Development of an effective and potentially scalable weather generator for temperature and growing degree days

Abstract

The main purpose of this study is to develop an easy-to-use weather generator (WG) for the downscaling of gridded data to point measurements at regional scale. The WG is applied to daily averaged temperatures and annual growing degree days (GDD) of wheat. This particular choice of variables is motivated by future investigations on temperature impacts as the most important climate variable for wheat cultivation under irrigation in Iran. The proposed statistical downscaling relates large-scale ERA-40 reanalysis to local daily temperature and annual GDD. Long-term local observations in Iran are used at 16 synoptic stations from 1961 to 2001, which is the common period with ERA-40 data. We perform downscaling using two approaches: the first is a linear regression model that uses the ERA-40 fingerprints (FP) defined by the squared correlation with local variability, and the second employs a linear multiple regression (MR) analysis to relate the large-scale information at the neighboring grid points to the station data. Extending the usual downscaling, we implement a WG providing uncertainty information and realizations of the local temperatures and GDD by adding a Gaussian random noise. ERA-40 reanalysis well represents the local daily temperature as well as the annual GDD variability. For 2-m temperature, the FPs are more localized during the warm compared with the cold season. While MR is slightly superior for daily temperature time series, FP seems to perform best for annual GDD. We further assess the quality of the WGs applying probabilistic verification scores like the continuous ranked probability score (CRPS) and the respective skill score. They clearly demonstrate the superiority of WGs compared with a deterministic downscaling.