Study of Different Image Windowing on Turbidity Regression Model using Remotely Sensed Data

Abstract

The Thane creek region, near Mumbai city is being used as dumping site for treated and untreated effluents by government agencies and private industries for the last several decades. This coastal water is very important from environmental point of view since it supports a vast area of mangrove forest besides a wide variety of flora and fauna. Turbidity, an important marine physical pollution parameter, affects the growth of mangroves, causes loss of swamps and poses threat to aquatic life. The work presented discusses the effect of 'variations in sampling time' on Turbidity regression model using Remotely Sensed Data. Marine water samples were collected synchronous to pass of Landsat satellite and Turbidity (NTU) was measured (During the post monsoon season of 1996/97 window of sample collection was ± 1 hour, which was reduced during the post monsoon season of 1997/98 to ± 15 minutes). The digital satellite images were corrected initially for geometric, sun angle and atmospheric errors. From the corrected remotely sensed data, DNs values were extracted. Multiple regression model was developed between water quality parameter, turbidity and extracted Digital Numbers (DNs) from corresponding sampling locations by varying image window sizes (i.e. 1x1, 3×3 and 5×5 pixels). It was deduced that averaging 3×3 window corresponding to water sample collection locations followed by multiple regression with water quality parameter, turbidity, gave best results of regression coefficient.