Skilful seasonal forecasts of streamflow over Europe?

Posted on June 20, 2018 by Rebecca Emerton

Contributed by Louise Arnal, University of Reading & ECMWF

Editor’s Note: Don’t miss the brilliant cartoon summary at the bottom of this post!

Over recent decades, seasonal streamflow forecasting methods have evolved and diversified, reflecting changes in our scientific understanding of streamflow predictability on seasonal timescales and our increasing computer power. The first operational model-based ensemble seasonal streamflow forecast, called the ESP1,2 (ensemble streamflow prediction), relies on the correct knowledge of the initial hydrological conditions (IHC; i.e. of snowpack, soil moisture, streamflow and reservoir levels, etc.) and a large land surface memory, and contains no information on the future climate (if you’d like to know more about the ESP, I recommend reading this very good HEPEX blog post about it!).

In basins where the meteorological forcings drive the streamflow predictability however, this last point is a limitation of the ESP. This motivates the use of seasonal climate forecasts to feed hydrological models and extend the predictability of hydrological variables on seasonal timescales3, which we refer to as climate-model-based seasonal streamflow forecasts (CM-SSF). While studies exploring the skill of CM-SSF are abundant outside of Europe, they are still relatively scarce in this part of the world, partly due to the limited quality of seasonal climate forecasts (particularly for the variables of interest to hydrology: precipitation and temperature) for the extra-tropics.

In our recent paper, published in the HESS special issue on Sub-seasonal to seasonal hydrological forecasting, we carried out a Europe-wide analysis of the skill of the newly operational EFAS (European Flood Awareness System) CM-SSF (produced using the raw ECMWF System 4 seasonal climate forecast [Sys4] as an input to the Lisflood hydrological model), benchmarked against the ESP (produced using historical meteorological observations as an input to Lisflood; both forecasts go out to 7 months lead time). Below are the two main questions tackled in this paper.

C:\Users\sd869122\Documents\conferences&co\EGU 2018\photos\Louise A\IMG-20180409-WA0006.jpg

Louise presenting results from this paper at EGU 2018.

Does seasonal climate information improve the predictability of seasonal streamflow over Europe?

Overall, we found that Sys4 improves the predictability over historical meteorological information for pan-European seasonal streamflow forecasting for the first month of lead time only (in terms of accuracy, sharpness and overall performance). This shows the importance of the IHC and the land surface memory for seasonal streamflow forecasting in Europe. However, the predictability varies per season and we show that the CM-SSF is more skilful at predicting autumn and winter streamflows than for the spring and summer.

Our findings suggest that patterns in the CM-SSF skill are not mirrored in the Sys4 precipitation and temperature hindcasts, which calls for a more in depth look into the link between meteorological and hydrological variables on seasonal timescales over Europe.

What is the potential usefulness of the EFAS seasonal streamflow forecasts for flood preparedness?

As the quality of seasonal streamflow forecasts increases, their usability for decision-making has lagged behind. Translating forecast quality into added value for decision-making is not straightforward for short-range forecasting, let alone on seasonal timescales. While this has been explored for many water-related applications, such as navigation4, reservoir management5, water resource management6 and hydropower7, among others, seasonal streamflow forecasts have yet to be adopted by the flood preparedness community. In this paper, we additionally investigated the ability of the CM-SSF and the ESP to predict lower and higher than normal streamflow conditions up to 7 months ahead, which we term potential usefulness.

Here, we show results for higher than normal streamflows only (see the figure below); results for lower than normal streamflows can be found in the paper. Overall, we found that at least one of the two forecasting systems is capable of predicting higher than normal streamflows months in advance, with the ESP the most potentially useful system generally. However, our results suggest that the CM-SSF is more potentially useful than the ESP beyond 1 month of lead time for certain European regions and seasons, noticeably in winter for ~40 % of Europe.

Following the Red Cross Red Crescent Climate Centre Ready-Set-Go! approach8, seasonal streamflow forecasts could complement existing forecasts at shorter timescales and provide monitoring and early-warning information for flood preparedness.

Maps of the most potentially useful forecasting system (based on its ability to predict higher than normal streamflows, as measured with the ROC score) for all four seasons. The pie charts (also known as camembert plots in France!) display the ‘best’ system for each lead time (i.e. 1 to 7 months), as shown in the example pie chart. There are three possible cases: (1) neither the ESP nor the CM-SSF is potentially useful (red), (2) the ESP is potentially useful and better than the CM-SSF (yellow), or (3) the CM-SSF is potentially useful and better than the ESP (blue).

If you’d like to know more about this research, you can read the full paper here.

And if you’ve always wondered what a paper would be like if it was written as a cartoon with superheroes, check out the cartoon I made to summarise this paper!

C:\Users\sd869122\Documents\sciart projects\paper#3\comic_2.png

References:

1Twedt, T. M., Schaake, J. C., and Peck, E. L.: National Weather Service extended streamflow prediction, Proceedings Western Snow Conference, Albuquerque, New Mexico, 52–57, April 1977.

2Day, G. N.: Extended streamflow forecasting using NWSRFS, J. Water Res. Plan. Man., 111, 157–170, https://doi.org/10.1061/(ASCE)0733-9496(1985)111:2(157), 1985.

3Pagano, T. C. and Garen, D. C.: Integration of climate information and forecasts into western US water supply forecasts, Climate variations, climate change, and water resources engineering, edited by: Garbrecht, J. D. and Piechota, T. C., American Society of Civil Engineers location, Reston, