The Farmer and the Engineer: What AI Can and Cannot Tell India About the Monsoon

Source: Original editorial ↗ Hindustan Times

Resources for this Article

Download PDF Print-ready A4 notes · Free

The Lift Line

“The farmer can expect better answers soon. The engineer must build now, with room for changes as the answers improve.”

Why This Editorial Matters for Your Exam

The monsoon is core GS1 Geography; forecasting, climate change and disaster preparedness are GS3. This article, by Tapio Schneider (Caltech) and Sandeep Juneja (Ashoka University), gives a clear framework that answers often lack: the difference between weather prediction (days ahead, where AI is strong) and climate projection (decades ahead, where only physics can reach). It also offers concrete Prelims facts on why models struggle with the monsoon.

GS Paper 1: Important geophysical phenomena; the Indian monsoon and its mechanism. GS Paper 3: Climate change; science and technology in everyday life; disaster management; agriculture and weather risk.

Concept	Meaning	Why it is testable
Weather forecast	Prediction of conditions over hours to about two weeks	Where AI now performs well
Climate projection	Scenario-based estimate of conditions decades ahead	Needs physics-based models
Active and break spells	Wet and dry phases within the monsoon season	Crucial for sowing decisions
Convection	Rising warm, moist air that forms rain clouds	Too small for global model grids
Aerosols	Fine particles that reflect sunlight and seed clouds	Pull the monsoon in the opposite direction to greenhouse gases

Background and Context

What drives the monsoon. The authors list four influences: the land-ocean temperature contrast, sea surface temperatures in the Pacific and Indian Oceans, the Himalayas steering winds, and aerosols from air pollution over the Indo-Gangetic Plain, which reflect sunlight and can serve as cloud seeds.

Why models struggle. Monsoon rain is produced by convective clouds a few kilometres across, while global climate models work on grids with cells 10 to 100 kilometres wide. Because the monsoon is the net result of competing influences, small errors in one can change where and how much it rains.

The India context. As background, the India Meteorological Department (IMD), under the Ministry of Earth Sciences, issues the official forecasts, with research support from the Indian Institute of Tropical Meteorology (IITM), Pune. The Union Cabinet approved Mission Mausam in September 2024 to upgrade observations, modelling and forecasting. The IndiaAI Mission provides shared computing capacity, which the authors put at more than 45,000 GPUs.

The Analysis

1. Two questions, two levels of confidence. A Maharashtra farmer wants rain for the next three to ten days; an engineer wants to know how intense downpours will be in 2070. Science can answer the first far better than the second.

2. Models have a known bias. Climate models usually predict too little rain over the subcontinent and too much over the surrounding ocean, with seasonal totals off by tens of per cent in places. Most failed to reproduce the observed 5 to 10 per cent decline in monsoon rainfall between 1950 and 2000.

3. AI changes short-range forecasting. With over 40 years of observations, AI models learn how the atmosphere evolves from day to day. They beat conventional forecasts on many measures and may soon predict active and break phases. Rare extremes are thin in any one location’s data, but AI can transfer lessons learned elsewhere.

4. AI cannot see a climate that has not happened. A warmer atmosphere holds more moisture, so extremes intensify. In a world two or three degrees warmer, the authors write, a once-in-a-decade downpour will be 15 to 20 per cent more intense. No data exist to train AI for that, so physics-based models, perhaps hybrids with AI, remain necessary.

5. Frontier modelling is now within reach. AI and GPUs let universities and start-ups build regional models without national supercomputing centres. India has the talent, the computing and the longest observational record of the monsoon anywhere.

The precision that earns marks. The physical basis of the intensification claim is the Clausius-Clapeyron relation: saturated air holds about 7 per cent more water vapour per degree Celsius of warming. Name it in a Mains answer on extreme rainfall.

