Atmospheric Temperature Prediction Using Ensemble Deep Learning Technique

Abstract

For the research of climatology, temperature is the indispensable parameter that plays a significant role in measuring climate changes. Climate change is required to understand the variability in climate and its impact on various activities such as agriculture, solar energy production, travel, climate conditions in extreme cold or hot places, etc. Thus, time and again, researchers have tried to find methods to predict temperature with increasing accuracy. Hitherto many scientists used to perform on high-powered HPC systems using complex and dynamic climatology models to compute temperature at future time stamps. However, in such models, the accuracy plays game adversely with prediction lead time such that sometimes with increasing lead time as need of application, the model fails to result in acceptable outcomes. With advancement of artificial intelligence, nowadays, deep learning technique and, especially, long short-term memory (LSTM) are providing better solution with higher efficiency for time series prediction problems. We have enhanced the base LSTM and introduced certain changes to solve the prediction of time series data, specially temperature in alignment with scientific applications. The experimental outcome of our proposed technique convincingly justifies the logic behind the enhanced technique, and also, it has been compared with existing approaches to be found unmatched.