Science · Weather · 1875–2026

Indian Monsoon Forecasting: IMD’s Journey and Forecast Misses

📅 Updated 19 October 2026🌧️ 2026 season: 87% of LPA🎯 Forecast 90% in May

Written & reviewed by Nazish Fatima · Updated October 19, 2026

In short

Indian monsoon forecasting explained: how IMD forecasts evolved since 1886, why 2002, 2004 and 2009 missed, and how the 2026 season ended at 87% of LPA.

Latest Story

EntertainmentWho Fired at Tiger Shroff?What's the story behind this?SocietyWhat Replaced MGNREGA in 2026?Why does this matter right now?SocietyWhat Actually Worked?Why does this matter right now?PoliticsWhy Seats Still Follow 1971What's really at stake here?BusinessFrom the 2G Row to 5GWhat's driving this market shift?

India has forecast its monsoon for 140 years, longer than almost any country has issued seasonal forecasts. Indian monsoon forecasting began with one man noticing that heavy Himalayan snow tended to precede a weak monsoon, and it now runs on coupled climate models, supercomputers and, since 2026, artificial intelligence. The record is not a smooth climb. The India Meteorological Department (IMD) missed severe droughts in 2002, 2004 and 2009, then rebuilt its system. This timeline traces that journey through the deficient 2026 monsoon, which IMD saw coming.

Quick Answers How It Works Timeline Pick a Year Hits and Misses Chart What % of LPA Means Why Forecasts Miss What Next FAQ

💡 Short Answer

IMD has issued seasonal monsoon forecasts since 4 June 1886. Statistical models failed to foresee the droughts of 2002 (forecast 101% of the Long Period Average, actual 81%), 2004 and 2009 (forecast 96%, actual about 77%). Since switching to a multi-model ensemble in 2021, its average all-India error has fallen from 7.5% of the LPA in 2017–20 to 2.28% in 2021–24. In 2026 it forecast a below-normal 90%, and the season ended at 87%.

⚡ Monsoon Forecasting: Quick Facts

First regular forecast4 June 1886, by H.F. Blanford

Long Period Average87 cm (1971–2020 base)

Biggest modern miss2002: forecast 101%, actual 81%

Forecast strategy since 2021Multi-model ensemble with MMCFS

Average error 2023–251.9% of LPA

2026 monsoon87% of LPA vs 90% forecast

⚡ Quick Answers — AI Overview Ready

Monsoon Forecasting: Key Questions

How accurate is IMD’s monsoon forecast?

For the all-India seasonal total, much more accurate than a decade ago. The government told Parliament the average error fell from 7.5% of the Long Period Average in 2017–20 to 2.28% in 2021–24, and was 1.9% in 2023–25. Regional and district forecasts are less precise.

Which monsoon forecasts went most wrong?

2002, when IMD forecast 101% and rain was about 81%; 2009, forecast at 96% against about 77%; and 2004, forecast near 100% against about 87%. All three were drought or deficient years that the statistical models of the time did not anticipate.

Did IMD predict the weak 2026 monsoon?

Yes. It forecast a below-normal 92% of the LPA in April and 90% at the end of May, as El Niño developed. The season ended on 30 September at 87%, the weakest since 2015, inside the May forecast’s ±4% range.

What changed in IMD’s forecasting?

It moved from Himalayan snow rules (1886) to statistical regression (1988), two-stage forecasts (2003), a dynamical model from the Monsoon Mission (2012 onward) and a multi-model ensemble (2021). AI-based forecasts of monsoon advance were added in May 2026.

📚 Key Takeaways

140 Years of Forecasts in Ten Points

The start: IMD was founded in 1875 and issued its first regular monsoon forecast on 4 June 1886.
The first predictor: Henry Blanford linked heavy Himalayan snow to weaker monsoons.
Walker’s legacy: the search for better predictors in 1904–24 uncovered the Southern Oscillation.
Numbers, not words: the 1988 power regression model gave a forecast as a percentage of the LPA.
The big misses: 2002, 2004 and 2009 each produced deficient rain after near-normal forecasts.
Dynamical models: the Monsoon Mission from 2012 built a coupled ocean-atmosphere model for India.
The 2021 reset: a multi-model ensemble cut average error from 7.5% to 2.28% of the LPA.
Three clean years: forecasts for 2023, 2024 and 2025 all fell inside their stated error range.
2026: IMD forecast a below-normal season; it ended at 87%, the weakest since 2015.
The limit: a correct national number can still hide floods in one state and drought in another.

