Hot school grounds, mismatched evidence: campus-referenced satellite thermal screening in Barcelona and Madrid

Abstract Urban heat is often mapped at pixel or neighborhood scale, although adaptation decisions are implemented through bounded public assets. We test a facility-aligned screening framework using Landsat 8 Collection 2 Level-2 imagery, Local Climate Zone context, and five surface indicators for 158 public primary schools in Barcelona and 276 in Madrid during July-August 2014–2024 (21,438 complete pixel-year observations). Median summer school-ground LST was 31.8°C in Barcelona and 37.2°C in Madrid; median Theil-Sen slopes were 0.09 and 0.18°C/year, while standard Mann-Kendall tests with within-city FDR control classified positive monotonic trends in 23.4% and 40.6% of schools, respectively. Under five-fold school-grouped cross-validation, Extra Trees performed best for held-out schools (R²=0.78 ± 0.05; RMSE = 2.41 ± 0.18°C), compared with R²=0.89 under a random pixel split. NDMI and NDVI ranked first and second in pooled predictive importance. Aligning measurement, validation, and interpretation with the campus changes the evidential basis for screening. The results support field-audit prioritization, not causal retrofit effects or children's thermal comfort.