SOIL FERTILITY EVALUATION FOR FERTILISER RECOMMENDATION USING HYPERION DATA |
Paper ID : 1091-SMPR-FULL |
Authors: |
RANENDU GHOSH *1, Nityanand Padmanabhan2 1faculty Block 4, DAIICT, Near Indrora Circle, Gandhinagar, Gujarat 382 007 2http://iet.ahduni.edu.in/people-details/faculty-list/dr--n--padmanabhan |
Abstract: |
Soil fertility status is the most critical input for fertilizer recommendation under different cropping system. Traditionally soil fertility is established through soil testing which is very cumbersome process. The study was focused at exploring the utility of satellite based hyperspectral data for establishing fertility status of the soil. Hyperion data over Udaipur city and adjoining area was selected for the study. Hyperion data in the form of radiance image acquired during January 2004 was available in the spectral region between 426 – 2395 nm and spectral resolution of 10 nm. Preprocessing, atmospheric correction and spectral analysis of the satellite data was carried out in ENVI software. Spectral Feature Fitting (SFF) was done with four dominant clay mineral signature e.g. Quartz, Kaolinite, Illite and Montmorrilonite. SFF image was converted to point vector after appropriate thresholding. Similarly Pixel Purity Index (PPI) analysis was done on the radiance image to identify the most likely end member location. Both PPI and SFF information was used to select the ground truth sites for soil sampling. Soil samples were collected from forty-one GT sites for soil chemical analysis. Soil chemical analysis was carried to establish the fertility status with respect to seven soil nutrients e.g. organic carbon (OC), available nitrogen (N), available phosphorus (P), available potassium (K), exchangeable calcium (Ca), exchangeable magnesium (Mg) and available sulphur (S). Results of the chemical analysis show that the soil of the area is rich in OC and K while rest five nutrients are present in medium proportion. Statistical analysis was performed to optimize the number of spectral bands and spectral parameter to be used for estimating soil nutrient content for the unknown pixels of the image. Spectral bands were optimized using correlogram and spectral parameters were optimized using multiple regression analysis. Multiple regression coefficients corresponding to different nutrients were used for generating soil nutrient image of the study area. Coefficient of prediction in most cases was found to be more than 90% and RPD was found to be better than 2 which indicate that predictability of the regression model would be satisfactory and reliable for the unknown pixels. |
Keywords: |
Hyperion, soil fertility, spectral feature fitting, pixel purity index |
Status : Paper Accepted (Oral Presentation) |