Incompatible models, infra shortage: Why IMD struggles with the weather
A working paper shows that relatively poorer countries such as India do better than richer countries in at least the seasonal forecasts of temperature.
That the India Meteorological Department (IMD) is far from reliable in predicting rain is well-known. Why this inaccuracy exists is something more perplexing. A recent paper by two environmental economists suggests that there are both non-material and material factors behind IMD’s weather struggles: existing weather models find it hard to simulate tropical weather and the lack of comparable infrastructure with richer countries adds to this disadvantage.
A newer version of the working paper called “Global inequalities in weather forecasts” by environmental economists Manuel Linsenmeier and Jeffrey Schrader was uploaded on the online repository SocArXiv on November 25. The paper is under review and may undergo changes before a final version is published.
The paper shows that relatively poorer countries such as India do better than richer countries in at least the seasonal forecasts of temperature. The bad news is that poorer countries are worse off than their richer peers in short-term forecasts of both temperature and rain, which matter more for extreme weather events as well as public perception of weather forecasting.
The paper also studied forecasts different from those made by local agencies (more on this later). However, it is worth engaging with the paper beyond its headline findings to understand how forecasts can be improved in countries such as India.
First the bad news, which is the main finding of the paper. The researchers checked how much the time series of short-term forecasts of temperature anomalies (deviation from normal) – forecasts for the seven days ahead – was correlated with the anomalies observed at local weather stations. They found that the forecast for the seventh day in a high-income country was on average more accurate than the forecast for the first day in a low-income country. Simply speaking it means London can predict next Friday’s weather more accurately than Khartoum’s forecast of tomorrow’s weather, and Delhi’s forecast accuracy is somewhere in between. This analysis was also done for rainfall and yielded a similar result.
To be sure, India is a lower-middle income country and this group performed better than low-income countries in at least predicting rain, although it was still behind high-income countries. The inaccuracy of forecasts in poorer countries also does not mean that there have been no improvements over time. It is just that the forecasting gap between richer and poorer countries has not decreased in the 1985-2020 period the researchers studied.
What can be done to close the accuracy gap in forecasting between countries? There are some challenges that are harder to fix than others for individual countries. The harder challenge is that there is a large gap in accuracy between the tropics (the region between the Tropic of Cancer and Tropic of Capricorn) and extra-tropics (regions beyond the tropics). The accuracy gap by location is not surprising. The researchers were studying forecasts of the European Centre for Medium-Range Weather Forecasts (ECMWF) against observations from local weather stations, although its findings held even when using the forecasts of the Global Forecast System (GFS) of the US government’s National Oceanic and Atmospheric Administration (NOAA). This is because most national agencies adapt these or similar models to their local knowledge and needs. Unfortunately, these adaptations are sometimes needed because global weather models find it harder to simulate tropical weather. For example, the Bharat Forecast System (BFS), an adaptation of NOAA’s GFS that was launched by the IMD this year, is an attempt to fix some problems. That this was not enough can be gauged from the fact that more than a few extreme rainfall events this monsoon were not predicted accurately.
The researchers also quantified the contributions of various factors to the variation in accuracy. As expected from the difference in accuracy in tropical and extra-tropical regions, 54% of the variation could be attributed to just the latitude and longitude for which the forecast is made. However, 7% could be attributed to the weather observation infrastructure, which is the input that determines forecasts. Interestingly, if one looks at tropics and extra-tropics separately, the contribution of this infrastructure increases for both (to 12% in the tropics). This means that countries such as India can beat their lower middle-income peers by making their observation network denser, by adding more weather stations. Some improvements can also be made by increasing the frequency of observations, which is lower in poorer countries. A higher frequency of observation matters more when the infrastructure density is low.
To be sure, there is good news from the paper, too. Poorer countries outperform richer ones in seasonal forecasts (the period one to six months ahead). The paper suggests that this could be because of the difference in factors – such as observations from sea buoys, which are managed and funded by multiple countries, and satellites -- that determine the accuracy of seasonal forecasts. This fact, however, should not make India complacent. When the India Meteorological Department (IMD) issued its first forecast of the 2025 monsoon in April, HT reported that the accuracy of this first forecast is worse than predicting a coin toss in the 2001-2024 period.
