Remote sensing-based estimation for Gaussian distribution parameters of vertical structure of algal biomass in Lake Chaohu

Abstract

藻类垂向分布异质性导致了遥感反演的湖泊表层叶绿素a浓度结果与单元水柱内藻类生物量间不存在一一对应的关系，因此有效确定藻类垂向分布结构是遥感反演湖泊藻类生物量的基础.受自身因素和外环境条件的影响，藻类垂向分布结构呈现出多种类型，其中高斯类型应用最广.本文基于3200组HydroLight模拟的高斯垂向数据构建BP神经网络，实现用MODIS数据相对应的3个波段的遥感反射比R_rs（469）、R_rs（555）、R_rs（645）和表层叶绿素a浓度共同估算高斯垂向分布结构参数h和σ.经巢湖地面实测数据验证显示，h和σ的估算值与实测值的相关系数分别为0.97和0.95，对应的相对误差分别为13.20%和12.36%，两者相对误差同时小于30%的占总数据量的87.5%，表明该BP神经网络估算巢湖藻类高斯垂向分布结构的有效性和准确性，为基于卫星遥感数据获取湖泊藻类生物量提供了重要的理论基础.;The relationship between water surface reflectance and total depth integrated algae biomass can be very complex as different kinds of algal vertical distributions can occur. For this reason, effectively identifying the algae vertical profiles is fundamental to estimate algal biomass. Gaussian profiles are the most typical algae vertical profiles which occur in most environmental conditions (including external and internal system). In this research, a back propagation (BP) neural network was established to estimate Gaussian distribution parameters of the vertical structure h and σ by wave bands R_rs(469), R_rs(555), R_rs(645) and chlorophyll-a concentration band C_Chl.a(0). The BP neural network was trained by using 3000 simulated datasets (radiative transfer simulation based on in-situ measured data by HydroLight), and verified by another 200 groups of simulated data and measured data. The correlation coefficient between estimated and measured h and σ were 0.97 and 0.95, while the relative errors were 13.20% and 12.36%, respectively. The relative error of h and σ was mostly less than 30%. This indicated that it is a good effectiveness of BP neural networks to estimate the vertical distribution parameters and able to explore the three dimensional algal distribution in Lake Chaohu, thereby providing a significant theoretical basis for remote sensing estimation of algal biomass.