基于非线性特征的干旱影响评估研究

Drought Impact Assessment Based on Nonlinear Characteristics of Drought

摘要:

客观地认识干旱的非线性特征是干旱影响评估的关键，对制定抗旱减灾策略具有重要指导意义。以陕西省关中平原为研究区域，以核函数方法为非线性算法，基于核主成分分析方法(KPCA)，将遥感反演的条件植被温度指数（VTCI）映射到高维特征空间下对其进行特征提取，并结合Copula函数构建主成分间的联合分布模型，确定2008—2013年冬小麦主要生育期的综合VTCI；构建综合VTCI与冬小麦单产间的线性回归模型，评估干旱对冬小麦产量的影响。结果表明，相比于传统的主成分分析方法(PCA)，KPCA能有效地提取干旱的非线性特征，且降维效果更好。与PCA—Copula方法构建的回归模型相比，应用KPCA—Copula方法所建综合VTCI与单产间的回归模型的拟合度明显提高，决定系数达到0.608(p<0.001)，对应模型的估测单产与实测单产之间的均方根误差（RMSE）为298.1kg/hm2，相比于PCA—Copula的结果降低了60.1kg/hm2，且KPCA—Copula获取的综合VTCI更符合关中平原实际的干旱特征。这表明KPCA—Copula方法能够较好地体现干旱的非线性特征，更加适用于干旱影响评估研究。

Abstract:

Drought is a typical complex system, and nonlinear characteristics of drought are the concentrated reflection of its complexity. Therefore, objectively understanding of complex nonlinear characteristics of drought is the key approach of assessing the effects of drought, which can provide guideline for making drought mitigation strategies. Guanzhong Plain was chosen as study area, and the kernel method was applied as a nonlinear algorithm. Based on the kernel principal component analysis (KPCA), vegetation temperature condition index (VTCI) retrieved from MODIS was projected into a highdimensional feature space for feature extraction, and then the joint distribution model of principal components with Copula function was built. Comprehensive values of VTCIs at main growth stages from 2008 to 2013 were determined by using the joint distribution model (the KPCA—Copula method). Linear regression models between the comprehensive VTCIs and wheat yields were established to assess the effect of drought on wheat yields. The results showed that the KPCA could effectively extract the nonlinear characteristics of drought, and it had better performance in dimension reduction compared with the principal component analysis (PCA). Compared with the PCA—Copula method, the determination coefficient of regression model between wheat yields and comprehensive VTCIs with KPCA—Copula method reached 0.608 (p<0.001), which indicated that the fitting degree of the model was improved, and the root mean square error (RMSE) between estimated yields and measured ones was 298.1 kg/hm2, which was about 60.1kg/hm2 lower than the RMSE by using PCA—Copula method. The comprehensive VTCIs with KPCA—Copula method were more in line with actual drought characteristics of Guanzhong Plain. These results indicated that the KPCA—Copula method could well reflect nonlinear characteristics of drought, and it had good applicability in drought impact assessment.