Key points are not available for this paper at this time.
Monitoring agricultural land over vast geographical areas presents challenges due to the absence of accurate, comprehensive and precise data, which has become a complex process that is difficult to do in terms of both timespans and consistency. Hence, this study presents an improved approach for the identification of agricultural land by utilizing the capabilities of Sentinel-1 and Sentinel-2 satellites with a variety of vegetation and non-vegetation indices and machine learning algorithms. The Multispectral Correlation Mapper (MCM) and Random Forest (RF) algorithms are adopted to train different agricultural lands, crop types and sowing and cultivation seasons. The 45-bands mega-file data cube (MFDC) fusion for each season incorporates essential indices and features derived from the Sentinel-1 and Sentinel-2 datasets for both seasons, i.e. Rabi (winter-spring season) and Kharif (summer-autumn season). The proposed method demonstrated resilience when applied to satellite datasets while effectively reducing the impact of non-agricultural elements such as shrubs, grass, bare soil and orchards. The results demonstrate a notable ability to differentiate between the Rabi and Kharif seasons, resulting in a high level of precision in measuring the extent of cultivated land during the Rabi and Kharif seasons with an area of 626,947 acres and 590,858 acres, respectively. The total land area, ascertained from the observation of the comprehensive cropping pattern and agricultural modifications during the entire year (June 2021–May 2022) is 635,655 acres. The validation exercise shows the higher accuracy of this method for cropland, with an overall accuracy of 98.8%, kappa of 0.97, user accuracy of 98.69% and producer accuracy of 99.13%. Additionally, it was spatially compared with the ESRI, ESA and MODIS cropland layers and government statistical data. Furthermore, the research investigates the temporal dynamics of agricultural growth phases using spectral bands and indices. This approach improves the accuracy of precise cropland identification and provides useful insights into crop phenology.
Raza et al. (Mon,) studied this question.