Abstract: In estimating responses of crops to future climate realisations, it is necessary to understand and differentiate between the sources of uncertainty in climate models and how these lead to errors in estimating the past climate and biases in future projections, and how these affect crop model estimates. This paper investigates the complexities in using climate model projections representing different spatial scales within climate change impacts and adaptation studies. This is illustrated by simulating spring barley with three crop models run using site-specific observed, original (50•50 km) and bias corrected downscaled (site-specific) hindcast (1960-1990) weather data from the HadRM3 Regional Climate Model (RCM). Original and bias corrected downscaled weather data were evaluated against the observed data. The comparisons made between the crop models were in the light of lessons learned from this data evaluation. Though the bias correction downscaling method improved the match between observed and hindcast data, this did not always translate into better matching of crop models estimates. At four sites the original HadRM3 data produced near identical mean simulated yield values as from the observed weather data, despite differences in the weather data, giving a situation of ‘right results for the wrong reasons’. This was likely due to compensating errors in the input weather data and non-linearity in crop models processes, making interpretation of results problematic. Overall, bias correction downscaling improved the quality of simulated outputs. Understanding how biases in climate data manifest themselves in crop models gives greater confidence in the utility of the estimates produced using downscaled future climate projections. The results indicate implications on how future projections of climate change impacts are interpreted. Fundamentally, considerable care is required in determining the impact weather data sources have in climate change impact and adaptation studies, whether from individual models or ensembles. No Label

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)