How should I split spatiotemporal data for a random forest time‑series regression to avoid "leakage" from spatial and temporal autocorrelation? – stats.stackexchange.com

I am training a random forest to predict hourly land‑surface temperature (LST) from hourly environmental covariates on a 0.05° grid. I used a random row‑wise 70/30 train/test split but suspect leakage ...