A detailed statistical study on selection of optimum IRS LISS pixel configuration for development of water quality models
Abstract. Whenever remote sensing data are used in conjunction with in situ measurements for modelling a physical phenomenon, the major problem is in the selection of digital data which are representative of the ground points under consideration. This problem can be solved in most cases, including the present study on water quality modelling, by placing dependence not on a single pixel number of the estimated point in question. The study reports the analyses of different size pixel arrays for date of 11 April 1993 in the form of CCT obtained from NRSA, Hyderabad. In order to select the best pixel array configuration to represent the sample station point on the multi-band image, nested arrays of pixels for 4 bands with 7 different sizes are sampled at 52 different sampling stations in the Gautami-Godavari river estuary where water quality variables had been measured. These seven array sizes, namely three symmetric arrays of sizes: 5 by 5, 3 by 3 and 1 by 1 and 4 non-symmetrical arrays of 2 by 2 with the centre pixel located at different corners of each 2 by 2 arrays are analysed using Analysis of Variance \[ANOVA] and t -test. The data bank of 52 by 7 by 4 average pixel values was available for ANOVA and t -test. The results of ANOVA highlight the importance of considering all four IRS bands and each band contains different information regarding physical conditions of the study points, sensor characteristics and water and atmospheric responses. The pixel data are also significantly different at 99 per cent confidence level among sample locations. This indicated that all the four bands have to be considered in developing water quality models. The results of t -test proved that the 5 by 5 and 3 by 3 array sizes are the best among the 7 configurations, but no significant difference is indicated between the two array types. The 3 by 3 window is selected for water quality modelling because it is smaller and involves less computational work.