MEAN AND VARIANCE CHANGE IN CLIMATE SCENARIOS - METHODS, AGRICULTURALAPPLICATIONS, AND MEASURES OF UNCERTAINTY

Our central goal is to determine the importance of including both mean and variability changes in climate change scenarios in an agricultura l context. By adapting and applying a stochastic weather generator, we first tested the sensitivity of the CERES-Wheat model to combinations of mean and variability changes of temperature and precipitation for two locations in Kansas. With a 2 degrees C increase in temperature wi th daily (and interannual) variance doubled, yields were further reduc ed compared to the mean only change. In contrast, the negative effects of the mean temperature increase were greatly ameliorated by variance decreased by one-half. Changes for precipitation are more complex, si nce change in variability naturally attends change in mean, and constr aining the stochastic generator to mean change only is highly artifici al. The crop model is sensitive to precipitation variance increases wi th increased mean and variance decreases with decreased mean. With inc reased mean precipitation and a further increase in variability Topeka (where wheat cropping is not very moisture limited) experiences decre ase in yield after an initial increase from the 'mean change only' cas e. At Goodland Kansas, a moisture-limited site where summer fallowing is practiced, yields are decreased with decreased precipitation, but a re further decreased when variability is further reduced. The range of mean and variability changes to which the crop model is sensitive are within the range of changes found in regional climate modeling (RegCM ) experiments for a CO2 doubling (compared to a control run experiment ). We then formed two types of climate change scenarios based on the c hanges in climate found in the control and doubled CO2 experiments ove r the conterminous U. S. of RegCM: (1) one using only mean monthly cha nges in temperature, precipitation, and solar radiation; and (2) anoth er that included these mean changes plus changes in daily (and interan nual) variability. The scenarios were then applied to the CERES-Wheat model at four locations (Goodland, Topeka, Des Moines, Spokane) in the United States. Contrasting model responses to the two scenarios were found at three of the four sites. At Goodland and Des Moines mean clim ate change increased mean yields and decreased yield variability, but the mean plus variance climate change reduced yields to levels closer to their base (unchanged) condition. At Spokane mean climate change in creased yields, which were somewhat further increased with climate var iability change. Three key aspects that contribute to crop response ar e identified: the marginality of the current climate for crop growth, the relative size of the mean and variance changes, and timing of thes e changes. Indices for quantifying uncertainty in the impact assessmen t were developed based on the nature of the climate scenario formed, a nd the magnitude of difference between model and observed values of re levant climate variables.