How Monsoon Forecasting Works

A seasonal forecast is a single number for the whole country, not a prediction of rain where you live.

Each April IMD estimates how much rain India as a whole will get from June to September, expressed as a percentage of the Long Period Average, about 87 cm based on 1971–2020. It attaches a model error, usually ±4% or ±5%, and updates the forecast in the last week of May. A forecast of 106% does not mean every district will get 6% more rain than usual, or that the rain will come evenly through the season.

The seasonal forecast is one layer in a stack. Shorter forecasts answer different questions, and for a farmer deciding when to sow, or a city preparing for a cloudburst, they matter more.

Forecast type	Range	What it answers
Nowcast	Next 3 hours	Is a thunderstorm about to hit this district or city?
Short and medium range	Up to 7 days (heavy rain warnings up to 7 days, up from 5 in 2014)	Where will heavy rain fall this week?
Extended range	Up to 4 weeks	When will the monsoon reach my area? Is a dry spell coming?
Monthly	One month ahead	Will July be wetter or drier than normal?
Seasonal (long range)	June to September, issued in April and May	Will India’s monsoon as a whole be normal?

The Timeline: From the 1875 Founding to the 2026 Monsoon

Newest first. Tags mark forecasts that landed within range, forecast misses, changes of method and institutional milestones.

15 Oct
2026

The monsoon withdraws after a deficient season 2026

Southwest monsoon withdrew from the whole country on 15 OctoberNortheast monsoon outlook for South Peninsular India: normal (88–112% of LPA)

The withdrawal closed a season that ended well short of average. For Tamil Nadu, coastal Andhra Pradesh and south interior Karnataka, attention now turns to the October–December northeast monsoon, for which IMD issued a separate outlook on 1 October. Reservoir levels and rabi sowing in the south, which had a 24% shortfall in the main season, will depend heavily on it.

30 Sep
2026

2026 ends at 87% of the LPA: below normal, as forecast Within range

759.4 mm against a normal of about 868.6 mmWeakest monsoon since 2015Fourth lowest since 2001Regions: Northwest 94%, Central 97%, South Peninsula 76%, Northeast 74%

The season started with June rainfall about 40% below normal, one of the driest Junes on record, as El Niño took hold; NOAA declared El Niño conditions on 11 June. July recovered to around normal, but August and September were deficient again. The result, 87%, sat inside the May forecast’s range of 86–94% and on the lower edge of the April range. Unlike 2002 and 2009, IMD had warned of a poor year in advance, although it did not foresee how badly the south and north-east would fare, having expected the north-east to be near normal.

29 May
2026

The update cuts the forecast to 90% Method

90% of LPA ±4%Below normal over Central, South Peninsular and Northwest IndiaMonsoon Core Zone, where most rain-fed farming is, below normalJune outlook: below 92%

The second-stage forecast went further than April’s, putting the most likely outcome close to the deficient threshold of 90%. It also flagged rain-fed farming areas specifically, the kind of regional guidance IMD did not issue before its 2021 strategy change.

12 May
2026

AI enters operational monsoon forecasting Method

AI-enabled forecast of monsoon advance, issued every Wednesday up to four weeks ahead16 states, more than 3,000 sub-districts1 km rainfall forecast pilot for Uttar Pradesh, up to 10 days ahead

Launched by Earth Sciences Minister Jitendra Singh and built by IMD, IITM Pune and NCMRWF, the system blends AI models with extended-range and statistical forecasts and is sent to farmers through the agriculture ministry. At the launch the minister said India had around 50 Doppler weather radars, up from 16 or 17 a decade earlier, with another 50 planned.

13 Apr
2026

IMD calls a below-normal season in April 2026

92% of LPA ±5%MMCFS model signals El Niño developing during the monsoonPositive Indian Ocean Dipole expected late in the seasonSnow cover January–March slightly below normal

This was the test the 2002 and 2009 forecasts failed: anticipating a weak El Niño-affected monsoon months ahead. IMD’s April statement relied on its dynamical models, which showed the Pacific warming even while weak La Niña-like conditions were still present.

