利用安徽淮北平原1990年－2007年的粮食产量资料，采用滑动平均法分离出趋势产量，结合因旱减产量得到历年的因旱减产率；为提高信息扩散在旱灾损失风险评估中的精度，运用基于交叉验证的窗宽优化进行改进，并通过计算机仿真验证了其在P-III分布中的适用性。在此基础上，构建旱灾损失风险评估模型，计算了因旱减产率的概率分布，并对该区域旱灾风险分布进行了分析。对比发现，基于交叉验证与信息扩散的旱灾损失风险评估模型构建合理，对解决旱灾风险评估中普遍存在的小样本问题具有一定参考价值。

According to the data of grain output from 1990 to 2007 in Anhui Huaibei Plain, tendency yields were got by wielding sliding average method, and the annual yield reduction rates caused by drought were obtained by output reductions. The bandwidth was optimized by means of cross-validation for the improvement of accuracy of the information diffusion method in the drought loss risk assessment, and the applicability of information diffusion method based on cross-validation for the P-III distribution population in terms of computer simulation was proved. On this basis, the assessment model of drought loss risk was established, and the probability distribution of yield reduction rates caused by drought was caculated with information diffusion. Meanwhile, the risk distribution of regional drought was analyzed. It was found that the model in this paper was built reasonably and appropriately by comparison with others, and provided a certain reference value for solving common small sample size problems in the drought risk assessment.

国家自然科学基金项目 (51409002, 51409001, 71273081); 水利部重大基建前期项目“全国干旱区划及旱灾风险评估研究”;中国气象局成都高原气象开放实验室基金课题 (LPM2011002)