepanski RJ, Stoss FW, eds. 1994. Trends '93: a compendium of data on global change. Oak Ridge (TN): Carbon Dioxide Information Analysis Center, Environmental Sciences Division, Oak Ridge National Laboratory. Bretherton FP, Bryan K, Woods JD. 1990. Time-dependent greenhouse-gas-induced climate change. Pages 173-194 in Houghton JT, Jenkins GJ, Ephraums J J, eds. Climate change: the IPCC scientific assessment. Cambridge (UK): Cambridge University Press. Chapin FS III, Shaver GR, Giblin AE, Nadelhoffer KJ, Laundre JA. 1995. Responses of arctic tundra to experimental and observed changes in climate. Ecology 76: 694-711. Collatz GJ, Berry JA, Farquhar GD, Pierce J. 1990. The relationship between the Rubisco reaction mechanism and models of photosynthesis. Plant, Cell, and Environment 13: 219-225. Crocker RL, Major J. 1955. Soil development in relation to vegetation and surface age at Glacier Bay, Alaska. Journal of Ecology 43: 427-448. Drake DR, Mueller-Dombois D. 1993. Population development of rain forest trees on a chronosequence of Hawaiian lava flows. Ecology 74: 1012-1019. Emerson R, Arnold W. 1932. A separation of the reaction in photosynthesis by means of intermittent light. Journal of General Physiology 15: 391-420. Farquhar GD, von Caemmerer S. 1982. Modelling of photosynthetic response to environmental conditions. Pages 549-588 in Lange OL, Nobel PS, Osmond CB, Ziegler H, eds. Physiological plant ecology II: water relations and carbon assimilation. Encyclopedia of Plant Physiology, New Series. Vol. 12B. Berling (Germany): Springer-Verlag. Fastie CL. 1995. Causes and ecosystem consequences of multiple pathways of primary succession at Glacier Bay, Alaska. Ecology 76: 1899-1916. Francey RJ, Tans PP, Allison CE, Enting IG, White JWC, Trolier M. 1995. Changes in oceanic and terrestrial carbon uptake since 1982. Nature 373: 326-330. Fritts HC, Lough JM. 1985. An estimate of average annual temperature variations in North America, 1602 to 1961. Climate Change 7: 203-224. Graumlich LJ, Brubaker LB. 1986. Reconstruction of annual temperature (1590-1979) for Longmire, Washington, derived from tree rings. Quaternary Research 25: 223-234. Harley PC, Thomas RB, Reynolds JF, Strain BR. 1992. Modeling photosynthesis of cotton grown in elevated CO2. Plant, Cell, and Environment 15: 271-282. Houghton RA. 1993. Changes in terrestrial carbon over the last 135 years. Pages 139-157 in Heimann M, ed. The global carbon cycle. Berlin (Germany): Springer-Verlag. -----. 1995. Effects of land use change, surface temperature, and CO2 concentration on terrestrial stores of carbon. Pages 333-350 in Woodwell GM, Mackenzie FT, eds. Biotic feedbacks in the global climatic system: will the warming feed the warming? New York: Oxford University Press. Karl TR, Easterling DR, Knight RW, Hughes PY. 1994. U.S. national and regional temperature anomalies. Pages 686-737 in Boden TA, Kaiser DP, Sepanski RJ, Stoss FW, eds. Trends '93: a compendium of data on global change. Oak Ridge (TN): Carbon Dioxide Information Analysis Center, Environmental Sciences Division, Oak Ridge National Laboratory. Kauppi PE, Mielikainen K, Kuusela K. 1992. Biomass and carbon budget of European forests, 1971 to 1990. Science 256: 70-74. Keeling CD, Whorf TP. 1994. Atmospheric CO2 records from sites in the SIO air sampling network. Pages 1627 in Boden TA, Kaiser DP, Sepanski RJ, Stoss FW, eds. Trends '93: a compendium of data on global change. Oak Ridge (TN): Carbon Dioxide Information Analysis Center, Environmental Sciences Division, Oak Ridge National Laboratory. McKane RB, Rastetter EB, Melillo JM, Shaver GR, Hopkinson CS, Fernandes DN. 1995. Effects of global change on carbon storage in tropical forests of South America. Global Biogeochemical Cycles 9: 329-350. McMurtrie RE. 1993. Modelling of canopy carbon and water balance. Pages 220-231 in Hall DO, Scurlock JMO, Bolhar-Nordenkampf HR, Leegood RC, Long SP, eds. Photosynthesis and production in a changing environment: a field and laboratory manual. London (UK): Chapman and Hall. McMurtrie RE, Cumins HN, Kirschbaum MUF, Wang YP. 1992. Modifying existing forest growth models to take account of effects of elevated CO2. Australian Journal of Botany 40: 657-677. Melillo JM, McGuire AD, Kicklighter DW, Moore B III, Vorosmarty CJ, Schloss AL. 1993. Global climate change and terrestrial net primary production. Nature 363: 234-240. Mitchell JFB, Manabe S, Tokioka T, Meleshko V. 1990. Equilibrium climate change. Pages 131-172 in Houghton JT, Jenkins GJ, Ephraums JJ, eds. Climate change: the IPCC scientific assessment. Cambridge (UK): Cambridge University Press. Mooney HA, Drake BG, Luxmoore RJ, Oechel WC, Pitelka LF. 1991. Predicting ecosystem responses to elevated CO2 concentrations. BioScience 41: 96-104. Neftel A, Friedli H, Moor E, Lotscher H, Oescher H, Siegenthaler U, Stauffer B. 1994. Historical CO2 record from