12 Feb
2026

Government reports a 1.9% average error for 2023–25 Within range

Rajya Sabha reply by Jitendra Singh2023: 96 forecast, 95 actual2024: 106 and 1082025: 106 and 108All three inside the ±4% model error

The reply also listed why forecasts still go wrong: the evolution of ENSO, the Indian Ocean Dipole, large swings within the season and models’ difficulty simulating monsoon low-pressure systems over central India. On IMD’s 151st foundation day on 15 January 2026, the minister had said seasonal and monthly forecast errors had fallen from about 7.5% to about 2.5%, and announced 50 automatic weather stations each for Delhi, Mumbai, Chennai and Pune.

2025

A wet year, forecast correctly Within range

Forecast 105% (15 April) and 106% (27 May), ±4%Actual 108%Bharat Forecast System launched 26 May 2025AI onset forecasts sent to about 3.9 crore farmers

The Bharat Forecast System, developed at IITM Pune, runs a global model at a 6 km grid on the Arka supercomputer. Separately, the agriculture ministry, working with the University of Chicago, texted local forecasts of monsoon onset to 3,88,45,214 farmers in 13 states, based on a blend of AI weather models including Google’s NeuralGCM. A follow-up survey found that 31–52% of farmers changed planting decisions.

2024

Mission Mausam and a close forecast Within range

Forecast 106% ±4% in April and MayActual 108%Mission Mausam approved 11 September 2024, 2,000 crore rupees over two yearsArka (11.77 petaflops) commissioned September 2024

A La Niña-leaning season produced above-normal rain, as forecast. Mission Mausam funds new radars, satellites and supercomputers and better Earth-system models. By 2025, IMD’s rainfall monitoring stations had grown to 6,727 from 3,980 in 2015, and its high-performance computing capacity to about 22 petaflops.

An IMD Doppler weather radar on Kailasagiri hill in Visakhapatnam. India had about 15 such radars in 2014 and around 50 by 2026, with about 50 more planned under Mission Mausam. Photo: IM3847, CC BY-SA 4.0, via Wikimedia Commons.

2023

An El Niño year handled well Within range

Forecast 96% ±4%Actual 94.4%August 2023 was the driest August since records began in 1901

A seasonal total close to the forecast hid a dramatic month: August rainfall was about 36% below normal before a wetter September pulled the season back. It is a good example of why IMD now publishes monthly forecasts as well as a seasonal number.

2022

Third year of the new strategy Within range

Forecast 99% (April), raised to 103% ±4% on 31 MayActual about 106.5%

The late-May upgrade moved the forecast closer to the outcome. Heavy rain in central and southern India offset deficits in the Gangetic plains, another case of a national figure that masked regional extremes.

2021

IMD switches to a multi-model ensemble Method

New strategy adopted for the 2021 seasonMonthly forecasts addedForecast 101% (June update), actual 99%

Instead of relying mainly on statistical regression, IMD now combines coupled global climate models from several prediction centres, including its own MMCFS, and gives probabilities for each rainfall category and maps of where rain is likely to be above or below normal. Average absolute error fell from 7.5% of the LPA in 2017–20 to 2.28% in 2021–24.

2017–
2020

Four years, three big errors Forecast miss

2017: forecast 98, actual 952018: 97 and 912019: 96 and about 1102020: 102 and 109Average error 7.5% of LPA

The misses ran in both directions. In 2019, a positive Indian Ocean Dipole helped deliver the wettest season in 25 years after IMD had forecast a normal one. These years were the immediate case for the 2021 overhaul.

2012–
2017

The Monsoon Mission builds a dynamical model Institution

Ministry of Earth Sciences mission, phase one 2012–2017Coupled ocean-atmosphere Monsoon Mission Climate Forecast System (MMCFS)Supercomputing capacity expanded

The mission partnered with the US National Centers for Environmental Prediction to adapt its Climate Forecast System to Indian conditions. Dynamical models simulate the physics of ocean and atmosphere rather than relying on past correlations, which is what allows them to see an El Niño developing in a way statistical predictors could not in 2002 or 2009.

2009

Another drought the forecast missed Forecast miss

Forecast 96% ±5% (17 April), cut to 93% ±4% (24 June)Actual about 77%Error against April forecast about 19 points

Even the late-June update, issued after the season had begun badly, was well above the outcome. The government later told Parliament that El Niño had had a stronger impact than the models expected. Together with 2002 and 2004, the 2009 failure discredited purely statistical seasonal forecasting.

