Geospatio-Temporal Weather Generator Based on Correlation Matrix Factorization.

Weather generators have bee used in the last decades to extend the available historical record length in order to have enough data to test simulation models using sensitivity analyzes, and to explore possible scenarios of climate variability and change. If ones interest is in temporal properties of rainfall, temperatures and its effects in crop production at points of fields, the standard point-specific weather generators works well; but if spatially independent generated data are used to aggregate rainfall, temperatures or model outputs over space for subsequent analyses, spatial correlations of the variables must be taken into account to for the same time scale at which the data are used as inputs to models. New advances in generating synthetic data of precipitation structured in time and space have been recently developed at the University of Florida. The objective was to extend the method to other meteorological variables to increase the use of this methodology in crop and environmental modeling research. The method was developed based on the Cholesky’s factorization of the monthly geospatio-temporal correlation matrices of daily rainfall events and amounts, and maximum and minimum temperatures among the weather stations. Generation of both temperature values was driven by rainfall occurrence. This permitted to take into account the correlation among variables. Daily data for five weather stations located in North Central Florida were obtained from the National Climate Data Center and used in this study. Results were compared to those produced by an existing weather generator. The spatial structure was measured by Moran’s I test after generating rainfall and temperatures using uni- and multi-site weather generators. Main statistics obtained from individual weather stations by using both data generation methodologies, matches those from the observed climatology; however, the spatial structure of the generated data was only preserved by the new geospatio-temporal weather generator.