Impact of Urbanization on Land Surface Temperature Using Deep Learning Approach

摘要

Land use change and urbanization are the two astonishing factors can introduce ambiguity to the estimation of global temperature trends related to climate change. In this work, we investigate the characteristics of urban expansion and its impact on land surface temperature using a time series of Landsat images. Acurate land cover information required in order to relate land surface temperature with land cover features. Even with increased number of satellite systems and sensors acquiring data with improved spectral, spatial, radiometric and temporal characteristics of remotely sensed image with an ambiguous accuracy. One major approach for improving accuracy is to develop an accurate and effective image classification algorithm. This work incorporates a long short-term memory (LSTM) recurrent neural network (RNN) model to take advantage for time series images to improve accuracy and reduce complexity. Network trained thoroughly using state of the art techniques of deep learning and finally, we tested our model on multiple Landsat images to derive urban land cover changes (ULCC). Then, Land surface temperature has been calculated from thermal data of Landsat TM/TIRS using emissivity derived from NDVI images. Urban density as a measure of urbanization has been derived from gravity model. Later, land surface temperature data associated with land use and land cover information for further investigation of the relationship between land surface temperature behavior with land cover features. The results provide a scientific reference for policy makers and urban planners can work towards a sustainable and healthy environment.