Abstract Extreme rainfall events in India, particularly in vulnerable mountainous regions such as Wayanad (Kerala), have become more frequent and intense, often leading to floods and landslides. Accurate simulation of such events using numerical weather prediction (NWP) models (such as the weather research and forecast (WRF) model) is essential to better understand the role of atmospheric dynamics in such events. This study examines the influence of radiation-scheme update frequency on WRF simulated rainfall during the 2024 Wayanad extreme event using three physics suites – tropical, Kerala, and Uttarakhand. Each suite is tested with default (recommended) and frequent (every time step) radiative configurations. The model fidelity was evaluated against IMERG observations using root mean square error (RMSE), correlation coefficient (CORR), structural similarity index measure (SSIM), mutual information (MI), and object based verification through the contiguous rain area (CRA) method. Our results indicate that the tropical suite with recommended updates (Case 1) achieved the lowest RMSE (0.1766), highest CORR (0.65), and realistic rainfall peaks (738.12 mm/hr versus 716.31 mm/hr in IMERG) though CRA revealed large displacement errors. The Kerala suite configuration exhibited more balanced CRA error contribution, with moderate spatial displacement and enhanced informational similarity, consistent with higher MI. The Uttarakhand suite showed minimal displacement errors (3–4%) but was dominated by large volume errors (80%), explaining high SSIM (0.8620 in Case 5) despite poor intensity representation. Frequent radiation updates improved early convection and spatial coherence in tropical domains but degraded mature-phase intensity and orographic rainfall. The findings highlight that frequent updates benefit tropical convection, whereas recommended intervals remain preferable for complex terrain. Research highlights <UnorderedList Mark="Bullet"> <ItemContent> Frequent radiative scheme calls generally improve spatial similarity and information content in rainfall simulations. </ItemContent> <ItemContent> Structural similarity ensures coherent spatial rainfall patterns, while mutual information captures variability and detail. </ItemContent> <ItemContent> A trade-off may exist between enhancing rainfall variability and maintaining spatial coherence. </ItemContent> <ItemContent> Dual-metric validation using SSIM and MI provides deeper insight into model performance. </ItemContent> <ItemContent> Radiative call frequency should be tuned based on regional meteorological dynamics rather than a fixed configuration. </ItemContent> </UnorderedList>

Structural similarity and mutual information-based evaluation of radiative call frequency in WRF: insights from the 2024 Wayanad rainfall

期刊论文
发表日期：

2026年5月25日

查看全文

M R Mohamed Aksath Rahil,
Srinivasa Ramanujam Kannan

摘要

Abstract

Extreme rainfall events in India, particularly in vulnerable mountainous regions such as Wayanad (Kerala), have become more frequent and intense, often leading to floods and landslides. Accurate simulation of such events using numerical weather prediction (NWP) models (such as the weather research and forecast (WRF) model) is essential to better understand the role of atmospheric dynamics in such events. This study examines the influence of radiation-scheme update frequency on WRF simulated rainfall during the 2024 Wayanad extreme event using three physics suites – tropical, Kerala, and Uttarakhand. Each suite is tested with default (recommended) and frequent (every time step) radiative configurations. The model fidelity was evaluated against IMERG observations using root mean square error (RMSE), correlation coefficient (CORR), structural similarity index measure (SSIM), mutual information (MI), and object based verification through the contiguous rain area (CRA) method. Our results indicate that the tropical suite with recommended updates (Case 1) achieved the lowest RMSE (0.1766), highest CORR (0.65), and realistic rainfall peaks (738.12 mm/hr versus 716.31 mm/hr in IMERG) though CRA revealed large displacement errors. The Kerala suite configuration exhibited more balanced CRA error contribution, with moderate spatial displacement and enhanced informational similarity, consistent with higher MI. The Uttarakhand suite showed minimal displacement errors (3–4%) but was dominated by large volume errors (80%), explaining high SSIM (0.8620 in Case 5) despite poor intensity representation. Frequent radiation updates improved early convection and spatial coherence in tropical domains but degraded mature-phase intensity and orographic rainfall. The findings highlight that frequent updates benefit tropical convection, whereas recommended intervals remain preferable for complex terrain.

Research highlights

Frequent radiative scheme calls generally improve spatial similarity and information content in rainfall simulations.

Structural similarity ensures coherent spatial rainfall patterns, while mutual information captures variability and detail.

A trade-off may exist between enhancing rainfall variability and maintaining spatial coherence.

Dual-metric validation using SSIM and MI provides deeper insight into model performance.

Radiative call frequency should be tuned based on regional meteorological dynamics rather than a fixed configuration.