Skymet’s forecast for 102 per cent of the long period average (LPA) of the southwest monsoon was wrong. On September 30, the monsoon ended at 86 per cent of the LPA, leading to a second consecutive season with deficient rainfall (mild meteorological drought). Congratulations to the Indian Meteorological Department (IMD), which sounded out the country on below-normal rainfall at 93 per cent of the LPA and then downgraded it to 88 per cent.
Getting a monsoon forecast absolutely right is an impressive scientific achievement, one that should be celebrated.
Now, to why we failed. Skymet was successfully able to forecast the monsoon for the past three years using a hybrid methodology that combines statistics and dynamics. Our statistical method is El Nino southern oscillation (Enso)-based, which links the Indian southwest monsoon’s performance to two climactic phenomena — El Nino and La Nina. In the past 100 years, 80 per cent of all El Nino years have resulted in below normal monsoons, and 60 per cent in droughts. The opposite phenomenon in the Pacific Ocean is called La Nina, and is positively correlated with the monsoon.
We overlaid the Enso statistical method with the dynamical models, or numerical weather prediction. This breaks climate down into integral equations and runs it through supercomputers. Skymet has been using the climate forecasting system version two (CFSv2), run by the National Centre for Environmental Prediction at the National Oceanic and Atmospheric Administration in the US. We backtested the CFSv2 for 32 years and found that it is correct six out of 10 times. When you layer this with the Enso statistical method, the accuracy improves to 0.7.
The CFSv2 has previously shown that it is very sensitive to El Nino. It has successfully deciphered rainfall trends months in advance. It helped us accurately forecast below normal rainfall for 2012, normal rainfall for 2013 and deficient rainfall for 2014. It had caught these variations well in advance in those years. This year, the CFSv2 indicated normal to above normal till well into June. This was our first bias.
That assumption was strengthened by the historical context. First, back-to-back droughts are rare. In the past 144 years, two consecutive years of deficient rainfall have occurred only four times — 1904-05, 1965-66, 1985-86, 1986-87. Second, a continuing El Nino episode from the previous year is less likely to be a drought in the successive year. The present El Nino was formed in September 2014, and our assumption was that its impact would be less severe in 2015, its second year. Third, we believed the Indian Ocean Dipole (IOD) would insulate the monsoon from El Nino, like it did in 1997.
Till August 1, the monsoon was only 5 per cent less than the LPA. It seemed our assumptions were correct. But the 23 per cent deficit in August and the 26 per cent deficit in September sealed 2015’s fate. Even though the IOD became positive in September, it was too late for it to have a tangible counter-balancing effect.
We seem to have overapplied and overengineered our forecast this year. The monsoon is a beast we thought we knew intimately, but it seems to hold a few more secrets. It has also made us confront the fact that while our model has a 70 per cent chance of success, it also has a 30 per cent chance of failure. We have identified the gaps we need to fill, such as the need to model ocean dynamics more accurately.
In these columns, Ashok Gulati said that we should be “introspective about our forecasts”. I’d like to assure him that we are; this article comes from that place of reflection. Skymet is committed to decoding the Indian monsoon — to learning from the years we go wrong and the incorrect assumptions we might have made. In the meteorological scientific community, there are voices that call us disruptive and attention-seeking. I see things differently. Here’s why: Between 2000 and 2014, the IMD’s monsoon forecast mostly ranged within normal rainfall of the LPA.
There was only marginal variation from year to year, considering these 15 years saw four droughts — in 2002, 2004, 2009 and 2014. In 2002, against the forecast of normal rainfall — 102 per cent of the LPA — the season ended with a severe drought of 79 per cent of the LPA. This year marks a dramatic shift from that safe zone. In fact, this is the first time since it was established in 1875 that the IMD forecast for a deficient monsoon.
Although the El Nino-monsoon correlation has been known since 1980, it was never a driver of the IMD’s forecasts. As recently as March 24, 2014, a senior IMD official referred to El Nino as a Western conspiracy. El Nino fears help American and Australian interests to “drive down commodities and stocks in India”, this official had said. The IMD’s forecast this year was a success because it seems to have finally taken into consideration the strong El Nino correlation.
Its journey from El Nino sceptic to El Nino acolyte has been fast. The IMD’s ocean dynamics also seem to have massively improved over the past year. Its 2015 achievement is unqualified and absolute. But I believe media attention and competition too helped fast-forward its research, with spectacular results.
Forecasting the monsoon is a solvable problem that we are getting better at understanding. The attention, focus and critiquing of forecasts, as well as more competition among forecasting agencies, whether private or public, will help this journey.
The writer is founder and chief executive officer, Skymet.