Analysis and Forecasting of Temporal Rainfall Variability Over Hundred Indian Cities Using Deep Learning Approaches

Abstract India, a topographically and meteorologically rich country, has a vast range of rainfall variability. The impacts could be realized across various sectors, including agriculture, industry, tourism, etc. With the increasing impacts of changing climate, more intense extreme rainfall events ar...
Ausführliche Beschreibung

Abstract India, a topographically and meteorologically rich country, has a vast range of rainfall variability. The impacts could be realized across various sectors, including agriculture, industry, tourism, etc. With the increasing impacts of changing climate, more intense extreme rainfall events are expected to trigger severe floods, landslides, etc., in future. Therefore, it is imperative to make a precise prediction so that the intensity of the impacts on life and property could be reduced. Rather than using the computationally expensive conventional numerical modeling, the data driven AI/ML frameworks could be adopted in forecasting rainfall trends and patterns. The present work is an effort in this direction, which uses a monthly accumulated gridded rainfall dataset and a monthly averaged daily mean temperature dataset from 1901 to 2021 with a resolution of 0.5° × 0.5° for the analysis and prediction of yearly rainfall patterns across India through a city-specific approach. Accordingly, comparison of deep learning (DL) algorithms like Long Short Term Memory (LSTM), Bi-directional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and Convolution 1D LSTM (Conv1DLSTM) are performed for long-term rainfall prediction over hundred cities of India. The statistical parameters including Root means square error (RMSE), Mean absolute error (MAE), Coefficient of Determination ($ R^{2} $), and Nash–Sutcliffe efficiency (NSE) are estimated to assess the robustness of the considered DL models and identify the better performing one. The initial results indicated that for univariate forecasting of accumulated monthly rainfall, Conv1DLSTM performed better while for bivariate forecasting, GRU performed better than the others. City-based rainfall trend analysis using the seasonal Mann–Kendall (MK) test suggested increasing trend over northwestern region, decreasing trend over northeastern region and no significant trend over other cities. The DL model-based forecast realized that temporal rainfall variability may be altered in future over some cities, attributable to the changing climate scenario. These models could reasonably capture the low and moderate intensity rainfall variabilities, though the very high intensity scenarios exhibited indifferent results, where the performance of the considered DL frameworks is found to be limited. Ausführliche Beschreibung

Autor*in:	Singh, Sanjeev [verfasserIn] Mukherjee, Asmita [verfasserIn] Panda, Jagabandhu [verfasserIn] Choudhury, Animesh [verfasserIn] Bhattacharyya, Saugat [verfasserIn]

Format:	E-Artikel
Sprache:	Englisch

Erschienen:	2024

Schlagwörter:	Rainfall ML DL LSTM BiLSTM GRU Conv1DLSTM

Anmerkung:

© King Abdulaziz University and Springer Nature Switzerland AG 2024. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Übergeordnetes Werk:	Enthalten in: Earth systems and environment - Springer International Publishing, 2017, 8(2024), 3 vom: 20. Apr., Seite 599-625
Übergeordnetes Werk:	volume:8 ; year:2024 ; number:3 ; day:20 ; month:04 ; pages:599-625

Links:	Volltext

DOI / URN:	10.1007/s41748-024-00396-y

Katalog-ID:	SPR057375879