A spatial context-aware model for climate model data fusion

Abstract

Acquiring more accurate climate model data is crucial for conducting precise regional climate studies. Most studies use linear weighting methods or a single machine learning model fusing multiple climate model datasets to reduce uncertainty. However, these methods use the identical model globally and ignore local characteristics of various climate models. This study develops a spatial context-aware fusion (SCAF) model to fuse multi-source climate data by constructing distinct models at different spatial locations, capturing local climate features and enhancing the regional applicability of the climate data. The developed SCAF model is implemented in the fusion of radiation data using 22 CMIP6 climate models in the upper Yellow River. Results show that SCAF can effectively fuse regional radiation data with correlation coefficients higher than 0.95 between the fused data and observed data at any location across space. As such, SCAF can effectively capture regional climate characteristics through spatially local modelling. In addition, the analysis of future radiation trends shows that the rate of radiation decline accelerates with stronger scenario models, with decreases ranging from 0.123 to 0.7771 W/m2 per decade. The model demonstrates significant advantages and has a broad potential to effectively fuse regional climate models.