the Sieple Station ice core. Pages 11-15 in Boden TA, Kaiser DP, Sepanski RJ, Stoss FW, eds. Trends '93: a compendium of data on global change. Oak Ridge (TN): Carbon Dioxide Information Analysis Center, Environmental Sciences Division, Oak Ridge National Laboratory. Oechel WC, Hastings SJ, Vourlitis G, Jenkins M, Riechers G, Grulke N. 1993. Recent change of arctic tundra ecosystems from a net carbon dioxide sink to a source. Nature 361: 520-523. Oosting HJ. 1942. An ecological analysis of the plant communities of Piedmont, North Carolina. American Midland Naturalist 28: 1-126. Oreskes N, Shrader-Frechette K, Belitz K. 1994. Verification, validation, and confirmation of numerical models in the earth sciences. Science 263: 641-646. Platt JR. 1964. Strong inference. Science 146: 347-353. Popper KR. 1959. The logic of scientific discovery. New York: Basic Books. -----. 1979. Objective knowledge: an evolutionary approach. Revised ed. Oxford (UK): Clarendon Press. Rastetter EB, Houghton RA. 1992. Carbon budget estimates, letters to Science. Science 258: 382. Rastetter EB, Ryan MG, Shaver GR, Melillo JM, Nadelhoffer KJ, Hobbie JE, Aber JD. 1991. A general biogeochemical model describing the responses of the C and N cycles in terrestrial ecosystems to changes in CO2, climate, and N deposition. Tree Physiology 9: 101-126. Rastetter EB, McKane RB, Shaver GR, Nadelhoffer KJ, Giblin AE. In press. Analysis of CO2, temperature, and moisture effects on C storage in Alaskan arctic tundra using a general ecosystem model. In Oechel WC, Holten JI, eds. Global change and Arctic terrestrial ecosystems. New York: Springer-Verlag. Reynolds JF, Leadley PW. 1992. Modeling the response of arctic plants to changes in climate. Pages 413-438 in Chapin FS III, Jefferies RL, Reynolds JF, Shaver GR, Svoboda J, eds. Arctic ecosystems in a changing climate. San Diego (CA): Academic Press. Reynolds JF, Kemp PR, Acock B, Chen JL, Moorhead DL. 1996. Progress, limitations, and challenges in modeling the effects of elevated CO2 on plants and ecosystems. Pages 347-380 in Koch G, Mooney H, eds. Terrestrial ecosystem response to elevated carbon dioxide. San Diego (CA): Academic Press. Ryan MG, Hunt ER Jr., Agren GI, Friend AD, Pulliam WE, Linder SE, McMurtrie RE, Aber JD, Rastetter EB, Raison RJ. In press. Comparing models of ecosystem function for temperate conifer forests. I. Model description and validation. In Melillo JM, Breymeyer A, eds. Global change effects on coniferous forests and grasslands. London (UK): John Wiley and Sons. Rykiel E. 1994. The meaning of models: a response to Oreskes et al. Letters to Science. Science 264: 330331. Schlesinger WH. 1991. Biogeochemistry: an analysis of global change. San Diego (CA): Academic Press. Shaver GR, Chapin FS III. 1991. Production biomass relationships and element cycling in contrasting arctic vegetation types. Ecological Monographs 61: 1-31. Shaver GR, Billings WD, Chapin FS III, Giblin AE, Nadelhoffer K J, Oechel WC, Rastetter EB. 1992. Global change and the carbon balance of arctic ecosystems. BioScience 42: 433-441. Shugart HH, West DC. 1980. Forest succession models. BioScience 30: 308-313. Tissue DT, Oechel WC. 1987. Response of Eriophorum vaginatum to elevated CO2 and temperature in the Alaskan tussock tundra. Ecology 68: 401-410. [VEMAP] Vegetation/ecosystem modeling and analysis project members. 1995. Vegetation/ ecosystem modeling and analysis project (VEMAP): comparing biogeography and biogeochemistry models in a continentscale study of terrestrial ecosystem response to climate change and CO2 doubling. Global Biogeochemical Cycles 9: 407-437. Wahlen M, Allen D, Deck B, Herchenroder A. 1991. Initial measurements of CO2 concentrations (1530 to 1940 AD) in air occluded in the GISP2 ice core from central Greenland. Geophysical Research Letters 18: 1457-1460. Watson RT, Rodhe H, Oescher H, Siegenthaler U. 1990. Greenhouse gases and aerosols. Pages 1-40 in Houghton JT, Jenkins G J, Ephraums J J, eds. Climate change: the IPCC scientific assessment. Cambridge (UK): Cambridge University Press. Weinberg AM. 1972. Science and trans-science. Minerva 10: 209-222. Wilson H, Hansen J. 1994. Global and hemispheric temperature anomalies from instrumental surface air temperatures. Pages 609-614 in Boden TA, Kaiser DP, Sepanski RJ, Stoss FW, eds. Trends '93: a compendium of data on global change. Oak Ridge (TN): Carbon Dioxide Information Analysis Center, Environmental Sciences Division, Oak Ridge National Laboratory. ~~~~~~~~ By Edward B. Rastetter Edward B. Rastetter is an associate scientist at The Ecosystem Center of the Marine Biological Laboratory in Woods Hole, MA 02543. His major research interests are in modeling carbon and nitrogen interactions in terrestrial ecosystems and in the rates of carbon sequestration by terrestrial ecosystems. Copyright 1996 American Institute of Biological Sciences.