My research focuses on understanding and mapping urban near-surface air temperature (Ta) at high spatial resolution. In cities, Ta can change markedly over short distances because land cover and urban form are highly heterogeneous—dense built-up areas, vegetation patches, water bodies, and varying street canyons all modify how energy is absorbed, stored, and released. These fine-scale temperature differences shape the Urban Heat Island (UHI) and strongly influence heat-exposure assessments, outdoor thermal comfort, and building-energy analyses. A key objective of my work is to produce accurate, spatially continuous Ta fields that represent this variability in a physically meaningful way, rather than relying only on sparse weather stations or coarse reanalysis products.

To achieve this, I develop hybrid approaches that combine dense in-situ observations, multispectral satellite remote sensing, and simplified process-based modelling. Satellite imagery (e.g., Landsat and Sentinel-2) provides spatial predictors linked to urban climate drivers, such as land surface temperature (LST), vegetation and built-up indicators (NDVI, NDBI), terrain properties (DEM), and surface radiative parameters like albedo and absorptance. These datasets offer detailed and city-wide coverage, but translating them into near-surface Ta requires methods that can handle local effects and non-linear relationships.

A central aspect of my research is integrating machine learning (ML) with physical interpretability. I use ML models to learn the relationship between station-measured Ta and satellite-derived predictors, including approaches that explicitly account for spatial heterogeneity, such as geographically weighted methods. This allows the model to adapt to neighborhoods with different urban textures and land-cover regimes. At the same time, I develop and apply a simplified, dynamic Surface Energy Balance (SEB) model that estimates surface temperature states (Ts) and energy/flux indicators from minimal inputs (e.g., radiative properties from Sentinel-2 combined with meteorological observations). SEB-derived variables provide a process-based description of the local surface–atmosphere exchange and can be used as additional predictors—or as soft physical guidance—to improve robustness and ensure that Ta estimates remain physically plausible.

Overall, my work sits at the intersection of urban climate science, remote sensing, and interpretable modelling. The outcomes are high-resolution, uncertainty-aware Ta maps suitable for UHI characterization, hotspot identification, and microclimate-aware applications. By delivering spatial Ta fields that reflect both data patterns and underlying physics, this research supports more realistic urban heat-risk assessments and provides improved climatic forcing for building-energy modelling workflows (e.g., UBEM) and urban sustainability studies.

Click here to view my posters.