Prediction of the Future Temperature of Baghdad City by Land Surface Temperature (LST) Dynamics Using the BiLSTM Deep Learning Model

Abstract

In recent years, land surface temperature (LST) has become an increasing concern because of the rise in urban temperatures and the accompanying microclimatic warming. The utilization of artificial intelligence models to predict variations in LST is highly beneficial for assessing and forecasting the dynamic climatic changes occurring worldwide. However, the prediction of LST is a difficult task because slight errors in its short-term forecasts can accumulate to become significant errors over longer periods of time. In this paper, a hybrid model that utilizes a bi-directional long short-term memory (BiLSTM) framework is presented for improving the accuracy of long-term LST prediction. The goal is to forecast the future patterns of LST and their possible effects on the urban microclimate of Baghdad city. A high-resolution land cover and land use map for Baghdad City, as well as data collected from satellite photos, were used in this work to construct a surface temperature forecast model. Based on the data analysis, Baghdad experienced the greatest temperature rises from 2001 to 2018, where a fast staggering in LST occurred at >35°C in 2018 due to the net change in the Baghdad area, which was 12.8%. The prediction results show that the proposed BiLSTM model can significantly increase the accuracy of long-range weather forecasts for Baghdad. The results show that the mean-squared error of 0.53 and the correlation coefficient of 0.84 between the predicted and actual LST indicate good accuracy. Hence, the proposed model can be used to estimate future LSTs in Baghdad with low error.