MONSOON PREDICTION • MACHINE LEARNING

Indian Monsoon

Enhanced Agent-Based Modeling with Orographic Physics

Including Himalayan Barrier & Western Ghats Rain Shadow Effects

868.6 mm
Long Period Average
4 months
June-September
70%
Annual Rainfall

The Challenge

Predicting the Indian monsoon is one of the most challenging problems in meteorology

1.4 Billion People

Depend on monsoon for agriculture, water supply, and livelihoods

60% of Agriculture

Rain-fed agriculture directly dependent on monsoon timing and intensity

±5% Accuracy

IMD's current seasonal forecast skill (post-2007 improvements)

Current Prediction Methods

Statistical (SEFS)

Regression models using ENSO, IOD indices

Dynamical (MME)

Multi-model ensemble of global climate models

Key Factors

Multiple atmospheric and oceanic phenomena control monsoon variability

El Niño/La Niña (ENSO)

Pacific Ocean temperature patterns strongly influence monsoon strength

Indian Ocean Dipole (IOD)

Temperature gradient across Indian Ocean affects moisture transport

Ocean Currents

Somali Current and monsoon currents drive upwelling and SST patterns

Trade Winds

Cross-equatorial flow and westerlies determine moisture flux

Global Warming

Rising temperatures increase atmospheric moisture capacity (+2.5 mm/year trend)

Himalayan Barrier NEW

Orographic lifting + Tibetan Plateau heating enhance rainfall up to 75% in foothills

Western Ghats Shadow NEW

Windward rainfall enhancement (2-3x) and leeward rain shadow (40-60% reduction)

Enhanced Agent-Based Modeling

Our approach models these factors as interacting agents that collectively produce monsoon rainfall patterns. The enhanced version includes orographic physics from the Himalayan mountain barrier and Western Ghats rain shadow, improving test accuracy by 34%. By learning from historical data (2010-2019), the model predicts future monsoon behavior (2020-2024) with RMSE of just 33.7 mm.

Data & Training

Data Sources

IITM AISMR

All-India Summer Monsoon Rainfall (1871-2024)

IMD Gridded Data

Daily rainfall at 1° resolution

Atmospheric Indices

ENSO, IOD, EQUINOO time series

Ocean Data

SST, current patterns, upwelling indices

Training Approach

Split Strategy

Training: 2010-2019 (10 years)
Testing: 2020-2024 (5 years)

Features

ENSO, IOD, ocean currents, trade winds, warming trend

Objective

Minimize prediction error on unseen test years

15 years
Historical Data (2010-2024)
1,830 days
Daily Rainfall Records
+35 mm
Warming Impact (2010-2024)

Results & Validation

Enhanced Model Performance

34% RMSE Improvement with Orographic Physics
Test RMSE: 50.9 mm → 33.7 mm
33.7 mm
Enhanced Test RMSE
(2020-2024)
100%
Categorical Accuracy
Normal/Deficient/Excess
+0.94
R² Improvement
-0.673 → 0.267

IMD Reference (2021-2024)

100% forecast accuracy within ±5% margin

Typical error: ~43 mm (±5% of LPA 868.6 mm)

Methods: Statistical Ensemble (SEFS) + Multi-Model Ensemble (MME)

Source: India Meteorological Department

Training: 2010-2019 (10 years) | Testing: 2020-2024 (5 years) | Multi-CPU Optimized

Interpretability

Model reveals which factors (ENSO, IOD, warming) drive each year's prediction

Efficiency

Runs in seconds on laptop vs hours on supercomputer for dynamical models

Climate Awareness

Explicitly incorporates global warming trend (+2.5 mm/year increase)

While the model shows higher RMSE than operational forecasts, it provides valuable insights into monsoon drivers and can be rapidly improved through expanded training data and ensemble methods.

Climate Change Impact

+2.5 mm/year

Estimated monsoon rainfall increase due to global warming

Intensification

Warmer atmosphere holds more moisture (Clausius-Clapeyron: ~7% per °C)

Extremes

More intense rainfall events and longer dry spells between active periods

Total Impact (2010-2024)

2010 baseline: 868.6 mm
Warming contribution: +35.0 mm
2024 expectation: ~903.6 mm

Agriculture

Changing patterns require adaptive crop selection and irrigation strategies

Water Resources

Reservoir management must account for increased variability and extremes

Flood Risk

More intense rainfall events increase urban and riverine flooding

Sources & References

Data Sources

IMD Prediction Methods

Key Publications

  • Gadgil, S. (2003). "The Indian Monsoon and its Variability." Annual Review of Earth and Planetary Sciences
  • Rajeevan, M. et al. (2007). "New statistical models for long-range forecasting of southwest monsoon rainfall"
  • Webster, P. J. et al. (1998). "Monsoons: Processes, predictability, and prospects for prediction"

Indian Monsoon Prediction Study | 2025