Understanding the Latest ECMWF Forecast Model Upgrade and What It Means for Weather Prediction

Author: Chris Nitsopoulos

Posted on: Wednesday, November 13th, 2024, 10:43:57 AM

Today, while all of us were keenly watching for storms, The ECMWF (European Centre for Medium-Range Weather Forecasts) implemented an upgrade to its Integrated Forecasting System, known as Cycle 49r1. This is a significant update that aims to improve the accuracy of weather forecasts, particularly for near-surface conditions such as temperature and wind. Here's a breakdown of the changes and how they could impact weather forecasting. This model update is live now on various weather sites and of course our own Weather Centre

What Has Changed?

• Better Temperature and Wind Forecasts Near the Surface
  One of the biggest improvements in this upgrade is in near-surface temperature and wind predictions, especially beneficial during winter months with large frontal systems and long drawn out sub tropical ridges. By including more data from weather stations and adjusting how this data is processed, ECMWF has improved short-range forecasts of 2-metre temperature and 10-metre wind speeds. This means more accurate forecasts for conditions that affect us daily, like cold snaps or windy days. The model also 'understands' that inner city areas are usually hotter than areas around them which also assists it in improving temperature forecasts for large cities and urbvan areas.

• New System for Predicting Uncertainties
  ECMWF has introduced a new method to handle uncertainties in weather forecasts by using a technique called Stochastically Perturbed Parametrizations (SPP) - check the explanation of what this is below. This approach better captures the unpredictable aspects of weather, especially in the lowest layers of the atmosphere. As a result, medium-range temperature and wind predictions are expected to be more reliable, helping forecasters provide better guidance.

So what is this new system that deals with unpredictable weather? let's explain it in words that the everyday person can understand

Those of you who know me, know I love rugby (both codes) so let me explain it the best way I can using a sport/s that I love. Predicting the weather is a bit like being a rugby coach and trying to predict what will happen in a rugby match—it’s tough because there are so many moving parts, as a coach you can't control what your players do on the field, you certainly can't control your opponents and small changes/errors from the players on the field or errors in officiating can really affect the final outcome.

To handle this, weather scientists use something called SPP (Stochastically Perturbed Parameters). Here’s how it works, using rugby as an example:

Imagine SPP as a "Changeable Game Plan"

Picture this: you're the coach, and you’re trying to figure out the best game plan for tomorrow’s rugby match. But there are some things you can’t be sure about—like whether the weather will be wet and muddy, or whether your star player will be at 100% health, will the opposition use a different player in a key poisition. what if the ref makes a crucial mistake? So, instead of relying on one strict plan, you try out a few different ones to see what might work best and most importantly so you can adjust the game plan on the run.

In weather forecasting, SPP does something similar. Scientists aren’t always sure about certain details in the atmosphere (like exactly how hot or humid or windy it’ll be), so they test different “game plans” for the weather model by adjusting these details. This helps them see a range of possible weather outcomes.

Why SPP is Smarter Than Older Methods

In the distant past, scientists only tweaked the starting conditions, like setting a single “game plan” for the weather for example what is the final game result if I start with a short kickoff vs a long kickoff or what if my star player is hurt at the kickoff what happens to the result of the game? As you can imagine this wasn't a really great way of dealing with things because if we got it wrong at the start we can basically throw the game plan out the window and wait until the next game to fix it (or in this case until the next model run). 

Then in the last few years scientists at the ECMWF and other agencies developed 

• SPPT (Stochastically Perturbed Parametrization Tendencies): This method perturbs (disturbs) the outputs of specific weather model processes. Imagine it like tweaking the final score in our rugby analogy—SPPT makes small adjustments to the outcome of different weather processes after they’ve run, like cloud formation or rainfall. However, this approach doesn’t address whether the initial “game plan” or parameters that drive these processes are exactly right; it just adjusts the final results. So this was a better method than the early methods where if we got the game plan wrong at the start we threw our hands up in the air and gave up until we got another chance 12 hours later to get it right, but it still looked at the end product (the 'Try' if you like) without really understanding why or how we got into the position to score that 'Try' or get that rainfall. 

