Spatial modeling of winter temperature and precipitation in Arizona and New Mexico, USA

Abstract

The development of a statistical modeling technique suitable for producing mean and interannual gridded climate datasets for a topographically varying domain is undertaken. Stepwise regression models at 1 x 1 km resolution are generated to estimate mean winter temperature and precipitation for the Southwest United States for the years 1961-1990. Topographic predictor variables are used to explain spatial variance in the datasets. Kriging and inverse distance weighting interpolation algorithms are utilized to account for model residuals. The final regression models show a high degree of explained variance for temperature (R2 = 0.98, mean bias error [MBE] = -0.15°C, root-mean-squared error [RMSE] = 0.74°C) and a moderate degree of explained variance for precipitation (R2 = 0.63, MBE = -1.4 mm, RMSE = 27.0 mm). Several smaller-scale precipitation regression models are developed for comparison to the domain-wide model, but do not show marked accuracy improvements. Observed values of winter temperature and precipitation from the years 1961-1999 are compared to the 30 yr modeled means, and the differences are interpolated using kriging (temperature) and inverse distance weighting (precipitation). The result is a 39 yr time series of maps and datasets of winter temperature and precipitation at 1 x 1 km resolution for the Southwest United States.