Keywords: Neural Network, Chlorophyll-a, Suspended Solid, Back propagation, CASI.
Abstract
Two important parameters for monitoring water quality of inland water bodies are the concentrations of Chlorophyll a (Chl.a) and Suspended Sediment (SS) in the surface water. Due to the optically complex nature of inland water with high turbidity and dissolved organic matter this task becomes quite difficult in comparison to Case-I water. Neural Network have proven its ability successfully in modeling a variety of geophysical transfer function. A Back-Propagation Neural Network (BPNN) model with one hidden layer is employed for modeling the transfer function between the chlorophyll a and sediment concentrations and in situ upward reflectance radiance taken with a spectro-radiometer at the Lake Kasumi Gaura of Japan. The inputs to the network model are the in-situ upward radiance taken all over the Lake Kasumi Gaura and the outputs are the concentrations of Chl.a and SS. The trained and validated model is used to get the spatial distribution of Chl.a and sediment concentrations at the lake Kasumi Gaura using CASI(Compact Airborne Spectrographic Imager) imagery. The Neural Network model was able to model the transfer function to a much higher accuracy than multiple regression analysis. In case of Chl.a estimation, the RMS error for the neural network was 8.29mg/l, whereas the same for regression analysis were 26.498.29mg/l.. However, in case of estimation of SS by regression RMS errors was 3.94mg/l, whereas by Neural Network RMS error came out to be 2.90mg/l..
1. Introduction
Lake Kasumi Gaura is the second largest lake in Japan located about 60 km northeast of Tokyo.. It is a monomictic lake with average depth of about 4 m.. It faces acute problems of eutrophication and heavy sedimentation every year. The lake houses several important cities/towns around its bank which effect or are being effected by the water quality of the same. Keeping in view this very important fact and other, it becomes necessary to parameterize and estimate the water quality parameters as accurately as possible for effective and correct investigation of the present situation and possible solutions. The main factors effecting the water quality of lakes include the concentrations of Chlorophyll-a(Chl.a) and Suspended Solid(SS). This study tries to find a way to effectively estimate these two water quality parameters in the Lake Kasumi Gaura. Present day remote sensing makes it possible to monitor the Chl.a and SS concentration in wide spatial scale. Oki(1997) have developed spatial concentration maps for Chl.a and SS in Lake Kasumi Gaura using Landsat-TM and regression analysis. The authors believed that, conventional techniques like regression can not model the transfer function effectively in complex waters. The study is a result to eliminate this very drawback.
This study employs Neural Networks (NN) as a tool for effectively model the transfer function of Chlorophyll and Suspended Solid in the water of Lake Kasumi Gaura. The model is then is used in CASI image taken at the lake to get the picture of Chl.a and SS distribution.
2. In Situ Data
Water Quality samples for concentrations of Chl.a and SS were collected in the waters of Lake Kasumi Gaura by Oki(1997) for a total of 29 locations spread over the lake. Out of these 29 data set, data for 20 locations were collected on 10th Sept., 1993, 6 locations were collected on 22nd of April, 1994 and the data at remaining 3 locations were collected on 5th of September, 1996, coinciding with the CASI flight over the lake. Measurements of water leaving radiance reflectance at the water surface were also done at all the 29 locations with a spectro-radiometer.