• SPP: This newer method instead adds randomness to the parameters or inputs driving the model processes. Rather than only changing the final “score” of each process, SPP tweaks the initial “game plan” as it goes. So, while the forecast is running, SPP adds small, random changes to parameters like humidity or wind, or temperature thresholds, which allows the model to simulate a wider range of realistic outcomes and understand why it's moving in the direction it is. i.e. it focuses on the process and the journey of winning the game/scoring the try (predicting the weather) without focusing on the win or the try itself. That old analogy if you work hard (high amounts of computational power in this case) and trust the processes you have in place the results will come but there might be setbacks and random problems on the way and we have to deal with those as they come and adjust what we do as a team because of them. That's the same analogy used here in SPP forecasting techniques. We no longer focus on scoring a try and winning the game (predicting the weather outcome), but the processes we have to go through to get there and we're ready to tweak those processes slightly and randomly as we need to.

The Result? A More Realistic Prediction of the Weather Game

By running multiple “game plans” with SPP, scientists (the rugby coach) can figure out the most likely weather outcome. This gives them a better idea of all the different ways the “match” could go, even when things in the atmosphere are hard to predict.

So, when you see a weather forecast that gives a range (like a 30-50% chance of rain), that’s because SPP and other tools are helping scientists see all the possible “game outcomes.” Just like in rugby, having a solid game plan and adjusting it on the run makes a big difference—SPP just makes sure we’re ready for any moves the weather throws at us

• Upgraded Ocean Wave Model
  The ocean wave model now uses a finer grid, allowing for more accurate wave predictions. This includes improved ways of modeling how wind influences waves, which is crucial for predicting storm surges and coastal conditions. For regions impacted by tropical storms or rough seas, this upgrade offers a more realistic view of wave conditions.

• Improved Data Assimilation – Using More Real-World Data
  Data assimilation, or the process of feeding real-world data into the model, has been enhanced. The model now incorporates more temperature and humidity observations from weather stations worldwide. This allows for quicker adjustments based on current conditions, making the first few days of a forecast more accurate, particularly for cold regions in winter.

• Higher Resolution for Ensemble Forecasts
  The model's Ensemble Data Assimilation (EDA) now has a higher resolution, meaning it can make finer predictions. This will improve forecasts for complex weather patterns, like tropical storms, by better capturing their uncertainty and potential paths. The increased detail could lead to earlier warnings and more accurate tracking.

What Has Improved?

• More Accurate Near-Surface Weather: The model now does a better job of predicting conditions close to the ground, such as temperature and wind, which has practical impacts on daily forecasts.

• Enhanced Wave and Ocean Predictions: With better wave modeling, we get a more reliable forecast of sea conditions, especially important for shipping, fishing, and coastal communities.

• Increased Consistency in Forecasts: The new uncertainty system (SPP) helps reduce errors, because remember we focus on the processes that are occuring in the atmosphere not the outcome of those processes making forecasts more consistent over time. We can trust the ensemble model 'average' or 'mean' a little more than we could prior to the update. 

Are There Any Downsides?

• Computational Demands: Higher resolution and more complex systems require more computing power, which could mean that updating forecasts takes longer. Although the supercomputers at ECMWF are very powerful so this is perhaps not going to be a major issue. 

• Potential for Minor Discrepancies in the Tropics: The new methods might need some adjustments for tropical areas where unique weather patterns make forecasting challenging. A slight degradation to Tropical weather forecasting for some parameters has been evaluated during the testing period (it is important to note that tropical surface winds and surface temp forecasts have still been improved). 

• Increased Tropical Cyclone model track spread: Tying in slightly with the previous negative, the evaluation period indicated that the spread in model forecasts for Tropical Cyclone motion in the new ensemble has increased slightly, making track forecasts potentially slightly less reliable. 

Does Weather IQ offer this model to its users?

We sure do - it is one of many models we offer to help our users predict the weather and we find if we use this model in combination with the Australian ADFD and the high resolution ACCESS C model suites we can get a really accurate picture of what the weather will be like. In the not too distant future we will also be providing our users with the ECMWF Ensemble data (expected implementation in December 2024). All in all we offer our users 12 weather forecast models from local scale to global scale as part of our Weather Centre.  All for less than the price of a cup of coffee a month

Weather IQ Product Comparison 

Where can I learn more about the ECMWF upgrade? 
https://www.ecmwf.int/en/about/media-centre/news/2024/forecast-upgrade-improves-wind-and-temperature-predictions