CLASSIFICATION OF RICE, COTTON, AND MAIZE CROPS IN NAGARKURNOOL DISTRICT OF TELANGANA STATE USING MULTI-TEMPORAL SENTINEL-1A SYNTHETIC APERTURE RADAR DATA AND MACHINE LEARNING TECHNIQUES

India's diverse agricultural landscape supports the production of a wide array of crops that are essential to its economy.Staple crops such as cereals, millets, and pulses are cultivated alongside key cash crops and oilseeds.Rice stands as a primary staple, cotton serves as a major cash crop, and maize is gaining importance for both food and fodder, especially in the poultry and industrial sectors.However, mapping these crops accurately during the kharif season using optical remote sensing data is often restricted by persistent cloud cover, highlighting the advantage of all-weather-capable SAR (Synthetic Aperture Radar) data for crop mapping.This study focuses on mapping kharif crops specifically cotton, rice, and maize in the Bijinapally and Peddakothapally mandals of Nagarkurnool district, Telangana, using multitemporal Sentinel-1A C-band SAR data.The SAR data, covering the crop growth period from June to December in the kharif-2023 season, was analyzed to generate mean backscatter profiles in VV (Vertical-Vertical) and VH (Vertical-Horizontal) polarizations for each crop type.Ground truth points were systematically collected from fields of cotton, rice, and maize, enabling precise backscatter profile extraction and facilitating crop differentiation based on temporal backscatter response.In Bijinapally mandal, cotton and rice were distinguishable during August and September.Cotton exhibited backscatter values ranging from -13.01 to -7.44 dB in VV and -16.30 to -13.62 dB in VH polarization, while rice showed values of -12.62 to -9.34 dB in VV and -20.65 to -17.70 dB in VH polarization.Similarly, in Peddakothapally mandal, maize and rice were differentiable from August to mid-September.Maize showed backscatter values between -7.52 and -9.08 dB in VV and -15.04 to -15.21 dB in VH, while rice showed -12.61 to -9.77 dB in VV and -20.49 to -17.84 dB in VH polarization.To classify these crops, two machine learning algorithms-Random Forest (RF) and Support Vector Machine (SVM)-were employed.Among these, RF with VH polarization achieved the highest accuracy.In Bijinapally, RF classification showed deviations of -17.67% and -9.7% for cotton and rice areas, respectively, when compared to official agricultural statistics (7596 and 7311 ha).In Peddakothapally, maize and rice areas showed minimal deviations of 1.1% and 3.7% (6131 and 4373 ha), respectively.The RF classifier using VH polarization achieved an overall accuracy of 94.44% with a kappa coefficient of 0.91, underscoring its effectiveness for crop classification in this region.