Spatial downscaling of climate variables using three statistical methods in Central Iran

Abstract

Spatial downscaling methods are widely used for the production of bioclimatic variables (e.g. temperature and precipitation) in studies related to species ecological niche and drainage basin management and planning. This study applied three different statistical methods, i.e. the moving window regression (MWR), nonparametric multiplicative regression (NPMR), and generalized linear model (GLM), to downscale the annual mean temperature (Bio1) and annual precipitation (Bio12) in central Iran from coarse scale (1 km × 1 km) to fine scale (250 m ×250 m). Elevation, aspect, distance from sea and normalized difference vegetation index (NDVI) were used as covariates to create downscaled bioclimatic variables. Model assessment was performed by comparing model outcomes with observational data from weather stations. Coefficients of determination (R2), bias, and root-mean-square error (RMSE) were used to evaluate models and covariates. The elevation could effectively justify the changes in bioclimatic factors related to temperature and precipitation. All three models could downscale the mean annual temperature data with similar R2, RMSE, and bias values. The MWR had the best performance and highest accuracy in downscaling annual precipitation (R2=0.70; RMSE=123.44). In general, the two nonparametric models, i.e. MWR and NPMR, can be reliably used for the downscaling of bioclimatic variables which have wide applications in species distribution modeling.