RESEARCH ARTICLE Modeling Biophysical Variables and Land Surface Temperature Using the GWR Model: Case Study—Tehran and Its Satellite Cities Zahra Alibakhshi 1 • Mahmoud Ahmadi 1 • Manouchehr Farajzadeh Asl 2 Received: 11 November 2018 / Accepted: 14 October 2019 Ó Indian Society of Remote Sensing 2019 Abstract The land-cover type plays a decisive role for the land surface temperature (LST). Since cities and their satellite cities are composed of varying covers, including vegetation, built-up areas, buildings, roads, and bare areas, the main purpose of this research is to examine the LST in Tehran and its satellite cities and the cover type that contributes to increased or decreased temperature. The study investigated the relationship between NDVI, SAVI, NDBI, and NDBaI indices, as four biophysical variables, and LST over a period of 15 years (2001–2015) by the geographically weighted regression (GWR) model using imagery of Landsat 7. The results showed that the relationship between LST and NDBI is stronger than the associations with other variables. In 2010, biophysical variables had the greatest effect on LST. Using the GWR model, the local R 2 map was drawn for the studied area, showing that the highest value for the coefficient of determination belonged to Islamshahr and Shahriar because of the homogeneity of the land cover in these cities. Keywords Land surface temperature Á Biophysical variables Á GWR model Á Tehran Á Satellite cities Introduction Urbanization and human activities affect the climate of cities and surface temperatures. Ground-based weather stations cannot provide sufficient land surface temperature (LST) data as they are not well distributed within the area (Hereher 2017). One of the most effective methods for measuring surface temperatures worldwide with high temporal and spatial resolution is remote sensing (Li et al. 2013). The spatial distribution of LST varies according to the type of land cover (Voogt and Oke 2003; Ali and Shalaby 2012). Geographically weighted regression (GWR) is a spatial statistical method for spatial modeling of heterogeneous processes, which allows the relationship between response variables and a set of auxiliary variables to be different across geographic locations (Brunsdon et al. 1996, 1998; Fotheringham et al. 1996, 1997, 2003). A major component of GWR is the space weight by which the spatial relationships are created. Usually, space weights are defined by spatial nuclear functions such as Gaussian or bisquare functions (Fotheringham et al. 2003), in which weights are related to closer observations. The GWR model provides a more precise prediction for the response vari- able (Hession and Moore 2011; Chu 2012). This model can estimate regression coefficients in each situation (Ahmadi et al. 2018b). GWR is a new approach to modeling heterogeneous spatial processes and, due to its greater analytical capa- bility and further details, leads to increased accuracy and efficiency (Ahamdi et al. 2018b). GWR approaches are methods of exploring spatial variations (Is ¸ik and Pinar- ciog ˘lu 2006; Mennis 2006; Wen et al. 2010). Application of the GWR model is limited for certain reasons: First, the results of the model are very sensitive to the kernel type and bandwidth methods (Wheeler and Tiefelsdorf 2005; Wu and Qiu, 2011); second, nonlinear relations cannot be added to the model, and its inference does not occur in the model (Fotheringham et al. 2003). The LSTs of Tehran (as the capital of Iran) and its neighboring cities have undergone changes in recent years due to the population growth, built-up areas, and changes & Zahra Alibakhshi zalibakhshi7@gmail.com 1 Faculty of Earth Sciences, Shahid Beheshti University, Tehran, Iran 2 Department of Natural Geography, Tarbiat Modares University, Tehran, Iran 123 Journal of the Indian Society of Remote Sensing https://doi.org/10.1007/s12524-019-01062-x