A NEW METHODOLOGY FOR STOCHASTIC SIMULATION OF DAILY CLIMATIC DATA PRESERVING THE INTERANNUAL VARIABILITY

In this work we propose a new methodology to reproduce, by means of simulations, the interannual variability of climatic variables which included only the minimum air temperature. To evaluate the performance of the proposed method, it was maked a comparison with other two weather generators (i.e., PGECLIMA_R and LARS-WG). Moreover, it was utilized the historical series of thirty years of five meteorological stations of the state of Parana - Brazil to generate ten sets of thirty years for each model, which were confronted with the respective historical series. The performance of the proposed model as well as weather generators was evaluated by applying tests of central tendency, variability and distribution. Furthermore, was utilized the statistical measures RMSE, MBE and Willmott agreement index (d). In the stations investigated, the proposed methodology reduced the total error and eliminated the negative bias of interannual variability. In only four (of 600) generated sequences the interannual variability differs significantly from the observed one. The series generated by PGECLIMA_R and LARS-WG presented rejection rate of 99% in the variability test. In this case, the bias was ten times greater and the RMSE was twice times greater than the proposed methodology. The d index was always greater than 0.98 for the five locations in the proposed methodology and around 0.83 in other models. Based on these results, the new methodology provides a relevant contribution concerning the interannual variability of climatic variables.