Remotely sensed retrieval of Local Climate Zones and their linkages to land surface temperature in Harare metropolitan city, Zimbabwe

This study sought to determine Local Climate Zones (LCZs) in Harare metropolitan City, using Landsat 8 multi-spectral and multi-temporal data. The World Urban Database and Access Portal Tool (WUDAPT) and Support Vector Machine classifiers were applied. Training datasets were extracted from Google Earth as prescribed by the WUDAPT procedure. Before image classification, we tested the separability of the LCZs, using the Transformed Divergence Separability Index (TDSI) based on the digitized training datasets and Landsat 8 data. Derived LCZs were then linked with Landsat 8 derived Land Surface Temperature (LST) for the cool and hot seasons. TDSI values greater 1.9 were obtained indicating that LCZs were highly separable. Comparatively, the WUDAPT method produced more accurate LCZs results (Overall accuracy = 95.69%) than the SVM classifier (Overall accuracy = 89.86%) based on seasonal Landsat 8 data. However, SVM derived accuracies were within the acceptable range of at least 80% (overall accuracy) in literature. Further, LST was observed to be high in LCZs with high built-up density and low vegetation proportion, when compared to other zones. Due to high proportion of vegetation, sparsely built areas were at least 1 degrees C cooler. Although LCZs are usually linked at 2 m air temperature, they also strongly explain LST distribution. This work provides insight into the importance of mapping LCZs in third world countries where such information remains scarce.