Water TOC and TP Concentration Estimation Using Landsat TM Data with Empirical Algorithms in Chagan Lake, China

Two important parameters used for monitoring inland water quality are the concentrations of total dissolved organic carbon (TOC), and total phosphorus (TP) in surface water. Remote sensing provides a convenient and synoptic method for determining those parameters from upwelling radiances. In this study, imagery data and in situ water sample concurrent with Landsat overpass was acquired. As neural networks has been proven successful in modeling a variety of geophysical transfer functions, here, both regression model and empirical neural network were established to simulate the relationship between these two water parameters and the satellite-received radiances. It was found that neural network model played much better in accuracy than regression model with Landsat TM visible and near infrared bands as inputs. The relative RMSE for the neural network were < 7%. while the RMSE for regression analysis were less than 25% in general. Contour maps of TOC concentration and TP were produced with regression and neural network models, these two water quality parameters distribution characteristics were analyzed with ambient background. Future work still need to be done for the dynamic characteristic of Chagan Lake water quality with Landsat TM data, and the algorithms developed in this study need to be tested with multi-date imagery data and other lakes in Songnen Plain, which have developed in similar geological and environmental backgrounds.