Improving the estimation of urban surface emissivity based on sub-pixel classification of high resolution satellite imagery

Information about the spatial distribution of urban surface emissivity is essential for surface temperature estimation. The latter is critical in many applications, such as estimation of surface sensible and latent heat fluxes, energy budget, urban canopy modeling, bio-climatic studies and urban planning. This study proposes a new method for improving the estimation of urban surface emissivity, which is primarily based on spectral mixture analysis. The urban surface is assumed to consist of three fundamental land cover components, namely vegetation, impervious and soil that refer to the urban environment. Due to the complexity of the urban environment, the impervious component is further divided into two land cover components: high-albedo and low-albedo impervious. Emissivity values are assigned to each component based on emissivity distributions derived from the ASTER Spectral Library Version 2.0. The fractional covers are estimated using a constrained least absolute values algorithm which is robust to outliers, and results are compared against the ones derived from a conventional constrained least squares algorithm. Following the proposed method, by combining the fraction of each cover component with a respective emissivity value, an overall emissivity for a given pixel is estimated. The methodology is applicable to visible and near infrared satellite imagery, therefore it could be used to derive emissivity maps from most multispectral satellite sensors. The proposed approach was applied to ASTER multispectral data for the city of Heraklion, Greece. Emissivity, as well as land surface temperature maps in the spectral region of 10.25–10.95 μm (ASTER band 13) were derived and evaluated against ASTER higher level products revealing comparable error estimations. An overall RMSE of 0.014776 (bias = −0.01239) was computed between the estimated emissivity obtained using the proposed methodology and the ASTER higher level product emissivity (AST05). The respective overall RMSE value for derived LST was found equal to 0.816935 K (bias = 0.67826 K).

► Development of a new method for deriving land surface emissivity over urban areas.
► Spectral mixture analysis approach to handle the complexity of urban environment.
► Applicable also to visible and near infrared satellite imagery.
► Least absolute value approach used to solve spectral mixing.
► Results present an improvement in emissivity estimation compared to existing methods.