A global regression method for thermal sharpening of urban land surface temperatures from MODIS and Landsat

• Main Authors: WANG, James. W., CHOW, Winston T. L., WANG, Yi-Chen
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2020
• Subjects: Impervious surface; Normalized differences; Regression method; Spatial resolution; Statistical scheme; Thermal infrared sensors; Urban applications; Urban land surface temperature; Environmental Sciences; Political Science; Urban Studies and Planning
• Online Access: https://ink.library.smu.edu.sg/soss_research/3048; https://ink.library.smu.edu.sg/context/soss_research/article/4305/viewcontent/urban1.pdf
• Institution: Singapore Management University

Land surface temperatures (LST) in urban landscapes are typically more heterogeneous than can be monitored by the spatial resolution of satellite-based thermal infrared sensors. Thermal sharpening (TS) methods permit the disaggregation of LST based on finer-grained multispectral information, but there is continued debate over which spectral indices are most appropriate for urban TS, and how they should be configured in a predictive regression framework. In this study, we evaluate the stability of various TS kernels with respect to LST at different spatial (Landsat 8) and diurnal (MODIS) scales, and present a new TS method, global regression for urban thermal sharpening (SGRUTS), based on these findings. Of the spectral indices examined, the normalized difference built-up index (NDBI) and the normalized multi-band drought index (NMDI) were the most spatially stable for Landsat 8 and MODIS overall. Kernel performance varied diurnally, with the index-based impervious surface index (IBI) and broadband α selected for 1030 h, NDBI and NMDI selected for 1330 h, and IBI and NMDI selected for 2230 h and 130 h, respectively. Over a range of field-validated metrics, the SGRUTS scheme comprising a two-factor interaction between NDBI and NMDI was competitive with the best alternative TS models compared. This SGRUTS model is essentially a refinement of the Enhanced Physical Method for urban applications in terms of kernel selection and configuration, and has interpretative advantages over more complex statistical schemes.