India’s diverse agricultural landscapes demand a single Land Use and Land Cover (LULC) product integrating both intra-annually static (built-up, tree cover, barren land) and dynamic (water seasonality, cropping intensity) classes for sustainable Natural Resource Management (NRM). Existing LULC products suffer from limited thematic coverage of dynamic processes, imprecise delineation of fragmented smallholder features, limited reproducibility, and poor performance of monolithic classifiers on spectrally similar categories. We introduce a hierarchical decision-tree framework that breaks complex classification tasks into targeted sub-tasks, offering methodological improvements in detecting monsoon water, delineating tree-croplands, and classifying cropping intensity. Class-wise evaluations demonstrate superior performance: SAR water detection achieves an NRMSE of 0.33 (vs. baselines 0.53–0.75), tree-cropland macro-average of 0.94, and cropping intensity macro-average of 0.88 (vs. baseline 0.77). Crucially, this study ensures transparency and reproducibility in LULC mapping. We publicly release four curated datasets alongside the entire classification pipeline implemented on the Google Earth Engine commodity platform. The resulting pan-India output maps at 10m resolution are hosted on CoRE-Stack (digital public good) for easy accessibility and analysis 24, already powering real-world applications in water-security planning 61 and agricultural studies 50.
Bansal et al. (2026) studied this question.