Seasonal Predictability of Drought and the Importance of Land-Atmosphere Interactions

Abstract

Hydrologic extremes in the form of flood and drought have large impacts on society. The ability to predict such extreme events at seasonal timescale allows for preparations that can reduce the risk of these events. However, seasonal prediction skill of global climate models varies seasonally and spatially, which severely limits their practical use. In this thesis a framework for assessing and attributing the seasonal predictability through a probabilistic predictability metric based on model skill across temporal and spatial scales; i.e. for the canonical events was developed and demonstrated. The attribution of predictability specific to land-atmosphere interactions and drought is also developed through a new classification of land-atmosphere interactions that includes the Coupling Drought Index (CDI). The CDI was used to understand the current predictability in NCEPs Climate Forecast System version 2 (CFSv2) and the new classification of coupling is used to develop statistical models to isolate attributes of predictability relevant to land-atmosphere interactions and drought. The results show clear seasonal and spatial patterns of predictability that vary with each forecast variable and provide a better understanding of when and where to have confidence in model predictions. The new classification of coupling indicates strong persistence and the CDI shows good agreement with the temporal and spatial variability of drought and highlights the role of coupling in drought recovery. The CDI in the CFSv2 forecasts indicates climatological bias toward the wet coupling regime that precludes the forecast model from consistently predicting and maintaining drought over the continental US. The attribution of the CFSv2 forecasts skill in the summer indicates that the local persistence of initial conditions provides some predictability over the hindcast period and for specific drought events, however the skill is greatly enhanced by the inclusion of spatial interactions. Furthermore, the statistical model based on correcting coupling bias in CFSv2 provides an unbiased prediction and maintained a similar level of skill and provided better precipitation predictions during the 1988 drought. This argues that the wet bias in the coupling limits the precipitation predictability during drought events. The synthesis and extension of the results is also discussed.