DSS stops predictions as another pollution forecasting model down

A pollution analysis system to estimate the proportion of smoke from farm fires in Delhi’s air has stopped offering predictions because it was inaccurate — the latest in a long line of examples of how efforts to tackle the annual public health crisis have either fallen short or proven to be entirely ineffective.
According to officials at the Indian Institute of Tropical Meteorology (IITM), the decision support system (DSS) they operate will no longer be issuing forecasts on the estimated contribution of farm fires to the PM2.5 (particulate matter roughly 2.5 micron thick) pollution in Delhi but will only be sharing an estimation of the previous day’s levels recorded.
The forecasting project comes under the Union ministry of earth sciences and is run as a collaboration with IITM in Pune. Till recently, it gave an estimate of sources that contribute to PM2.5 levels, which denote the concentration of microscopic particles roughly 30 times smaller than diameter of a strand of human hair.
“The decision [to stop predictions] has been taken due to low accuracy in the model at present. The accuracy of 55% has a model error of plus or minus 15%, which means it may be as low as 40%. This is why we are no longer issuing forecasts for the expected contribution from stubble burning, but will only give the actual contribution from the previous day after accounting in the actual farm fires,” said a senior IITM official, asking not to be named.
Invisible to the eye, these particles can penetrate deep into the lungs and even enter the bloodstream and come from a variety of sources, the most common of which is smoke – whether from vehicle tailpipe, garbage or stubble burning.
As recently as Wednesday, the model predicted the PM2.5 contribution from farm fires to be around 11.1%. However, the actual contribution was 15.96%, the same website declared on Thursday. Forecast for Tuesday had shown an estimated contribution of 3.5%, but it stood at 7.98%.
However, forecasts for other local and NCR sources will continue to be shared, the official added.
In the past week, several other measures rolled out over the last few years were proven to be ineffective. Late on October 20, the ban on firecrackers – in place since 2018 – was widely violated as pops and bangs echoed across the city on the occasion of Karva Chauth.
And in the days leading to it, the EWS – also operated by IITM – kept making inaccurate predictions about changes in air quality, delaying the roll-out of the graded response action plan (Grap) mandates by the Commission for Air Quality Management (CAQM). Grap lays down a series of curbs on activities to reduce emissions.
Year after year, administrations at the Centre, Haryana, Punjab, Delhi and Uttar Pradesh governments have been unable to meaningfully end the practice of paddy stubble burning that farmers often adopt to save money while clearing their fields for the winter crop sowing.
According to the official quoted above, the estimates for stubble burning in DSS were based on historical farm fire data along with meteorological forecasts like wind direction. But a change in the trend led to inaccurate estimates, the official said. “We have seen a change in the trend of burning, as concentrated fires sometimes occurred in a short window. This makes day-to-day forecasting tricky,” the official added.
Forecasts related to stubble burning contribution were part of DSS’s attempt of identify the impact of different sources to Delhi’s air in the coming days, thus allowing agencies, including the CAQM to take air quality related measures or decisions in advance.“If a particular source’s contribution is expected to rise, both micro or macro level action, including restrictions can be planned,” said the IITM official, adding this data is shared with the MoES, CPCB, the Union environment ministry and CAQM among others.
Till October 23, Punjab recorded 1,638 farm fires, the lowest till this time of the year in the last five years, data from the Indian Agricultural Research Institute (IARI) shows. For Haryana, the count is 680, also the lowest in the last four years. The only anomaly has been Uttar Pradesh, with 808 farm fires recorded so far, which is the highest in the last four years. The previous highest till this point was 670 farm fires in 2020. But experts fear the situation could worsen, particularly in Punjab, in the days to follow.
When asked about the DSS move, Sunil Dahiya, founder and lead analyst at Envirocatalysts, an organisation which researches air pollution levels across the country, said that putting out inaccurate data will not help any government body. “Attempts should be made to update the model and share data once the forecasting accuracy increases,” he said, adding that clear, accurate data will only help decision making for different agencies working on air pollution. 
“It is important that such reviews are carried out and if forecast accuracy is low, say below 70%, the data should be removed and the model should be reworked,” he said.