The Forecast isn’t Always Fair: Barriers to Accurate Weather Prediction in the Developing World
- Dr. Hansi Singh
- 7 days ago
- 5 min read
Weather Forecasting: An Achievement of Modern Physics
Numerical Weather Prediction (NWP) is one of the triumphs of modern physics, mathematics, and computing. In a nutshell:
Weather models combine the physics of fluid motion, represented by a specialized form of the Navier-Stokes equations, along with mathematical formulations for convection, clouds, radiation, chemistry, and aerosols (e.g. particulates) to represent the different physical processes in the atmosphere.
All physics equations and formulations are discretized over hundreds of thousands of spatial grid cells, covering a given forecast domain, using advanced numerical methods that are accurate and stable. This numerical discretization enables us to solve these complex equations without needing to come up with an analytical solution (which is often impossible to do, anyway).
High performance parallel computing is used to solve these equations simultaneously over thousands of processors, allowing an ensemble of forecasts to be produced in hours.
Several times a day, numerical weather models come up with updated forecasts using this extensive machinery, initialized using the latest on-ground and satellite observations of atmospheric conditions. This complex, coordinated, resource-intensive technological marvel gives us the quality 10-day forecasts that we take for granted on our favorite weather app.
As we described in an earlier post, the Earth System Models we use for long-range weather forecasting (beyond 10 days) are a more extensive, complex version of weather models, involving even more physical formulations to represent the ocean, land surface, and sea ice. Not surprisingly, these Earth System Models also rely on a similar (albeit, larger in scale) combination of physics, mathematics, and computing to deliver state-of-the-art predictions.
The Quality of Weather Forecasts in the Developing World
But the same quality weather forecasts that we take for granted in North America or Europe can’t be taken for granted in other parts of the world. Take India, for example, a developing country that is technologically advanced where nearly 1 in 2 people have a smartphone. But weather predictions over India have always been notoriously unreliable, with the start of the monsoon season representing one of the most difficult forecast targets to be able to hit accurately. But it’s not only forecasts that are poor; nowcasts over the Indian subcontinent are also inaccurate. On a sunny day, for example, ‘current conditions’ might indicate rain. Understandably, inaccuracies in nowcasts severely limit trust in forecasts.
It’s not only India where this is a problem. Around the developing world, from much of Asia and Africa to many parts of South America, weather forecasts are much less accurate than the forecasts we take for granted in North America and Europe. This lack of quality forecasts is particularly troubling given that the scale and frequency of extreme weather events are increasing all over the world. For areas of the world with fewer resources for repair and rehabilitation, including large swaths of developing nations, forecasts are even more critical for anticipating severe weather and taking measures to avoid harm before the fact.
Poor weather information in the developing world arises from several factors. First, many areas in the developing world lack observational networks, both on-ground and in space. Such networks are essential not only to create accurate nowcasts, but also for initializing forecast models. On-ground Doppler weather radars, used for real-time precipitation monitoring, are expensive to purchase, operate, integrate, and service, requiring a developed and resourced regional meteorological service. While hundreds of such radars track real-time rain and snow over the US and Europe, there are only sixteen over the enormous African continent, outside of South Africa and Morocco. Even basic in situ meteorological stations, which can measure quantities like wind, temperature, humidity, and solar radiation, are expensive and require frequent, skilled maintenance for continued operation. And while many remote sensing satellites record observations over the globe, many others are geostationary, meaning they are focused over select regions of interest, predominantly North America and Europe.
Furthermore, existing operational weather forecasting systems tend to be designed and tuned for the regions that they serve. Over the continental US, for example, the National Oceanic and Atmospheric Administration (NOAA) operates several global forecasting systems, including the Global Forecast System (GFS; for 14-day weather predictions) and the Climate Forecast System (CFS; for subseasonal and seasonal weather forecasts). Both of these systems are global but are designed to serve the U.S. When these models are tuned for forecast performance and accuracy before being operationalized, the quality metrics used to optimize the tuning are predominantly those that impact weather over the U.S. and its territories, meaning that the forecasts are most accurate for those regions. Similarly, forecast models produced by the European Centre for Medium-Range Weather Forecasting (ECMWF), like the Integrated Forecasting System (IFS), are optimized for predictive performance over the European Union. On the other hand, most developing countries do not have meteorological agencies with their own in-house operational forecasting systems that are optimized for maximum predictive performance over their own nations.
Indeed, many developing countries lack the computing infrastructure and specialized personnel to tune and maintain forecasting workflows that are accurate for their own regions (let alone building such a system from scratch). Even if out-of-the-box forecasting systems, like NOAA’s GFS or ECMWF’s IFS, produce poor forecasts over a particular region, they can be bias-corrected and calibrated during post-processing to improve forecast quality. Regional forecasting models, like the Weather Research & Forecasting model (WRF) can have parameters tuned to produce quality forecasts over any region (such as WRF’s CONUS configuration over the continental U.S.). But such tuning or post-processing requires extensive high-performance computing resources with sufficient highly-trained personnel to utilize them. The lack of such computational resources and technical personnel in many developing countries severely limits forecast quality.
The Path Forward: Investing in Global Forecasting Infrastructure
The common thread in all these factors is a lack of resources: lack of observational systems, lack of trained personnel, and lack of high performance computing systems. Thankfully, this is slowly changing in some areas of the developing world that are becoming wealthier. India, for example, is in the process of developing its own national weather prediction infrastructure for quality extreme weather forecasting. The Bharat Forecasting System (BFS), released just last month, provides quality, high-resolution, 10-day forecasts by integrating on-ground Doppler radar and weather station observations with an advanced physics-based modeling framework run on state-of-the-art supercomputing infrastructure. Designed by the Indian Institute of Tropical Meteorology, BFS already boasts much better forecasting capabilities for extreme precipitation and monsoon onset compared to previous systems.
There is still much work to do to bring quality weather forecasts, both conventional and long-range, to different parts of the world. Quality forecasts are necessary for building resilience not just in the developed world, but also in developing countries where weather impacts may be even more dire. At Planette, we are mission-driven to build climate resilience through quality forecasts that can inform data-driven decision-making. We hope that this is a mission that governments, international agencies, and philanthropic organizations get behind in the coming years.
