A complementary two-resolution assessment of ratoon rice suitability using agroclimatic predictors and satellite embeddings in central and southern China

Introduction Ratoon rice can increase annual rice production, but regional agroclimatic conditions and local landscape heterogeneity complicate suitability assessment. Methods We developed a complementary two-resolution framework for central and southern China. Phase I used 1 km maximum entropy (MaxEnt) models combining 1981–2010 climate normals, terrain, soil, and seasonal thermal-exposure proxies under alternative early- and late-window scenarios. Phase II independently used the 2024 Google Satellite Embedding V1 Annual product at 30 m, comparing random forest (RF) feature selection with principal component analysis (PCA). Fully nested four-fold spatial validation confined feature adaptation, screening, and tuning to outer-training data, using 618 occurrences and 9,937 background points. Results Topography had the greatest permutation importance, while climatic associations differed between seasonal windows. The two 1 km baselines averaged an area under the receiver operating characteristic curve (AUC) of 0.685, a continuous Boyce index (CBI) of 0.877, and an omission rate at the 10th-percentile training-presence threshold (OR10) of 0.147. PCA achieved AUC = 0.829, CBI = 0.912, and OR10 = 0.113; RF achieved AUC = 0.828, CBI = 0.929, and OR10 = 0.109. The embedding pathways showed similar discrimination, with RF providing numerically stronger ranking consistency and omission control. Discussion Historical agroclimatic baselines and contemporary satellite representations provide complementary spatial evidence. The 30 m embedding models outperformed the tested 1 km environmental baselines, but their latent dimensions do not measure crop physiology. Temporal mismatch among occurrence records, climate normals, and 2024 imagery limits inference about current cultivation and extreme-weather exposure. The maps support field-survey prioritization rather than estimates of planted area, yield, profitability, or expansion capacity.