Statistical downscaling of global climate models (GCMs) allows researchers tostudy local climate change effects decades into the future. A wide range ofstatistical models have been applied to downscaling GCMs but recent advances inmachine learning have not been explored. In this paper, we compare fourfundamental statistical methods, Bias Correction Spatial Disaggregation (BCSD),Ordinary Least Squares, Elastic-Net, and Support Vector Machine, with threemore advanced machine learning methods, Multi-task Sparse Structure Learning(MSSL), BCSD coupled with MSSL, and Convolutional Neural Networks to downscaledaily precipitation in the Northeast United States. Metrics to evaluate of eachmethod's ability to capture daily anomalies, large scale climate shifts, andextremes are analyzed. We find that linear methods, led by BCSD, consistentlyoutperform non-linear approaches. The direct application of state-of-the-artmachine learning methods to statistical downscaling does not provideimprovements over simpler, longstanding approaches.
展开▼