2007

A statistical ensemble system Method

New statistical forecasting system based on an ensemble techniqueIntroduced after the 2002 and 2004 misses

Instead of one regression equation, the system ran many statistical models with different combinations of predictors and averaged them. It reduced dependence on any single relationship, but as 2009 showed, all the models still drew on the same kind of historical correlations.

2004

A second miss in three years Forecast miss

Forecast about 100%Actual about 87%Weak July again the main cause

A paper in IMD’s journal Mausam by R. Jenamani and colleagues, ‘Another deficient monsoon 2004: a comparison with drought year 2002 and possible causes’, examined why. Both seasons had normal-looking forecasts and both broke down in July. The repetition showed that 2002 had not been a one-off.

2003

Two-stage forecasts begin Method

First-stage forecast in April, update before the season8-parameter power regression and linear discriminant analysis models, 2003–2006

The two-stage format, an April forecast followed by an update with more recent data, survives today, with the update now issued in the last week of May. The 16-parameter model was replaced by smaller models after several of its predictors lost their link with monsoon rainfall.

2002

The forecast that missed a severe drought Forecast miss

Forecast 101%Actual about 81%July rainfall roughly half of normalError about 20 points

IMD’s history describes how rapid warming of sea-surface temperatures in the central equatorial Pacific after the forecast was issued suppressed the monsoon, and that no group in India or abroad anticipated it. Kharif crops failed across large parts of the country. The episode triggered the review that produced the 2003 changes.

1988

The 16-parameter power regression model Method

Introduced by IMD in 1988Uses 16 predictors, from Pacific pressure to Eurasian snowGives a quantitative forecast as % of LPA

Associated with the scientist Vasant Gowariker, the model replaced qualitative forecasts with a number that could be checked. Between 1988 and 2001 it delivered a run of reasonable forecasts, which helped build the confidence that 2002 then shattered.

The Simla Office building in Pune, where IMD moved its headquarters in 1928 and where its climate research and long-range forecasting group still works. Photo: Phadke09, CC BY-SA 4.0, via Wikimedia Commons.

1904–
1924

Gilbert Walker and the Southern Oscillation Institution

Walker heads IMD as Director General of ObservatoriesSeeks objective predictors from worldwide data1922: forecasts for three regions of India

Working with hand calculations, Walker’s team searched global records for anything that correlated with the monsoon. Among the patterns they found was the Southern Oscillation, a see-saw of air pressure between the Pacific and Indian Oceans, now known to be linked with El Niño. His regional forecasts ran until the North-East India forecast was dropped in 1935.

Sir Gilbert Walker, Director General of Observatories in India from 1904 to 1924, whose search for monsoon predictors led him to the Southern Oscillation. Photo: unknown author, 1925, public domain, via Wikimedia Commons.

1892–
1893

Forecasts widen Method

1892: forecast for the second half of the monsoon seasonDecember 1893: first forecast of winter rainfall

Within a few years of the first forecast, IMD had added mid-season and winter outlooks, the beginning of a forecast calendar that now runs from April to September.

4 Jun
1886

The first regular monsoon forecast Method

Issued by Henry Francis BlanfordBased on Himalayan snowfall and other observed conditions

India became one of the first countries to issue a regular long-range seasonal forecast. Blanford’s method rested on one empirical link: more snow in the Himalayas in winter and spring tended to mean a weaker monsoon. Versions of that snow predictor were still in IMD’s models more than a century later.

Henry Francis Blanford, the first Meteorological Reporter to the Government of India, who issued tentative monsoon forecasts from 1882 and the first regular forecast on 4 June 1886. Photo: India Meteorological Department, public domain, via Wikimedia Commons.

1882–
1885

Tentative forecasts from Himalayan snow Method

Blanford tests the snow–monsoon linkForecasts not yet issued regularly

After the famine of 1877, the government wanted warning of a failing monsoon. Blanford’s early trials gave enough promise for the department to commit to a regular forecast.

1875

The India Meteorological Department is founded Institution

Foundation day 15 JanuaryHenry Francis Blanford appointed Meteorological ReporterFollowed the 1864 Calcutta cyclone and famines in 1866 and 1871

Before 1875, weather observation in India was handled by separate provincial services with no common standards. A central department made it possible to collect comparable data across the country, the foundation for every forecast that followed. IMD moved from Calcutta to Shimla, then to Pune in 1928 and finally to New Delhi.

