Neural network technology and semi-analytical approach combined model for remote sensing chlorophyll-a concentration

Abstract

The accurate assessment of chlorophyll-a concentration in turbid coastal waters by means of remote sensing is quite challenging, due to the optical complexity of these waters. In this study, a semi-analytical approach is used to analyze the mathematical relationship between chlorophyll-a concentration and remote sensing reflectance, then neural network technology is proposed to simulate the mathematical relationship in the Yellow Sea and East China Sea (YS & ECS). Through evaluation by field measurements, it is shown that our model produces 31.4% uncertainty in quantifying chlorophyll-a concentration from the YS & ECS. Moreover, the performance of our model was compared with four existing models, and the results indicate that the use of our model for quantifying chla in the YS & ECS can decrease uncertainty by >58% in comparison to the four existing models. The atmospheric influences on MODIS data are removed using a near-infrared-shortwave infrared model, then the chlorophyll-a data is quantified from the atmospheric corrected results.