Kappa系数的修正及在干旱预测精度及一致性评价中的应用

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

大面积遥感面上干旱预测结果无法应用点上预测精度评价方法对每个像素进行评价,因此该研究引用Kappa系数对一阶自回归模型和季节性求和自回归移动平均模型的关中平原干旱预测结果进行精度和一致性评价。通过将面上VTCI干旱监测结果和预测结果划分为干旱与不旱两种类型,建立转换矩阵,得到了Kappa系数。由于Kappa系数存在两个明显的反论,因此在计算Kappa系数的同时计算了阳性一致率、阴性一致率、PI指数、BI指数、Kappa系数的最大、最小值和PABAK等系列指标,对Kappa系数的反论进行了检验与修正。文章通过对2种预测模型2009年4月上旬、中旬和下旬的干旱预测结果进行评价和对比,得到一阶自回归模型的总体精度优于季节性求和自回归移动平均模型,而季节性求和自回归移动平均模型预测旱情发生的一致性高于一阶自回归模型;得出Kappa系数或修正后的Kappa系数结合阳性一致率可用于干旱预测精度评价的结论。

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.