Abstract:
Salinization in the dry land is a common problem for agriculture and has greatly impacted land productivity and even brought about the abandonment of cropland. The spatial distribution of salinity and its changing pattern in space and time must therefore be quantified. This can be achieved by mapping the Electric conductivity of the soil using high resolution satellite imagery with remote sensing and GIS technique. Remote sensing and GIS has proven to provide a large-scale monitoring with the aided high accuracy and efficiency in this sector. The aim of this study was to use multitemporary remote sensing to Identify by assessing spectral indices, map, analyze, quantify salinity severity. Koru majorly known for sugarcane production is a ward in Muhoroni constituency in Kisumu County which was used as an example for carrying out this research project. Sentinel 2 Imagery characterized with 13 bands and a relatively high resolution of 10m that was acquired in February 2019 was used with combination of the field soil survey and sampling obtained in the same month within the same period. To avoid redundancy of derived data, principal Component Analysis was applied to the various sentinel bands used. Bands; blue, green, red and NIR were used to extract salinity indices. A series of 8 various documented and proven Salt features were extracted from the Imagery in which a correlation was done between the extracted features and the ground measurements (Electric conductivity values obtained from the field). To obtain the best salinity indices for bare-vegetation land analysis, two different regression techniques were used which were Multiple and geographically weighted. With a correlation coefficient of R2>0.7 and a spatial autocorrelation of p=0.09, three salinity indices proved to be attaining the required threshold: SI1, SI4 &SI14. A further support vector machine with Radial Basis Function was applied to improve the model findings. From the findings, the results showed that salinity can be predicted reliably by the established Salinity Model with a precision of 75%, suggesting that my mapping approach is valid and extensible to other similar environments.