Wednesday, March 27th, I’ll be giving a webinar on pre-processing in hydrologic forecasting. The webinar is part of a new series of online presentations that are part of the HEPEX initiative. It is hosted by ECMWF and starts at 4pm CET. Full details of how to join the webinar are found here.
The webinar’s full title is “Post-processing ECMWF precipitation and temperature ensemble reforecasts for operational hydrologic forecasting at various spatial scales”. The accompanying abstract reads as follows:
The ECMWF temperature and precipitation ensemble reforecasts are evaluated for biases in the mean and spread, and how these biases propagate to streamflow ensemble forecasts. The forcing ensembles are subsequently post-processed to reduce bias and increase skill, and it is investigated whether this leads to improved streamflow ensemble forecasts. Multiple post-processing techniques are used: quantile-to-quantile transform, linear regression with an assumption of joint normality and logistic regression. Both raw and post-processed ensembles are run through a hydrologic model of the river Rhine to create streamflow ensembles. The results are compared using multiple verification metrics and skill scores: relative mean error, Brier skill score and its decompositions, mean continuous ranked probability skill score and its decomposition, and the ROC score. Verification of the streamflow ensembles is performed at multiple spatial scales: relatively small headwater basins, large tributaries and the Rhine outlet at Lobith. The streamflow ensembles are verified against simulated streamflow, in order to isolate the effects of biases in the forcing ensembles and any improvements therein. The results show that the forcing ensembles contain significant biases, and that these cascade to the streamflow ensembles. Some of the bias in the forcing ensembles is unconditional in nature; this was resolved by a simple quantile-to-quantile transform. Improvements in conditional bias and skill of the forcing ensembles vary with forecast lead time, amount, and spatial scale, but are generally moderate. The translation to streamflow forecast skill is further muted, and several explanations are considered, including limitations in the modelling of the space-time covariability of the forcing ensembles.
The presentation is based on research that James Brown, Albrecht Weerts, Paolo Reggiani and myself have conducted. A manuscript describing this research has been accepted for publication in Journal of Hydrology, pending ‘moderate’ revisions.
The WordPress.com stats helper monkeys prepared a 2012 annual report for this blog.
Here’s an excerpt:
600 people reached the top of Mt. Everest in 2012. This blog got about 5,400 views in 2012. If every person who reached the top of Mt. Everest viewed this blog, it would have taken 9 years to get that many views.
Click here to see the complete report.
This afternoon, at the FloodRisk2012 conference in Rotterdam, I will present on the findings of the Probabilistic Forecast Use project that we’re currently finalising.
There is a strong theoretical rationale for using probabilistic rather than deterministic forecasts. Currently, however, there exist no best practices for effectively using probability forecasts. In the project, we tried to contribute to the development thereof. I think we achieved quite a lot. The main findings I am presenting on this afternoon are:
- Hydrological forecasting community supplies hazards whereas often, users are more interested in consequences
- Manipulating – not understanding – probabilities is an issue; asking the right question of a forecast largely resolves this.
- Disclaimers apply to the risk rationale
There’s more to say about the topic, obviously. Have a look at attached presentation slides and let me know what you think! If you can make it to the conference: today, Wednesday November 21st, 4pm, room 6.
Wednesday update: there has been an eruption alright… of neighbouring Tongariro!
A few years ago my partner and I hiked a six-day tour of Mt Ruapehu. This is a beautiful mountain, probably best known outside of New Zealand for its role as Mount Doom in one of the Lord of the Ring films.To the geological community, it is known as an active volcano which’ eruptions has lead to lahars in the past, sometimes with many fatalities as a result.
The title of this blog post is an exaggeration of the truth, of course, but the fact is that the NZ Department of Conservation has issued a warning:
A Volcanic Alert Bulletin issued today by GNS Science summarising recent measurements at Ruapehu indicates the likelihood of eruptions from the mountain has increased.
There is no certainty that an eruption will take place of course, and I am guessing that the uncertainty cannot be characterised by a probability estimate.
Below image was posted on Twitter yesterday by Environment’s Agency David Troup. I like it a lot. To me, it gives an instant overview of flow levels across England and Wales. I’d love to have a similar graphic available in the two forecasting systems I use (the Dutch system for Rhine and Meuse, and EFAS, the European Flood Awareness System).
Some thoughts on below picture:
- I wonder who the target audience for this graphic is. The percentiles are maybe a bit complicated for those not used to it.
- The colours used are different from what I would use. For my application (flood forecasting), I’d express flooding as red, not black.
- More and more often, I see observations and forecasts of hydrological variables expressed as relative values to some reference, rather than in absolute values. Here, the baseline is the climatology of flow at the hydrological stations. In the US, forecasts are often expressed relative to “normal”. This development, I think, comes from Anglo-Saxon environments.
- This overview is well suited if there are not too many stations on the map.
- The map does not indicate whether these flow levels cause flooding or not. At different locations, flooding may occur at
- If similar graphs were to be used for forecasting, one will quickly run out of available dimensions. Forecasts would require the leadtime dimension to be indicated somehow, and also the uncertainty in the forecast. Edwin Welles -a colleague of mine at Deltares- and I have developed some ideas about this, which I’ll share later.
Again, I think Dave’s map is a great way of showing spatially variable information. Above items should be seen as considerations in case similar graphs were to be implemented in the systems I use.
November 12 update: Dave let me know that these maps are used to show water resources situation. Makes perfect sense!
Last January, water board Noorderzijlvest experienced near-flooding. Large amounts of precipitation coincided with above normal tides. For a few days, the water board was unable to use its pumping stations for drainage, and the polders gradually filled with water. At some point, emergency managers decided to evacuate some of the polders and villages. As this doesn’t happen very often, this was widely reported in the national media.
The regional security authorities (“Veiligheidsregio Groningen“) made a 15 minute video (in Dutch only) about this event. It’s interesting to see that a lot of elements of flood risk management come together:
- flood hazard and flood risk
- flood stage and levee strength
- warning and response
- managing uncertainties in decision making
(if you like the science of hydrology and decision-making, check out this page)
For almost a year now, I’ve been a member of the Dutch River Forecasting Service, which is responsible for monitoring and forecasting of water levels and discharges of rivers Rhine and Meuse. For this, we have a wonderful tool available: FEWS Rivers. The tool imports meteorological as well as hydrological observations and forecasts and allows one to run these through hydrological and hydrodynamic rivers to produce estimates of future stage and flow. There is really a wealth of information available, making it difficult sometimes to find one’s way.
Rhine-Meuse delta, just downstream of my forecasting locations
So how do I go about making a forecast? Rather than simply looking at the latest available forecasts and choosing one that fits my beliefs, I try to dig a little deeper. Here’s the list of questions I try to answer:
- What does the meteorological history look like?
- What do the meteorological forecasts say?
- What’s the hydrological history?
- What do the hydrological models say?
- What is the current action level? What are the criteria for moving to the next level and have these been met?
Typically, this analysis will take about one hour for each of the two rivers. Usually I only do this at the beginning of my shift, though. Once I (think I) know what the situation is like, I can suffice by looking at incremental information only, which takes a lot less time. Obviously, this is true only if nothing much is happening. If any flood waves are coming my way, analysis takes longer.
If I can find the time, I’ll go a bit deeper into each of above questions.