基于GF-1 WFV影像和随机森林算法的总氮反演研究

Retrieval Concentration of TN Using Random Forest Algorithm Based on GF-1 WFV Remote Sensing Data

基于高分一号卫星影像遥感数据和水库水质实测数据,利用随机森林回归算法建立遥感反射率与总氮浓度的定量反演模型,以获取东风水库总氮浓度的时空分布情况,进而对水库总氮污染情况进行分析。结果表明,总氮反演模型精度较高,决定系数R2为0.879,均方根误差为0.169 mg/L,但仍有提升空间。将模型运用于2016-2018年GF-1 WFV影像遥感数据,反演得到东风水库总氮浓度时空分布情况。结果表明,2016年水库总氮浓度呈波动性变化,2017-2018年除个别月份外,水库总氮浓度整体呈下降趋势。从水质标准来看,水库总氮浓度介于Ⅲ~Ⅴ类水质标准,且存在富营养化风险。从不同季节变化情况来看,春冬季总氮浓度较高,夏秋季浓度较低。从空间差异性来看,区域变化规律性不太明显。

Based on the remote sensing data from Gaofen-1 satellite images and measured water quality data of Dongfeng Reservoir,a quantitative retrieval model which links the remote sensing reflectance and TN concentration is established by using the random forest regression,for the purpose to access the temporal and spatial distribution of total nitrogen(TN)concentration and analyze the overall nitrogen pollution of the Reservoir.The results show that the TN quantitative retrieval model has high accuracy,the coefficient of determination R2 is 0.879,and the root means square error(RMSE)is 0.169 mg/L.However,there is still room for improvement.The model is applied to the remote sensing data of GF-1 WFV images from 2016 to 2018 to retrieve the spatial and temporal distribution of the TN concentration in the Reservoir.The results show that the TN concentration fluctuated in 2016.Except for several months,the TN concentration of the Reservoir shows a downward trend from 2017 to 2018.As for water quality indicators,the TN concentration is between GradeⅢ~Ⅴwater qaulity standard,with a risk of eutrophication.From the view of varieties of season,the TN concentration tends to be higher in spring and win⁃ter and lower in summer and autumn.From the perspective of spatial scale,the regional changes are not remarkable.