Modeling diurnal land temperature cycles over Los Angeles using downscaled GOES imagery

Land surface temperature is a key parameter for monitoring urban heat islands, assessing heat related risks, and estimating building energy consumption. These environmental issues are characterized by high temporal variability. A possible solution from the remote sensing perspective is to utilize geostationary satellites images, for instance, images from Geostationary Operational Environmental System (GOES) and Meteosat Second Generation (MSG). These satellite systems, however, with coarse spatial but high temporal resolution (sub-hourly imagery at 3–10 km resolution), often limit their usage to meteorological forecasting and global climate modeling. Therefore, how to develop efficient and effective methods to disaggregate these coarse resolution images to a proper scale suitable for regional and local studies need be explored. In this study, we propose a least square support vector machine (LSSVM) method to achieve the goal of downscaling of GOES image data to half-hourly 1-km LSTs by fusing it with MODIS data products and Shuttle Radar Topography Mission (SRTM) digital elevation data. The result of downscaling suggests that the proposed method successfully disaggregated GOES images to half-hourly 1-km LSTs with accuracy of approximately 2.5 K when validated against with MODIS LSTs at the same over-passing time. The synthetic LST datasets were further explored for monitoring of surface urban heat island (UHI) in the Los Angeles region by extracting key diurnal temperature cycle (DTC) parameters. It is found that the datasets and DTC derived parameters were more suitable for monitoring of daytime- other than nighttime-UHI. With the downscaled GOES 1-km LSTs, the diurnal temperature variations can well be characterized. An accuracy of about 2.5 K was achieved in terms of the fitted results at both 1 km and 5 km resolutions.