Weather model accuracy and data shortages

Global #precipitation products leave gaps and biases, especially over mountains and at high rain rates. Check this new paper in Communications Earth & Environment where we introduce an open-access framework that merges multi-source observations using regional-scale intelligent optimization and explicit #topographic factors within an end-to-end neural pipeline. The approach reconstructs missing values and corrects biases in global, time-varying precipitation fields, yielding stronger correlations and reduced errors versus standard satellite estimates. Results suggest immediate value for #storm monitoring and #flood forecasting in data-sparse, complex terrain. Paper: Communications Earth & Environment (2025), doi:10.1038/s43247-025-02624-3. https://lnkd.in/dKp6NsyY Hohai University Consiglio Nazionale delle Ricerche

I've recently had a fascinating conversation about the difficulties of real-time weather forcasting. With the rise of AI weather forecasts, one might think real-time is solved easily! But ironically, many of the problems around real-time forecasting in classical NWP persist in data-driven weather forecasts. If you've ever worked in physical simulation, you know, those problems usually need data on big regular grid. Our weather observations don't usually come in this format, between planes, satellites, and weather balloons, you can directly see these are as diverse as the weather itself. So even before one can run a weather forecast, a massive amount of work, expertise, and compute goes into creating a best guess of what the current weather looks like. This takes time, but it gets worse! These systems work in a way as a buffer against unforeseen circumstances! In 2020, suddenly most planes were grounded, which is a huge source of weather data! If anything, AI needs data, so how could we even make an AI system resilient to this type of catastrophe? You don't want your real-time system to just tell you "sorry, I know you rely on me, but I can't find enough planes in the area!" Or even worse, we live in a world with inequities, where manz areas in Europe and North America have a great coverage of diverse sensors for any kind of data. Rural and developing areas in the global South, however, are currently building up infrastructure and don't always have a good coverage of weather sensors or the money and people to for example send out weather balloons. Currently a lot of physical knowledge, expertise, and statistical system come together as a moat against these two problems of both temporal and spatial inconsistencies. But there are multiple fascinating developments in this sector: 1. Continuous data assimilation, where classical systems are updated, while new data comes in. 2. Using AI systems in this data assimilation. 3. Using observations directly in AI Can you imagine though? Not only longer acurate weather predictions, but you know exactly, if it's worth it to fire up the grill, because you know exactly if your garden is being watered in an hour. I'm seeing more and more people work on these types of problems and incrementally crack some problems in this space that were previously close to impossible or very expensive to solve. What a fascinating time we live in. And like an audience member said after one of my talks "it's nice to see the AI can be used for the benefit of society as well". I'm sure there's more, but what an interesting topic to explore!

This week, a series of continued powerful atmospheric rivers could fall anywhere from western Canada to Southern California, as strong jet stream passes over unusually warm waters in the Pacific, via El Niño. The landfall is poised to bring heavy rain, snow, and high winds to a large portion of the West. However, in 2024, the actual path and intensity of this weather is far from certain. How can that be? The answer is that we don’t have complete enough environmental data to predict weather precisely. This means that cities and communities don’t know if they’ll be affected, or to what degree. How do we fix this? It turns out we need data from every part of the Earth’s system, from the ground, to the atmosphere, and cloud tops, to form the most accurate forecasts. Today, the world’s biggest missing data set is the atmosphere, the vast majority of which is rarely or never observed, especially over oceans, where many major events like atmospheric rivers (ARs) develop. This is why we’re laser-focused at WindBorne on expanding our constellation of smart, long-duration balloons to completely unlock our intelligence into the atmosphere. Our balloon network travels autonomously all around the globe, directly gathering this critical data in real-time. To better predict ARs, we’ve been partnering with the Scripps Institution of Oceanography to gather data across the Pacific through the entire 2023-2024 AR season, from September through March. You can check out a snapshot below of a few of our missions, launched from South Korea and Palo Alto as part of this campaign. These observations will help us understand complex interactions between the oceans and atmosphere. They’ll ultimately produce insights that help us predict and mitigate the effects of climate change on a global scale. We’ll keep you posted on our progress. In the meantime, here’s the latest 12hr AR forecast from ECMWF:

🌩️ We Are Revolutionizing Weather AI 🚀 AI weather models are only as good as the data they train on. With Tomorrow.io’s new global Microwave Sounder satellite data (beautiful data sample attached), we’re reshaping AI-driven weather forecasting through better training, real-time inference, and reinforcement learning. Here's how: ✅ Training: Multi-altitude data (temperature, humidity, and more) helps models learn complex atmospheric interactions, improving predictions for rain, storms, and heatwaves. ✅ Inference: Real-time satellite streams provide live, high-fidelity data—closing gaps where ground-based observations are missing, ensuring more accurate forecasts globally. ✅ Reinforcement Learning: Continuous satellite feedback allows models to adapt and self-correct. AI can "learn" from past mistakes, refining forecasts for events like floods or heavy rainfall. 🌍 From emergency response and agriculture to aviation and renewable energy, this is the next leap in weather prediction.

Each week brings another extreme weather event - but behind the headlines is a quieter, systemic crisis: a global shortage of high-frequency, high-quality atmospheric observations to power accurate forecasts where they’re needed most. Forecast accuracy starts with assimilating the right data into numerical weather prediction models - a process often misunderstood, but foundational to effective early warning systems worldwide. Hear from Ryan Honeyager, Tomorrow.io’s Senior Data Assimilation Scientist, on how this process works and why Tomorrow.io’s satellite-derived microwave data is uniquely positioned to close critical observational gaps for national meteorological agencies and resilience teams across the globe. Watch now: https://okt.to/dGZq6j #DataAssimilation #SatelliteData #ForecastAccuracy #DaaS #WeatherResilience #GlobalForecasting #EarlyWarningSystems