Simulated Breakpoint Precipitation from Global Climate Products based on Machine Learning Regressions Calibrated with Ground Network Information.
Abstract
Runoff and water erosion computer models perform better when driven by precipitation data that has sufficient temporal resolution to represent true peak intensity, and ideally should be driven by breakpoint precipitation data that provides a continuous precipitation hyetograph. A methodology was developed to simulate breakpoint data using the point-scale CLIGEN stochastic weather generator and downscale the required CLIGEN input parameters derived from global climate products. The ERA5 climate reanalysis and GPM satellite-based radar gave sufficient record durations to determine 20-year CLIGEN input parameters. The derived CLIGEN inputs were spatially and temporally downscaled using machine learning regression (MLR) and ground-networks as targets for the MLR during the training phase. The methodology was applied to produce gridded datasets of CLIGEN inputs for different regions of the world, and has the potential to facilitate hydrological and erosion modeling in locations where long-term precipitation records are unavailable.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H25R1234F