Presentation description

Cool-season precipitation plays a significant role in the complex terrain of the Western contiguous United States (CONUS), impacting water resources, recreation, and public safety. However, the coarse grid spacing of current operational modeling systems cannot sufficiently resolve fine-scaled precipitation patterns in these regions. Convolutional neural networks (CNNs) can be utilized to downscale winter precipitation in mountainous terrain based on orographic precipitation gradients (OPGs), allowing for more fine-scaled prediction. Such a CNN has been tested in the Northern Rockies region of the CONUS, accounting for 34% of OPG variance with a mean absolute error (MAE) of about 2.9 mm km-1. However, facets with fewer OPG observations had higher MAE. This research aims to expand the application of the CNN tested in the Northern Rockies to predict OPGs for the entire Western CONUS and to improve model performance via hyperparameter tuning and custom loss functions. Hyperparameter tuning was implemented on four hyperparameters in the CNN to automate the process of optimizing hyperparameters and improve the CNN. To reduce errors in facets with fewer OPG observations, two custom loss functions were tested: one based on spatial correlation and the other on the absolute difference in OPGs. By penalizing deviations from a relationship based on facets with at least a 75% length of record, facets with fewer observations could benefit from neighboring facets with more observations. However, the original model predicted OPGs with more (less) correlation (absolute difference) than was observed. Thus, the model with the custom loss function added random heterogeneity that was effective at decreasing (increasing) correlation (absolute difference), but ineffective at improving OPG prediction. Ultimately, the CNN for the entire Western CONUS only accounted for 15% of OPG variance with a MAE of 4.69 mm km-1, indicating that it struggled to predict OPGs in a region as large as the Western CONUS.