Cropping Pattern Mapping in an Agro-Natural Heterogeneous Landscape Using Sentinel-2 and Sentinel-1 Satellite Datasets

The quantity of land covered by various crops in a specific time span, referred to as cropping pattern, dictates the level agricultural production. However, retrieval this information at landscape scale can be challenging, especially when high spatial resolution imagery is not available. This study hypothesized that utilizing unique advantages multi-date and medium freely available Sentinel-2 (S2) reflectance bands (S2 bands), their vegetation indices (VIs) phenology (VP) derivatives, Sentinel-1 (S1) backscatter data would improve pattern mapping heterogeneous landscapes using robust machine learning algorithms, i.e., guided regularized random forest (GRRF) for variable selection (RF) classification. study’s objective was map patterns within three sub-counties Murang’a County, typical African smallholder farming area, Kenya. Specifically, performance eight classification scenarios compared, namely: (i) only S2 bands; (ii) VIs; (iii) VP; (iv) S1; (v) bands, VIs, (vi) VP, (vii) (viii) S1. Reference dominant non-croplands were collected. GRRF algorithm used select optimum variables each scenario, RF perform scenario. highest overall accuracy 94.33% with Kappa 0.93, attained GRRF-selected scenario S2, Furthermore, McNemar’s test significance did show significant differences (p ≤ 0.05) among tested scenarios. demonstrated strength selecting most important synergetic advantage S1 derivatives accurately small-scale farming-dominated landscapes. Consequently, approach other sites relatively similar agro-ecological conditions. Additionally, these results understand sustainability food systems model abundance spread crop insect pests, diseases, pollinators.