Deep learning models comparable assessment and uncertainty analysis for diurnal temperature range (DTR) predictions over Indian urban cities

Abstract

For monthly time spans of Indian cities (Ahmedabad, Bengaluru, Guwahati, Kolkata, and New Delhi), this article used daily (max and min) temperature data from 1951 to 2020 and approximated diurnal temperature range (DTR). RClimDex (a user interface for extreme computing indices) was used to do statistical analysis and comparisons of climatological characteristics such time series, means, extremes, and trends. During these years, the DTR trend in the researched area was more variable. This research examines the appropriateness of three deep neural networks [recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU)] for one-step-ahead DTR time series (DTRTS) prediction. To test the efficiency of models in the testing set, six statistical error metrics were used [Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of correlation (R), Percent Bias (PBIAS), modified index of agreement (md), and relative index of agreement (rd). The forecasting inaccuracy in predicting the outcome The Wilson score approach was used to do a quantitative uncertainty analysis on DTR. The results show that the LSTM outperforms the other two models in terms of forgetting, remembering, and updating information. It is more accurate on datasets with longer sequences and exhibits substantially higher volatility throughout its gradient descent. It got concluded with LSTM being well adapted to learning from experience to categorize, analyze, and predict time series, and it may be employed as a new dependable artificial intelligence technique for DTRTS forecasting.