Variation and Prediction Analysis of 2m Air Temperature for Different Zones of the Indian Region

Abstract

Time series forecasting is a method that predicts future values by analyzing past values. Temperature alarms are valuable predictions because they are used to safeguard life and property and to increase operational performance. Here 2m air temperature refers to the temperature of air recorded at 2 meters from the ground. Paper consists of a time-series prediction of temperature data that has been taken from automatic weather stations(AWS) installed by the Indian Space Research Organisation. Given paper presents the applicability of different machine learning(ML) algorithms like convolutional neural network(CNN), long short term memory(LSTM), and autoregressive integrated moving average(ARIMA) algorithms for the validity of temperature prediction over four different stations which are Ahmedabad, Balasore, Coimbatore, and Udaipur. These stations record the hourly-based temperatures. Based on different datasets, the prediction accuracy of algorithms is compared. The paper discusses the results showing that by applying different algorithms to the different datasets with different characteristics, it is observed that the various algorithms behave distinctly with numerous 1-dimensional datasets based on the variation in recorded values, location, or type of the input data i.e hourly input data or daily input data. This analysis shows that different machine learning algorithms have a different performance ratio while applied to various data sets. Locally weighted regression is performed over all the datasets and the results are compared.