WRF model evaluation for the urban heat island assessment under varying land use/land cover and reference site conditions

Urban heat island effect in Delhi has been assessed using Weather Research and Forecasting (WRF v3.5) coupled with urban canopy model (UCM) focusing on air temperature and surface skin temperature. The estimated heat island intensities for different land use/land cover (LULC) have been compared with those derived from in situ and satellite observations. The model performs reasonably well for urban heat island intensity (UHI) estimation and is able to reproduce trend of UHI for urban areas. There is a significant improvement in model performance with inclusion of UCM which results in reduction in root mean-squared errors (RMSE) for temperatures from 1.63 A degrees C (2.89 A degrees C) to 1.13 A degrees C (2.75 A degrees C) for urban (non-urban) areas. Modification of LULC also improves performance for non-urban areas. High UHI zones and top 3 hotspots are captured well by the model. The relevance of selecting a reference point at the periphery of the city away from populated and built-up areas for UHI estimation is examined in the context of rapidly growing cities where rural areas are transforming fast into built-up areas, and reference site may not be appropriate for future years. UHI estimated by WRF model (with and without UCM) with respect to reference rural site compares well with the UHI based on observed in situ data. An alternative methodology is explored using a green area with minimum temperature within the city as a reference site. This alternative methodology works well with observed UHIs and WRF-UCM-simulated UHIs but has poor performance for WRF-simulated UHIs. It is concluded that WRF model can be applied for UHI estimation with classical methodology based on rural reference site. In general, many times WRF model performs satisfactorily, though WRF-UCM always shows a better performance. Hence, inclusion of appropriate representation of urban canopies and land use-land cover is important for improving predictive capabilities of the mesoscale models.