From Earth Observation to Landscape Units: A Framework for Urban Cooling Assessment in Polish Functional Urban Area

Urban environmental monitoring increasingly requires methods that translate Earth observation (EO) data into spatial units relevant to ecosystem-service assessment and planning. This study develops a landscape-unit framework for smart monitoring of urban cooling in two Polish Functional Urban Areas (FUAs): Zielona Góra and Gorzów Wielkopolski. Landsat-derived land surface temperature and surface urban heat island (SUHI) intensity were integrated with vegetation, water, imperviousness and NDVI indicators, together with the characteristics of directly adjacent landscape units. Spearman correlations, ordinary least squares and Spatial Error Models (SEMs) were used to diagnose the spatial structure of thermal conditions. The SEM variants removed residual spatial autocorrelation and showed the strongest full-sample fit, with RMSE values of approximately 0.85 SUHI units and MAE values of approximately 0.66; these statistics are used as model-fit diagnostics rather than as measures of out-of-sample predictive performance. The results were translated into a continuous weighted HeatCool Index and seven Jenks natural-break classes that represent the relative capacity of landscape units and their neighbourhoods to support cooling or reinforce warming. The classification is interpreted as an ecosystem-service screening product rather than as an independent validation of its component variables. Independently modelled annual and seasonal NO2 indicators were then aggregated to the same landscape units as a complementary pressure layer. After EDM correction, the NO2 model achieved R2 = 0.77, RMSE = 3.98 µg m−3 and MAE = 2.36 µg m−3 on the held-out validation subset. Across the HCI gradient, Classes 1–2 showed the lowest and least variable annual NO2 conditions, whereas Classes 6–7 exhibited the strongest winter deterioration and Class 7 contained the highest annual NO2 concentrations. Because NO2 was not used to construct the HCI, these gradients represent an independent comparison of co-occurring thermal and air-pollution pressures rather than evidence of a causal relationship. The proposed framework extends pixel-based thermal assessment by integrating landscape-unit classification, neighbourhood context and independently modelled air-pollution pressure within a common planning-oriented spatial framework.