The Scorecard: Forecast vs Actual

Pick a year to see how close IMD got, then compare the record before and after 2021.

🌧️ Interactive: Pick a Monsoon Year

Choose a year to compare IMD’s seasonal forecast with what actually fell.

Forecast (% of LPA)–

Actual–

Error–

–

The forecast shown is the latest one issued before the season (the late-May or June update where there was one). Actual figures are IMD’s all-India June–September totals.

Before 2021, actual rainfall often landed outside the forecast range, in dry years (2002, 2004, 2009) and wet ones (2019, 2020). Since the multi-model ensemble was adopted, every season has landed inside it, including the deficient 2026 monsoon.

Year	Forecast (% of LPA)	Actual	Verdict
2002	101	about 81	Major miss; severe drought
2004	about 100	about 87	Miss; deficient season
2009	96 (April), 93 (June)	about 77	Major miss; drought
2018	97	91	Miss
2019	96	about 110	Miss; wet season underestimated
2020	102	109	Miss
2021	101	99	Within range
2022	99 (April), 103 (May)	about 106.5	Within range
2023	96	94.4	Within range
2024	106	108	Within range
2025	105 (April), 106 (May)	108	Within range
2026	92 (April), 90 (May)	87	Within range; deficient season called in advance

Values are rounded all-India figures. The 2009 April forecast of 96% is used for the headline error; the June update was 93%.

How Much Has Accuracy Improved?

Period	Average absolute error	What changed
2017–2020	7.5% of LPA	Mainly statistical system; three large misses
2021–2024	2.28% of LPA	Multi-model ensemble with MMCFS, monthly forecasts
2023–2025	1.9% of LPA	Every year inside the ±4% range
2026	3 points (May forecast)	Deficient year forecast as below normal

These figures measure one thing: the error in the all-India June–September total. They are the government’s own numbers, given in replies to Parliament in December 2025 and February 2026. They do not show that every monthly, regional or heavy-rain forecast improved by the same margin, and a four-year average can be moved a lot by a single year, as 2019 showed.

What Does ‘% of LPA’ Mean?

📏 Interactive: What Does the Percentage Mean?

Slide to any all-India rainfall figure to see IMD’s category and what it means in millimetres.

Rainfall–

IMD category–

Seasons in this band–

LPA = 868.6 mm, the 1971–2020 June–September average for India as a whole.

Why Monsoon Forecasts Still Miss

The drivers IMD itself lists when explaining errors.

Pacific Ocean

El Niño can arrive late or strong

El Niño years tend to be drier, but timing and strength matter. In 2002 the warming came after the forecast; in 2009 its impact was stronger than expected. In 2026 the models saw it coming in April.

Indian Ocean

The Dipole can cancel El Niño

A positive Indian Ocean Dipole can boost rain even in an El Niño year. It helped make 2019 an excess season after a normal forecast.

Within the season

Breaks and bursts

A single dry month can sink a season, as July did in 2002 and 2004 and August did in 2023. Seasonal models struggle to predict these swings months ahead.

Weather systems

Low-pressure systems are hard to model

Monsoon depressions bring much of central India’s rain. The government’s February 2026 reply said models’ difficulty simulating them reduces early accuracy there.

Geography

One number, many monsoons

In 2026 Central India got 97% of its average and the south 76%. A national forecast cannot capture that spread.

Climate change

A shifting baseline

Heavier downpours with longer dry gaps between them make the distribution of rain harder to predict, even when totals are right.

What IMD’s Journey Teaches

Judge a forecast by the question it answers. An accurate seasonal number can sit alongside a district flood or a regional drought. In 2026 the national forecast was right while the north-east, expected to be near normal, ended 26% short.

Correlations age; physics travels better. The 1988 model worked for a decade and then broke down as its predictors lost their link with the monsoon. Dynamical models that simulate the ocean and atmosphere were what finally allowed IMD to see an El Niño year in advance.

Uncertainty is part of the forecast. Since 2021 IMD has published probabilities for each rainfall category alongside the headline number, which is more useful to a reservoir manager than a single percentage.

