A framework to compare the performance of spatial disaggregation methods of statistical climate downscaling in South America

Abstract

Study region Chile and South America. Study focus Climate change is simulated at a coarse scale with General Circulation Models. Impact and regional studies require downscaling to increase spatial resolution. Several downscaling methods exist, but most studies do not analyze or compare spatial disaggregation process. To overcome this a framework is proposed, which is tested comparing eight methods (Bilinear (BIL), Bicubic (BIC), Inverse Distance Interpolation (DIS), Largest Area Fraction (LAF), Nearest Neighbor (NN), First-Order Conservative (CON1), Second-Order Conservative (CON2), and Ordinary Kriging (KRIG)) using daily temperature and precipitation data at different spatial scales in Chile (using CR2MET V2.5 dataset) and South America (using ERA5-LAND dataset). The concept of maximum scaling factor is introduced: ratio between the coarse grid resolution and the high grid resolution after disaggregation. The maximum scaling factor allows recommending the resolution of disaggregation to avoid variability inflation. The framework consists in aggregating the data to different resolutions and then disaggregating them back to their original grids. Finally, performance is measured with indices to determine which spatial disaggregation method most accurately recovers the original climatic gridded information. New hydrological insights for the region When spatial disaggregating gridded data one should avoid increasing the spatial resolution more than ∼22 times for daily precipitation and ∼11 for temperature in Chile, and ∼14 for precipitation and ∼8 for temperature in South America to avoid variability inflation.