Applying artificial neural networks to estimate suspended sediment concentrations along the southern coast of the Caspian Sea using MODIS images

摘要

In this day and age, most environmental researchers use satellite data for monitoring and assessing of water quality indicators since the traditional methods are both time- and money-consuming. One of the most important water quality parameters that can be assessed in coastal waters and river estuaries using remote sensing techniques is suspended sediment concentration (SSC). It regulates primary production and has substantial influence on the migration of pollutants, temperature, and marine life. In this study, Moderate-Resolution Imagine Spectrometry (MODIS) images were used to retrieve the SSC along the southern coast of the Caspian Sea. MODIS of 250 m resolution data were utilized because they have the highest spatial resolution of all the MODIS bands. In situ data were gathered with multiple campaigns with fast motor boats, while the MODIS sensor was passing over the study area. The goal of this article is to apply artificial neural networks (ANN) to retrieve SSC from satellite remote sensing imagery. ANN function as an intelligent structure to model a variety of nonlinear relationships because iteration-based inversion methods need long computation times for common usage. Using a validation data set and a testing data set, the network could be validated. The learning process was more efficient which resulted in a shorter learning time. The validation data set played a vital role as a stopping criterion during the training procedure to overcome the overtraining problem. A robust relationship between MODIS bands 1 and 2 and in situ data was established based on a three-layer ANN with six neurons in the hidden layer. Root mean squared error and R-2 values for this model were 0.853 and 0.969 mg/L, respectively, for all data. Results of this study reveal that the SSC in the Caspian Sea gradually decreases from west to east.