Data and Institutions Vault

Prelims-grade facts:

The monsoon and models:

• Global climate models use grid cells 10 to 100 km wide; monsoon convective clouds are a few km across.
• Models tend to under-predict rain over the Indian subcontinent and over-predict it over the surrounding ocean.
• Observed monsoon rainfall fell by about 5 to 10 per cent between 1950 and 2000, which most models failed to reproduce.
• Aerosols over the Indo-Gangetic Plain reflect sunlight, cooling the land and weakening the land-ocean contrast.
• Warm air holds about 7 per cent more moisture per 1 degree Celsius (Clausius-Clapeyron).

Institutions:

• IMD and IITM Pune function under the Ministry of Earth Sciences.
• Mission Mausam was approved by the Union Cabinet in September 2024.
• Ahmedabad’s Heat Action Plan (2013) is widely cited as the first in South Asia.

Prelims, the traps:

• Weather is short-term; climate is the long-term average; AI models are trained on weather data.
• Aerosols and greenhouse gases have opposite effects on monsoon strength.

⚠️ Watch the trap: The 15 to 20 per cent figure refers to more intense extreme downpours in a two to three degree warmer world, not to a rise in total seasonal rainfall.

The Debate

For the authors’ view. AI makes accurate short-range forecasting cheaper and local; physics models remain necessary for the long run; and infrastructure cannot wait for perfect projections.

The complication. AI forecasts are only as good as observations, and India’s radar and rain-gauge coverage is uneven. Hybrid models are still maturing, and the uncertainty on onset shifts and regional intensity is large.

The balanced verdict. Invest in both: dense observations plus AI for the coming seasons, and physics models plus adaptive design for the coming decades. The authors add a third priority: getting forecasts to the farmer, engineer and district planner in usable form.

How to Think About This

Separate the time scale before judging the tool. Whenever a technology claims to solve a prediction problem, ask what time horizon it serves and whether the future it predicts resembles its training data. Data-driven tools excel where the past is a good guide; they fail where conditions are new. That test applies equally to credit scoring, disease surveillance and demand forecasting.

Diagram-in-Words

Takeaway Box

• Authors: Tapio Schneider (Caltech) and Sandeep Juneja (Ashoka University), writing in the Hindustan Times.
• Why models struggle: clouds are smaller than model grids; competing influences; observed 1950-2000 decline missed by most models.
• AI: better, faster short-range forecasts, including active and break phases.
• Limits: no training data for a warmer climate; extremes 15 to 20 per cent more intense at 2 to 3 degrees of warming.
• Way forward: more observations, hybrid models, last-mile advisories, infrastructure with adjustable margins.

Sources: Hindustan Times, India Meteorological Department

Source: The Farmer and the Engineer: What AI Can and Cannot Tell India About the Monsoon — Ujiyari.com | Free UPSC & State PCS Editorial Analysis

Key Arguments at a Glance

Writing in the Hindustan Times, Tapio Schneider of Caltech and Sandeep Juneja of Ashoka University argue that AI trained on four decades of observations will soon give Indian farmers better short-range monsoon forecasts, but that planning for a warmer climate, where no training data exist, must still rely on physics-based models, and infrastructure should be built now for rainfall extremes 15 to 20 per cent more intense.

✓ Supporting

• Monsoon rain comes from convective clouds a few kilometres across, while global climate models use grid cells 10 to 100 km wide, so errors in representing clouds compound; models usually under-predict rain over the subcontinent and over-predict it over the surrounding ocean.
• Most models failed to reproduce the observed 5 to 10 per cent decline in monsoon rainfall between 1950 and 2000, partly because greenhouse gases and aerosols pull in opposite directions.
• AI weather models now beat conventional forecasts on many measures and run in minutes on GPUs, and may predict the active and break phases that farmers need for planting.
• In a world two or three degrees warmer, the authors write, a once-in-a-decade downpour will be 15 to 20 per cent more intense, a climate for which no training data exist.

⚠ Counter

AI forecasts depend on dense, high-quality observations that India still lacks in parts; rare extremes are thin in the training data; and physics models themselves remain uncertain about onset shifts and regional intensity.

→ Way Forward

More radars, rain gauges and ocean buoys; last-mile systems such as farmer advisories and heat action plans; hybrid AI-physics models; and drains and reservoirs sized for more intense extremes with margins that can be raised later.