松花江哈尔滨段悬浮物含量遥感反演与克里格插值预测精度对比

Comparison of Prediction Accuracies of Contents of Suspended Solids by Remote Sensing Inversion and Kriging Spatial Interpolation at Harbin Section of Songhua River

以松花江干流哈尔滨段水体中悬浮物为研究对象,2012年7月14~15日,设置了11个采样断面,共33个采样点,测量得到总悬浮物含量和江面高光谱反射率数据;选用时间上较为接近的TM影像,分别用遥感反演模型（PSO-LSSVM模型）和克里格空间插值方法,对总悬浮物含量进行预测,并对使用两种方法的结果进行比较。结果表明,使用两种方法的结果的预测精度较为接近,遥感反演模型预测值的平均绝对百分误差为11.21%,均方根误差为10.05 mg/L;克里格空间插值预测结果的平均绝对百分误差为11.32%,均方根误差为10.93 mg/L;遥感反演模型预测值能较好的反映出松花江干流哈尔滨段总悬浮物含量的总体分布特征和变异特征;克里格空间插值对空间自相关小范围内具有较好的预测能力,超出自相关范围不能预测。

Harbin section of Songhua River was taken as the study region, 33 sampling points were arranged at11 monitoring section. The contents of suspended solids and the hyperspectral reflectances of river surface were measured on July 14-15, 2012, used and sampling time was relatively close to the TM images. The contents of suspended solids were predicted by both the methods of remote sensing inversion model and kriging interpolation method, meanwhile, a comparison of prediction accuracies between these two methods was carried out. The results showed that prediction accuracies of the contents of suspended solids by two methods were similar. Mean absolute percent error and root mean square error of the results by the remote sensing inversion model were 11.21% and 10.05 mg/L, respectively. Mean absolute percent error and root mean square error of the results by the Kriging interpolation method were 11.32% and 10.93mg/L, respectively. The overall distribution characteristics and variation characteristics of different concentration for contents of suspended solids were represented fairly well by prediction values of remote sensing inversion model. Kriging interpolation method had better predictive ability at the range of spatial autocorrelation, and it could＇t forecast beyond the range.