What Comes Next

The next frontier is resolution and lead time rather than the all-India total. The Bharat Forecast System runs at 6 km, the Uttar Pradesh pilot at 1 km, and the AI monsoon-advance forecast reaches sub-district level four weeks ahead. Mission Mausam is funding more radars and automatic weather stations to feed them.

AI models are now part of the operational mix, but IMD uses them alongside physical models, and their record over a full range of monsoon years is still short. The real test is whether these products give farmers, cities and disaster agencies usable warning of dry spells and extreme rain, not just a better national number.

Explore More Timelines

PoliticsHow Did They Get Inside?What's really at stake here?TechnologyFrom Ethics Pledges to Binding AI LawWhere is this technology headed?ScienceThe Tiger Count Meter: Pick a CensusWhat does the evidence show?ScienceShe Helped Build India's Science. Name Her?What does the evidence show?PoliticsUS Secondary Sanctions: A Timeline of Financial WeaponryWhat's really at stake here?

Frequently Asked Questions

When did India start forecasting the monsoon?

The India Meteorological Department issued the first of its regular seasonal monsoon forecasts on 4 June 1886. Henry Francis Blanford had issued tentative forecasts from 1882 to 1885, using the observation that heavy winter and spring snow in the Himalayas tended to precede weaker monsoons over the plains.

When was the India Meteorological Department founded?

IMD was set up in 1875 and celebrates 15 January as its foundation day; it marked its 150th anniversary in January 2025. It brought scattered provincial weather services under one central department after the 1864 Calcutta cyclone and famines linked to monsoon failure in 1866 and 1871.

What is the Long Period Average (LPA) of monsoon rainfall?

The LPA is the average all-India rainfall for June to September over a fixed base period. IMD currently uses 1971 to 2020, which gives about 87 cm (868.6 mm). Forecasts and results are expressed as a percentage of this figure, so 100% means rainfall exactly at the long-term average.

How does IMD classify a monsoon as normal or deficient?

For the country as a whole IMD uses five categories: deficient below 90% of the LPA, below normal from 90 to 95%, normal from 96 to 104%, above normal from 105 to 110%, and excess above 110%. A drought year for the country is usually one in which rainfall is deficient and a large area is affected.

What does a model error of plus or minus 4% mean?

It is the uncertainty band IMD attaches to its quantitative forecast. A forecast of 106% with a model error of 4% means IMD expects rainfall between 102% and 110% of the LPA. IMD treats the forecast as accurate when the actual figure falls inside that band.

Why did the 2002 monsoon forecast fail?

IMD forecast 101% of the LPA, but the season ended at about 81%, one of the worst droughts on record. July rainfall collapsed to roughly half of normal. IMD’s own history links the failure to rapid warming in the central equatorial Pacific that developed after the forecast was issued and that no forecasting group, in India or abroad, anticipated.

What went wrong with the 2004 forecast?

IMD again forecast a normal season of about 100% of the LPA, and actual rainfall was about 87%. As in 2002, a weak July did most of the damage. A study in IMD’s journal Mausam compared the two years and found that the predictors used by the statistical models did not capture the conditions that suppressed rainfall.

How wrong was the 2009 monsoon forecast?

IMD forecast 96% in April and cut it to 93% on 24 June 2009, but the season ended at about 77% of the LPA, far below the lower limit of the forecast range. The error against the April forecast was about 19 percentage points. IMD later said El Niño had a stronger impact than its models expected.

Has IMD’s monsoon forecasting become more accurate?

On the official all-India measure, yes. The government told Parliament that the average absolute error of the seasonal forecast was 7.5% of the LPA in 2017 to 2020 and 2.28% in 2021 to 2024, after IMD adopted a multi-model ensemble strategy. For 2023 to 2025 the average error was 1.9%, with every year inside the forecast range.

What is the multi-model ensemble (MME) forecast?

Since 2021 IMD has combined output from several coupled ocean-atmosphere climate models run by global prediction centres, including its own Monsoon Mission Climate Forecast System, with its statistical models. Averaging many models reduces the influence of any single model’s bias and also gives probabilities for below-normal, normal and above-normal rainfall.

What is the Monsoon Mission Climate Forecast System?

MMCFS is a coupled ocean-atmosphere model developed under the Ministry of Earth Sciences’ Monsoon Mission, whose first phase ran from 2012 to 2017 and second from 2017. It is adapted from the US Climate Forecast System, and IMD uses it in the dynamical part of its seasonal forecasts and to anticipate El Niño.

