基于地理加权回归模型的水深反演算法研究

摘要

Inversion of shallow seawater depth using satellite multi-spectral data is an important measure of water depth measurement.The existing water depth inversion method is to establish an inversion model of unified mathematical parameters in the study area, without considering the problem of spatial non-stationarity caused by changes in sea floor sediment and water quality.In this study, the Geographically Weighted Regression (GWR) model was used to estimate the regression parameters in space.For the influence of the bandwidth of the GWR model on the inversion accuracy, a Cross Validation (CV) method was used to determine the best bandwidth, taking the sea areas of Woody Island and Ganquan Island in the South China Sea as experimental areas, the feasibility and accuracy of the GWR model were verified based on WordView-2multi-spectral data.As a result of the experiment, the accuracy of the GWR model in the study area of Woody Island was improved by 36.05％compared with the linear regression model, and in the ranges of 0-5m, 5-10 m, 10-15 m, and 15-20 m, the precision was increased by 49.46％, 39.97％, 12.36％ and 49.68％respectively.The precision of GWR model in the study area of Ganquan Island was improved by 8.08％.In the ranges of 0-5m, 5-10 m, 10-15 m, and 15-20 m, compared with the linear regression model, the precision was improved by 12.05％, 16.23％, 4.49％and 12.23％respectively, indicating that the GWR model has a better water depth inversion performance.%利用卫星多光谱数据反演浅海水深是水深测量的一种重要手段.已有水深反演方法是在研究区建立统一的数学参数的反演模型, 未考虑由于海底底质和水质变化导致的空间非平稳性问题.本文使用地理加权回归模型 (Geographically Weighted Regression, GWR) 对回归参数在空间上进行估计, 针对GWR模型的带宽对反演精度的影响, 使用了交叉验证 (Cross Validation, CV) 的方法来确定最佳带宽, 并以南海永兴岛和甘泉岛海域为实验区, 基于WordiVew-2多光谱数据对使用GWR模型的可行性和精度进行了验证.实验结果表明:永兴岛研究区GWR模型精度较线性回归模型提高了36.05％, 在0～5, 5～10, 10～15和15～20m区间, 精度分别提高了49.46％, 39.97％, 12.36％和49.68％;甘泉岛研究区GWR模型精度较线性回归模型提高了8.08％, 在0～5, 5～10, 10～15和15～20m区间, 精度分别提高了12.05％, 16.23％, 4.49％和12.23％, 表明GWR模型具有更好的水深反演效果.