Urban Overheating Assessment through Prediction of Surface Temperatures: A Case Study of Karachi, Pakistan

• Main Authors: Bilal Aslam, Ahsen Maqsoom, Nauman Khalid, Fahim Ullah, Samad Sepasgozar
• Format: Article
• Language: English
• Published: MDPI AG 2021-08-01
• Series: ISPRS International Journal of Geo-Information
• Subjects: urban overheating; land surface temperature; China Pakistan Economic Corridor; Karachi city; long short-term memory; artificial neural network
• Online Access: https://www.mdpi.com/2220-9964/10/8/539

Global climate has been radically affected by the urbanization process in recent years. Karachi, Pakistan’s economic hub, is also showing signs of swift urbanization. Owing to the construction of infrastructure projects under the China-Pakistan Economic Corridor (CPEC) and associated urbanization, Karachi’s climate has been significantly affected. The associated replacement of natural surfaces by anthropogenic materials results in urban overheating and increased local temperatures leading to serious health issues and higher air pollution. Thus, these temperature changes and urban overheating effects must be addressed to minimize their impact on the city’s population. For analyzing the urban overheating of Karachi city, LST (land surface temperature) is assessed in the current study, where data of the past 20 years (2000–2020) is used. For this purpose, remote sensing data from the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model (ASTER GDEM) and Moderate-Resolution Imaging Spectroradiometer (MODIS) sensors were utilized. The long short-term memory (LSTM) model was utilized where the road density (RD), elevation, and enhanced vegetation index (EVI) are used as input parameters. Upon comparing estimated and measured LST, the values of mean absolute error (MAE), mean square error (MSE), and mean absolute percentage error (MAPE) are 0.27 K, 0.237, and 0.15% for January, and 0.29 K, 0.261, and 0.13% for May, respectively. The low MAE, MSE, and MAPE values show a higher correlation between the predicted and observed LST values. Moreover, results show that more than 90% of the pixel data falls in the least possible error range of −1 K to +1 K. The MAE, MSE and MAPE values for Support Vector Regression (SVR) are 0.52 K, 0.453 and 0.18% and 0.76 K, 0.873, and 0.26%. The current model outperforms previous studies, shows a higher accuracy, and depicts greater reliability to predict the actual scenario. In the future, based on the accurate LST results from this model, city planners can propose mitigation strategies to reduce the harmful effects of urban overheating and associated Urban Heat Island effects (UHI).