How accurate was the 2026 monsoon forecast?

IMD forecast 92% of the LPA in April 2026 and 90% at the end of May, both below normal. The season ended at 87%, the weakest since 2015. That was inside the updated forecast’s range of 86 to 94% and at the edge of the April range of 87 to 97%, so IMD correctly called a poor year.

Why was the 2026 monsoon deficient?

El Niño developed during the season; the US agency NOAA declared El Niño conditions on 11 June 2026. June rainfall was about 40% below normal. July recovered to around normal, but August and September were again deficient, and South Peninsular and Northeast India ended near 76% and 74% of their averages.

Which forecast misses since 2017 pushed up the error?

The 2017 to 2020 period averaged a 7.5% error mainly because of three years. In 2018 IMD forecast 97% and rainfall was 91%. In 2019 it forecast 96% and rainfall reached about 110%. In 2020 it forecast 102% and rainfall was 109%. The last two were big underestimates of a wet season.

Why are monsoon forecasts so difficult?

The monsoon depends on interacting drivers: El Niño and La Niña in the Pacific, the Indian Ocean Dipole, Eurasian snow cover, land temperatures and short-lived weather systems such as monsoon depressions. The government’s February 2026 reply to Parliament named all of these, and said models still struggle to simulate low-pressure systems over central India.

Does El Niño always mean a bad monsoon?

No. El Niño years are more likely to bring below-normal rain, but the link is not fixed. A positive Indian Ocean Dipole can offset El Niño, as in 1997 and 2019, while some droughts happened without a strong El Niño. That is why IMD now forecasts using many climate models instead of a single relationship.

What role does Himalayan snow play in monsoon forecasts?

It was the first predictor ever used. Blanford noticed in the 1880s that heavy snow over the Himalayas tended to be followed by weaker monsoons. IMD still tracks Northern Hemisphere and Eurasian snow cover from January to March, which has a general inverse relationship with the following monsoon, as its April 2026 forecast noted.

Who was Sir Gilbert Walker?

Walker was Director General of Observatories in India from 1904 to 1924. Searching for better monsoon predictors, he identified the Southern Oscillation, the see-saw in air pressure across the Pacific and Indian Oceans that is now understood as the atmospheric half of El Niño. In 1922 he began issuing forecasts for separate regions of India.

What is Mission Mausam?

Mission Mausam is a programme approved by the Union Cabinet on 11 September 2024 with an outlay of 2,000 crore rupees over two years. It funds more radars, satellites, automatic weather stations and supercomputers, and better Earth-system models, with the aim of making India weather-ready and climate-smart.

What is the Bharat Forecast System?

Launched on 26 May 2025 and developed by IITM Pune, the Bharat Forecast System is a global weather model running at a 6 km grid, about twice the detail of the older 12 km model. It runs on the Arka supercomputer, which has 11.77 petaflops of capacity, and is used for short- and medium-range rainfall forecasts.

Is IMD using artificial intelligence for the monsoon?

Yes, alongside physical models. On 12 May 2026 the government launched an AI-enabled forecast of monsoon advance, issued every Wednesday up to four weeks ahead for more than 3,000 sub-districts in 16 states, and a 1 km rainfall forecast pilot for Uttar Pradesh. In 2025, AI-based onset forecasts were sent by SMS to about 3.9 crore farmers.

Does a correct seasonal forecast mean every state got the right rain?

No. The seasonal figure is an average for the whole country. In 2026 Central India received about 97% of its average while South Peninsular India got 76%. A season can be near normal nationally while one region floods and another is in drought, so IMD also issues regional, monthly and district forecasts.

Conclusion: From a Snow Rule to Seeing El Niño Coming

India’s monsoon forecast began in 1886 as a single empirical rule and became, by 2026, an ensemble of climate models, satellites, radars and AI. The failures of 2002, 2004 and 2009 were the turning point: they showed that correlations alone could not see a drought coming, and they led to the Monsoon Mission and the 2021 overhaul.

The 2026 season was the clearest test yet. IMD forecast a below-normal monsoon in April, and it was right. The challenge now is to be as right about where and when the rain falls as about how much falls across India.

Sources & further reading

Every dated entry above was checked against these references. Last reviewed 19 October 2026.