抚仙湖氨氮遥感估算研究

Study on Remote Sensing Estimation of Ammonia Nitrogen in Fuxian Lake

遥感估算湖泊水质对了解湖泊水质的空间分布特征具有重要意义,本研究利用2013~2017年的Landsat 8 OLI和GF-1 WFV卫星数据,对抚仙湖的氨氮浓度开展遥感估算。使用反向传播神经网络(BPNN)和多元线性回归(MLR)模型开展交叉对比研究。研究结果表明:在样本数较小的情况下BPNN比MLR有绝对的优势,其验证误差RMSE和rRMSE分别为0.0246 mg/L和40.74%。MLR过拟合现象严重,虽建模线性较好,误差较小,但验证误差较大。将风速作为输入参数可有效提高建模的线性,但也会增加过拟合风险。本研究的将对贫营养湖泊水质的遥感估算有一定的参考和借鉴作用。

Understanding the spatial distribution characteristics of lake water quality requires a quantitative remote sensing estimation of lake water quality. Landsat 8 OLI and GF-1 WFV satellite data from 2013 to 2017 are presented in this study. Multiple linear regression (MLR) and back propagation neural network (BPNN) models were employed. The results indicate that BPNN has an absolute advantage over MLR despite the limited sample size, and the verification error RMSE and rRMSE are 0.0246 mg/L and 40.74% respectively. Even though the linear modeling is accurate and the error is minor, the MLR over fitting phenomenon is serious because the verification error is large. Wind velocity as an input parameter can enhance the linearity of the model, but it also increases overfitting. The results of this study can be used as a reference for the remote sensing estimation of the water quality of oligotrophic lakes.