Wallach, D., Thorburn, P., Asseng, S., Challinor, A. J., Ewert, F., Jones, J. W., et al. (2016). Estimating model prediction error: Should you treat predictions as fixed or random. Env. Model. Softw., 84, 529–539.
Abstract: Crop models are important tools for impact assessment of climate change, as well as for exploring management options under current climate. It is essential to evaluate the uncertainty associated with predictions of these models. We compare two criteria of prediction error; MSEPfixed, which evaluates mean squared error of prediction for a model with fixed structure, parameters and inputs, and MSEPuncertain(X), which evaluates mean squared error averaged over the distributions of model structure, inputs and parameters. Comparison of model outputs with data can be used to estimate the former. The latter has a squared bias term, which can be estimated using hindcasts, and a model variance term, which can be estimated from a simulation experiment. The separate contributions to MSEPuncertain(X) can be estimated using a random effects ANOVA. It is argued that MSEPuncertain(X) is the more informative uncertainty criterion, because it is specific to each prediction situation.
|
Maiorano, A., Martre, P., Asseng, S., Ewert, F., Müller, C., Rötter, R. P., et al. (2016). Crop model improvement reduces the uncertainty of the response to temperature of multi-model ensembles. Field Crops Research, 202, 5–20.
Abstract: To improve climate change impact estimates and to quantify their uncertainty, multi-model ensembles (MMEs) have been suggested. Model improvements can improve the accuracy of simulations and reduce the uncertainty of climate change impact assessments. Furthermore, they can reduce the number of models needed in a MME. Herein, 15 wheat growth models of a larger MME were improved through re-parameterization and/or incorporating or modifying heat stress effects on phenology, leaf growth and senescence, biomass growth, and grain number and size using detailed field experimental data from the USDA Hot Serial Cereal experiment (calibration data set). Simulation results from before and after model improvement were then evaluated with independent field experiments from a CIMMYT world-wide field trial network (evaluation data set). Model improvements decreased the variation (10th to 90th model ensemble percentile range) of grain yields simulated by the MME on average by 39% in the calibration data set and by 26% in the independent evaluation data set for crops grown in mean seasonal temperatures >24 °C. MME mean squared error in simulating grain yield decreased by 37%. A reduction in MME uncertainty range by 27% increased MME prediction skills by 47%. Results suggest that the mean level of variation observed in field experiments and used as a benchmark can be reached with half the number of models in the MME. Improving crop models is therefore important to increase the certainty of model-based impact assessments and allow more practical, i.e. smaller MMEs to be used effectively.
|
Zhao, G., Hoffmann, H., Yeluripati, J., Xenia, S., Nendel, C., Coucheney, E., et al. (2016). Evaluating the precision of eight spatial sampling schemes in estimating regional means of simulated yield for two crops. Env. Model. Softw., 80, 100–112.
Abstract: We compared the precision of simple random sampling (SimRS) and seven types of stratified random sampling (StrRS) schemes in estimating regional mean of water-limited yields for two crops (winter wheat and silage maize) that were simulated by fourteen crop models. We found that the precision gains of StrRS varied considerably across stratification methods and crop models. Precision gains for compact geographical stratification were positive, stable and consistent across crop models. Stratification with soil water holding capacity had very high precision gains for twelve models, but resulted in negative gains for two models. Increasing the sample size monotonously decreased the sampling errors for all the sampling schemes. We conclude that compact geographical stratification can modestly but consistently improve the precision in estimating regional mean yields. Using the most influential environmental variable for stratification can notably improve the sampling precision, especially when the sensitivity behavior of a crop model is known.
|
Hoffmann, H., Zhao, G., Asseng, S., Bindi, M., Biernath, C., Constantin, J., et al. (2016). Impact of spatial soil and climate input data aggregation on regional yield simulations. PLoS One, 11(4), e0151782.
Abstract: We show the error in water-limited yields simulated by crop models which is associated with spatially aggregated soil and climate input data. Crop simulations at large scales (regional, national, continental) frequently use input data of low resolution. Therefore, climate and soil data are often generated via averaging and sampling by area majority. This may bias simulated yields at large scales, varying largely across models. Thus, we evaluated the error associated with spatially aggregated soil and climate data for 14 crop models. Yields of winter wheat and silage maize were simulated under water-limited production conditions. We calculated this error from crop yields simulated at spatial resolutions from 1 to 100 km for the state of North Rhine-Westphalia, Germany. Most models showed yields biased by <15% when aggregating only soil data. The relative mean absolute error (rMAE) of most models using aggregated soil data was in the range or larger than the inter-annual or inter-model variability in yields. This error increased further when both climate and soil data were aggregated. Distinct error patterns indicate that the rMAE may be estimated from few soil variables. Illustrating the range of these aggregation effects across models, this study is a first step towards an ex-ante assessment of aggregation errors in large-scale simulations.
|
Wallach, D., Thorburn, P., Asseng, S., Challinor, A. J., Ewert, F., Jones, J. W., et al. (2016). Overview paper on comprehensive framework for assessment of error and uncertainty in crop model predictions (Vol. 8).
Abstract: Crop models are important tools for impact assessment of climate change, as well as for exploring management options under current climate. It is essential to evaluate the uncertainty associated with predictions of these models. Several ways of quantifying prediction uncertainty have been explored in the literature, but there have been no studies of how the different approaches are related to one another, and how they are related to some overall measure of prediction uncertainty. Here we show that all the different approaches can be related to two different viewpoints about the model; either the model is treated as a fixed predictor with some average error, or the model can be treated as a random variable with uncertainty in one or more of model structure, model inputs and model parameters. We discuss the differences, and show how mean squared error of prediction can be estimated in both cases. The results can be used to put uncertainty estimates into a more general framework and to relate different uncertainty estimates to one another and to overall prediction uncertainty. This should lead to a better understanding of crop model prediction uncertainty and the underlying causes of that uncertainty. This study was published as (Wallach et al. 2016)
|
