Tandon et al. (2025) Rainfall Variability and Rising Extremes in Urbanizing Himalayan Foothills: A Machine Learning and data-driven Exploration of Hydroclimatic Shifts in Uttarakhand, India

Identification

• Journal: Earth Systems and Environment
• Year: 2025
• Date: 2025-10-17
• Authors: Aayushi Tandon, Amit Awasthi, Kanhu Charan Pattnayak, Sumanta Das
• DOI: 10.1007/s41748-025-00881-y

Research Groups

• Department of Applied Sciences, University of Petroleum and Energy Studies, Uttarakhand, India
• Hadley Centre for Climate, Met Office, Exeter, UK
• School of Environment and Disaster Management, Ramakrishna Mission Vivekananda Educational and Research Institute, Kolkata, India
• The University of Queensland, St. Lucia, Brisbane, QLD, Australia

Short Summary

This study investigates rainfall variability and hydroclimatic extremes in the urbanizing Himalayan foothills of Uttarakhand, India, using integrated statistical and machine learning approaches, revealing that urbanization significantly increases vulnerability to extreme rainfall events.

Objective

• To estimate spatial and temporal trends in precipitation using non-parametric statistical tools (Mann-Kendall and Sen’s Slope Estimator) across the districts of Uttarakhand.
• To characterize precipitation extremes through the distribution of quantile of extreme events and using ETCCDI-based indices such as Consecutive Dry Days (CDD) and Consecutive Wet Days (CWD).
• To examine the correlation between precipitation and key meteorological variables, thereby understanding underlying climatic controls.
• To examine the potential of machine learning models for accurately classifying and forecasting extreme rainfall events with improved precision at the regional scale.
• Principal Hypothesis: Urbanization-induced microclimatic changes, in synergy with regional climate variability, have significantly modified rainfall patterns and increased the occurrence of extreme hydroclimatic events in Uttarakhand.

Study Configuration

• Spatial Scale: 13 districts of Uttarakhand, India, covering both urban and non-urban areas. Data resolution is 0.5° × 0.625° (approximately 50 km × 70 km).
• Temporal Scale: 40 years (1984–2023) of daily meteorological data.

Methodology and Data

• Models used: Mann-Kendall (MK) trend test, Modified Mann-Kendall (MMK) trend test, Sen’s Slope Estimator (SSE), ETCCDI-based indices (Consecutive Dry Days (CDD), Consecutive Wet Days (CWD)), Pearson correlation analysis, Support Vector Machine (SVM), Random Forest (RF).
• Data sources: NASA Prediction of Worldwide Energy Resources (POWER) Project, based on Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis data. Variables include daily precipitation, dew point temperature, relative humidity, and surface pressure.

Main Results

• Urban areas exhibited significantly higher annual rainfall (e.g., Haridwar: 377.64 mm) compared to non-urban regions (e.g., Tehri Garhwal: 116.18 mm).
• Statistically significant increasing precipitation trends were observed across all districts (p < 0.05), with Sen's slope coefficients ranging from 3.80 × 10⁻⁵ mm/year to 9.06 × 10⁻⁵ mm/year (Dehradun).
• Extreme event analysis using ETCCDI indices revealed prolonged dry days (e.g., 123 days in 1999 in Almora, Pithoragarh, and Nainital) and extended wet days (e.g., 86 days in 2010 in Dehradun and Chamoli), with urban districts showing greater intensities.
• Relative humidity, dew point temperature, and surface pressure were identified as the most influential meteorological drivers of precipitation variability.
• The Random Forest (RF) machine learning model consistently outperformed the Support Vector Machine (SVM) for classifying extreme precipitation events, achieving test accuracies ranging from 0.786 (Dehradun) to 0.799 (Udham Singh Nagar) and lower Brier Scores (as low as 0.096).

Contributions

• Integration of non-parametric statistical methods (Mann-Kendall, Sen’s Slope) with machine learning algorithms (SVM, RF) to both detect and predict extreme rainfall events in a mountainous urbanizing context, identifying complex non-linear patterns.
• High-resolution, multi-district spatial analysis providing a geographically detailed assessment of precipitation shifts in Uttarakhand, uncovering localized manifestations of climate variability driven by urban microclimatic alterations.
• Characterization of rainfall extremes using standardized ETCCDI indices (CDD, CWD) coupled with data-driven classification, moving beyond basic precipitation totals to analyze the frequency, intensity, and persistence of extremes.
• Provides a scientific basis for localized climate adaptation strategies in the Himalayas and supports India’s National Action Plan on Climate Change by improving extreme weather prediction capabilities.

Funding

No external grant funding was received for this research.

Citation

@article{Tandon2025Rainfall,
  author = {Tandon, Aayushi and Awasthi, Amit and Pattnayak, Kanhu Charan and Das, Sumanta},
  title = {Rainfall Variability and Rising Extremes in Urbanizing Himalayan Foothills: A Machine Learning and data-driven Exploration of Hydroclimatic Shifts in Uttarakhand, India},
  journal = {Earth Systems and Environment},
  year = {2025},
  doi = {10.1007/s41748-025-00881-y},
  url = {https://doi.org/10.1007/s41748-025-00881-y}
}