Answers are grounded in this paper and cite their sources.
Accurate mapping of coffee production systems remains challenging in heterogeneous smallholder landscapes, where intercropping, variable shade, complex topography, and seasonal dynamics reduce class separability in remote-sensing data. This challenge is particularly critical in Vietnam's Central Highlands, where coffee underpins rural livelihoods but is increasingly affected by climate variability and land-use change.In this study, we develop an interpretable, provincial-scale workflow for mapping coffee production systems in Dak Lak, Vietnam, for the 2024 mapping year by integrating multi-temporal Sentinel-2, Landsat 8/9, Sentinel-1, and topographic predictors within a Random Forest framework, combined with systematic feature selection and SHapley Additive exPlanations (SHAP)-based interpretation. Spatial block cross-validation, the more demanding test of transferability in which training and validation blocks are geographically separated by 10-20 km, yielded coffee-subclass macro F1-scores of only 0.141-0.374, revealing limited ability to generalize the three-way coffee-system classification beyond reference-represented landscapes. By contrast, stratified five-fold cross-validation within those same landscapes achieved a substantially higher overall accuracy of 95.90 ± 1.13% and coffee-subclass macro F1-score of 0.909 ± 0.033, illustrating how validation design shapes the apparent reliability of coffee-system mapping.Model interpretation consistently identifies dry-season optical moisture, vegetation–soil contrast, and topographic variables as key drivers, with Sentinel-1 providing complementary structural information.The resulting map estimates 227,721 ha of coffee, 6.4% higher than the province-wide official total, although district-level residuals are substantially larger and vary in sign. An exploratory area-reweighting analysis further indicates potentially large class-specific discrepancies not evident from the province-wide total alone. Multi-sensor fusion provides a useful regional screening approach for mapping coffee extent and major land-cover classes, but production-system distinctions should be interpreted cautiously outside areas represented by the reference data, and a coordinate-based parcel-independence analysis showed that within-sample reliability for rubber and, to a lesser extent, newly planted coffee is also not well constrained by the current sample.
No takes yet. Share an insight, caveat, or question.
Đức et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: