Pending Issues in Applications of Ecophysiological Models 4331 Model Selection and Accuracy

One can imagine a well-structured process whereby a suite of potential models to be used in climate change research are tested for accuracy, considering the target crop(s) and production region. The best model or subset of models would then be used to estimate impacts. In practice, this process is seldom fully executed. The Fourth Assessment Report of the IPCC (Easterling et al. 2007) noted that previous "calls by the Third Assessment Report (TAR) to enhance crop model inter-comparison studies have remained unheeded; in fact, such activity has been performed with much less frequency after the TAR than before."

One obstacle to model intercomparisons is the lack of standardization for inputs and user interfaces. Using any given model requires a substantial effort to learn how to use the model and especially, prepare the input data for local model evaluation and scenario analysis. Thus, groups conducting impact studies often appear to select a model based on previous familiarity or general reputation of the model.

A related obstacle is that assessing model performance is less straightforward than it first seems. Comparisons of model predictions with observed data or outputs from other models require care. First, one should consider whether the evaluation datasets represent a valid sample for the target production situations. Ideally, the evaluation datasets should include conditions that test responses to elevated temperature and [CO2], but since datasets for such conditions are seldom available, most models are evaluated primarily for current conditions. Of course, sufficient calibration data must be available to ensure that one model is not benefited over another by having been pre-calibrated to the region or production system. This can involve subtle bias including weather and soil data and assumptions concerning initial conditions and crop management. Efforts to promote standards and data sharing have met limited success (Hunt et al. 2001; Bostick et al. 2004).

Faced with apparent data limitations for evaluation, users have several options. The foremost is to seek alternative sources of data. Reports from plant breeding trials can provide a wealth of usable data (e.g., Mavromatis et al. 20002; Anothai et al. 2008b; White et al. 2008). Agronomists familiar with a given region often can provide information on expected yields and responses to inputs. Sensitivity analysis, where model inputs or parameters are varied in a systematic fashion, can also provide useful information on model performance even in the absence of comparable field observations (White et al. 2005). Examples of papers describing relatively complete evaluations are White et al. (1995), Asseng et al. (1998), Soler et al. (2007, 2008). Hartkamp et al. (2002) illustrates strategies for evaluating a model for a crop where reliable field data were especially scarce.

The statistical assumptions used to evaluate models are often suspect as well. Comparisons of observed values vs. simulated data are often analyzed with linear regression, which assumes independence of values. Any dataset from multiple locations or years or involving samples over time is likely to violate this assumption. Multiple regression can overcome some of these problems and can be used to test explicitly whether one model provides better predictions than another (White et al. 2007).

4.3.3.2 Plant Processes

Aspects of the physiology represented in models remain problematic as evidenced by the debates over responses to elevated [CO2] (Long et al. 2006; Tubiello et al. 2007; Ziska and Bunce 2007). In simulating photosynthesis, there also is controversy concerning the role of rubisco enzyme activation in responses to heat and elevated [CO2] (Crafts-Brandner and Salvucci 2004). Furthermore, various responses to [CO2] seem independent of effects via photosynthesis and are not considered in models. Thomas and Harvey (1983) found that soybean leaves from plants grown at elevated [CO2] formed an extra layer of mesophyll tissue, which would affect leaf area expansion and gas exchange properties of the leaf. Bunce (2005) found that plants exposed to continuous [CO2] differed from plants only exposed to elevated [CO2] during the daytime, which again is suggestive of non-photosynthetic effects of [CO2]. Crops also vary in how elevated [CO2] affects time to first flower (Reekie et al. 1994), but the mechanisms of such responses are unclear. Plant biology offers great promise as a source of information on the genetic control and physiology of such processes (Hammer et al. 2006). Hoogenboom and White (2003) used information on the Tip locus in common bean to guide improvements in simulation of the temperature and photoperiod responses.

Surprisingly few models used in climate change research quantify the crop energy balance, which requires tracing flows and transformations of energy in the soil, plant, and atmosphere. While criticizable as introducing excessive complexity, an energy balance may make simulations more robust because it provides more realistic plant and soil temperatures and ensures that energy transfers through evaporation and transpiration are realistically constrained. The ecosys model has successfully reproduced performance of a wheat crop grown under free-air CO2 enrichment (FACE) conditions (Grant et al. 2001).

4.3.3.3 Model Design

Ecophysiological models evolve by having features added as understanding improves or limitations are identified. However, the modifications often emphasize expediency over robust software design. Reynolds and Acock (1997) outlined a modular modeling approach that allows components to be interchanged without requiring modifications to other parts of the model software. Modularity would greatly facilitate testing alternative physiological hypotheses related to issues such as temperature effects on photosynthesis. The approach, however, has seen only partial implementation. Individual models have become more modular in structure (e.g., Jones et al. 2003), but interchangeability of modules among models has not been attained.

The question of whether greater complexity improves model accuracy is frequently raised (e.g., Reynolds and Acock 1985; Passioura 1996). A simple model may have limited predictive capability because it does not describe a wide enough range of responses. A complex model may be inaccurate because of incorrect assumptions, programming errors, or propagation of errors from poorly estimated parameters. In the absence of rigorous model comparisons, however, it is difficult to endorse a specific level of complexity.

4.3.3.4 Application Scenarios

Issues such as what are the most accurate estimates for greenhouse gas levels or how to downscale climate change projections are dealt with in other chapters. However, many other aspects of scenario design merit review. Few studies consider how soil variability within a location might affect projections. Impact studies mainly consider crop species in isolation, yet in temperate regions the most dramatic changes in farming in temperate regions may involve changes from single crops to dual cropping and from short season cereal and oilseed crops with a spring habit to winter types.

Analyses of potential impacts of scenarios could be enhanced by greater consideration of associated responses rather than focusing on economic yield. Probably the most pressing topic is how water use might change. Assuming no adaptation in terms of cultivar type or planting date, the simplest expectation is that water use will decline due to the well documented reduction in stomatal conductance with increased [CO2]. However, if crops are selected for greater response of net photosynthesis to [CO2], the water-conserving response of stomata may decline, thus increasing water use. Furthermore, warmer temperatures would increase PET as well as lengthen the growing season, further increasing water use quantified on a seasonal or annual basis. Such interactions are readily simulated, but they involve plant and system responses that are still poorly understood.

Renewable Energy 101

Renewable Energy 101

Renewable energy is energy that is generated from sunlight, rain, tides, geothermal heat and wind. These sources are naturally and constantly replenished, which is why they are deemed as renewable. The usage of renewable energy sources is very important when considering the sustainability of the existing energy usage of the world. While there is currently an abundance of non-renewable energy sources, such as nuclear fuels, these energy sources are depleting. In addition to being a non-renewable supply, the non-renewable energy sources release emissions into the air, which has an adverse effect on the environment.

Get My Free Ebook


Post a comment