Using AI Techniques to Define the Roles of Temperature and Precipitation Anomalies in the Recent California Drought

The current California drought, which is part of the abnormal to extreme drought conditions affecting much of southwest USA, has lasted for 4 years (2011/12 – 2014/15). It has intensified steadily to what at present is likely the worst Californian drought since reliable instrumental records began in 1895. The uniqueness of this drought is demonstrated by assessing the Oct. – Mar. wet seasons for instances <25th percentile of precipitation and >75th percentile in average temperature. The latter has resulted in both higher evaporation rates and less snowpack in winter, providing positive feedback. Although previous multiyear droughts are present in a century long record in California, the present drought is the first one to combine a several consecutive years of abnormally warm climate with the precipitation deficit. Of the 8 seasons since 1895 that met these percentile conditions, only the present drought satisfied these criteria for more than one season. Predictions of California precipitation and temperature anomalies were made using linear regression (LR), regression trees and support vector regression (SVR) with several linear and non-linear kernels, applied to a range of climate drivers and local sea surface temperatures (SSTs). Cross-validated correlations were low (LR) to moderate (SVR) for precipitation, but were high (>0.7) for both temperature LR and SVR, with SVR marginally exceeding LR. The leading predictors were global warming and local SSTs near the California coast. Finally, the cool seasons were classified as dry/not-dry and hot/not-hot using logistic regressions and k-means classification clustering. Again, it was found that predictability was low for dry/not-dry classes but was high (>70% correct) for hot/not-hot classes. This research suggests that the climate system has warmed sufficiently so that drought can no longer be assessed solely by the lack of precipitation, but must consider the combination of low precipitation and abnormal warmth.