Adjustment of Kappa coefficient and its application in precision and agreement evaluation of drought forecasting models

Traditional methods for evaluating drought forecasting accuracies cannot be used to evaluate drought forecast results of large areas based on satellite remote sensing data. This paper is mainly on the accuracy evaluation and agreement analysis of the drought forecasting results of the autoregressive models (AR) and the seasonal integrated autoregressive moving average models (SARIMA) using the Kappa coefficient. The drought forecasting results and the drought monitoring results of VTCI were categories into two classes that are drought and normal, and were used to build the transformation matrices to calculate the Kappa coefficients. Two apparent paradoxes have been identified for the Kappa statistic, and some related indices, positive consistency, negative consistency, and prevalence index, bias index, maximum and minimum values of Kappa and PABAK coefficients were calculated and analyzed for verifying the paradoxes and modifying the Kappa coefficients. Drought forecasting results in first, middle and last ten days of April, 2009 in the Guanzhong Plain of China were selected for the evaluation. The results show that the overall accuracy of the AR(1) models are better than the SARIMA models, while the ability of forecasting drought situation of the SARIMA models is better than that of the AR(1) models. The Kappa or adjusted Kappa coefficients combined with the positive consistency can be used to evaluate the drought forecasting accuracy.