Spatiotemporal estimation of air temperature patterns at the street level using high resolution satellite imagery

Although meteorological monitoring stations provide accurate measurements of Air Temperature (AT), their spatial coverage within a given region is limited and thus is often insufficient for exposure and epidemiological studies. In many applications, satellite imagery measures energy flux, which is spatially continuous, and calculates Brightness Temperature (BT) that used as an input parameter. Although both quantities (AT-BT) are physically related, the correlation between them is not straightforward, and varies daily due to parameters such as meteorological conditions, surface moisture, land use, satellite-surface geometry and others. In this paper we first investigate the relationship between AT and BT as measured by 39 meteorological stations in Israel during 1984–2015. Thereafter, we apply mixed regression models with daily random slopes to calibrate Landsat BT data with monitored AT measurements for the period 1984–2015. Results show that AT can be predicted with high accuracy by using BT with high spatial resolution. The model shows relatively high accuracy estimation of AT (R² = 0.92, RMSE = 1.58 °C, slope = 0.90). Incorporating meteorological parameters into the model generates better accuracy (R² = 0.935) than the AT-BT model (R² = 0.92). Furthermore, based on the relatively high model accuracy, we investigated the spatial patterns of AT within the study domain. In the latter we focused on July–August, as these two months are characterized by relativity stable synoptic conditions in the study area. In addition, a temporal change in AT during the last 30 years was estimated and verified using available meteorological stations and two additional remote sensing platforms. Finally, the impact of different land coverage on AT were estimated, as an example of future application of the presented approach.