Blogs

Probabilistic forecasting – are we there yet and how can we best use it in the real world

Probabilistic forecasting – are we there yet and how can we best use it in the real world

Building on the opportunities provided by coming together for the FEWS International User Days, we held an impromptu informal knowledge sharing session to compare notes on the current state of play in probabilistic flood forecasting, current operational use and plans for the future. We heard from BoM Australia, Switzerland, SEPA, Environment Agency/FFC and of course Deltares.

So what have we learnt?

All operators are already using or trialling probabilistic forecasts, mostly in the form of ensemble NWP forecasts. Whilst it is acknowledged that such an approach only captures some of the uncertainties it has been the natural entry point to extend forecasting lead times. Capturing wider uncertainties probabilistically in a way that is meaningful for operational forecasting and decision making still emains challenging for most.

It is important to be clear upfront of the purpose for using probabilistic forecasting. For instance, is it to extend forecasting lead times or to capture all understanding key uncertainties (which may require different approaches).

Putting users and decision makers at heart will help to tailor capabilities and products to make sure value is gained from prob. forecasts. Working collaboratively with users in peace time what decisions should be informed / what is possible to forecast (lead times uncertainties) will help to ensure that benefits from prob. forecasting can be realised.

Techniques in use vary, from ensemble NWP, multi model approaches to Kalman Filters. Most user currently focus on ensemble NWP as a starting point. This leaves the question about whether other sources of uncertainties do practically matter and should also be considered.

As a starting point there is great interest conducting trials for selected catchments/situations to gain experience how it work, gain confidence in forecasts and crucially, seek user feedback how results can help inform practical decision making.

Regular performance monitoring will also help to test how to best use the new forecasting capability and become truly forecast led - asking the key question ‘could we have acted it earlier?’ If users ‘still wait and see’ until observations are in, valuable time to act could be lost. This will be part and parcel of shifting the culture towards being truly forecast led. The mantra in the EA has been ‘thinking big, acting early, be visible’ and it has helped us making this important shift, but there is more to!

We are keen on your thoughts, please share how you are using or planning to use probabilistic forecasts. Any examples, experiences good practice are very welcome from using it for national and local scale models, larger and smaller catchments and, crucially how it can help to enable earlier / improved decisions!

So, are we there yet? Not quite, but we are certainly a bit closer. Let’s work together so that the next step to gain benefits from prob. forecasting does not take as long.

Examples across some organisations

SEPA is using ensemble weather forecasts to run their national G2G model. In the next 6-12 months there are plans to trail running a number of ‘problem’ catchments with ensemble rainfall and using surge ensembles to help inform scenario planning. Rainfall ensembles have also been trialled since the Glasgow Commonwealth Games.

For the Environment Agency/FFC ensemble weather forecasts are already used to inform hydro-metrological advice and to run our national scale hydrological model G2G and surface water models. Going forward, our new Incident Management Forecasting System, IMFS, will be ‘ensemble ready’. A range of ensembles for rainfall, surge and wind will be available and be us to inform our provision of ‘Best Estimate and ‘Reasonable Worst Case Scenarios’ which will drive our local forecasting models. In future we also plan to trial the use of ensemble forecasts for running local models to explore how it best help inform decision making.

Switzerland – There are plans to use rainfall ensemble forecasts to drive the forecasts for their larger river catchments. The local Met Services are planning to move away from deterministic forecasts towards ensemble only forecasts, which will be a catalyst for this change.

Bureau of Meteorology, Australia are already using rainfall ensembles (lagged) as an extension to nowcasts at least once a day. The purpose is to increase lead time to enable earlier evacuation where required. The plan is to operationalise this service for selected locations next year. Through ensembles, extended lead times for 7 day flood scenarios are be possible. Increasingly the focus is on detecting thresholds crossings and the recession of hydrographs rather than ‘just peaks’.