Answers are grounded in this paper and cite their sources.
Field-scale soybean-yield models based on canopy imagery may miss soil-surface variation. We evaluated a two-flight drone framework combining pre-season bare-soil and pod-filling multispectral imagery at 101 locations in four slope units in Northeast China. Five random-forest scenarios were compared using nested feature selection and repeated outer-test evaluation. Adding the bare-soil proxy pool to pod-filling vegetation indices increased mean outer-test predictive R 2 from 0.240 to 0.325 and reduced mean absolute error from 573.06 to 518.76 kg ha −1 ; this error decreased in all 10 repeats. Model attribution showed contributions from canopy and bare-soil predictors. Predictions overestimated low yields and underestimated high yields. For this combined scenario, pooled predictive R 2 declined to 0.161 when each slope unit was excluded from model development. Pre-season bare-soil imagery therefore provided modest complementary information beyond a pod-filling acquisition, but the framework remains a field-scale proof of concept requiring multi-year and multi-site validation.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: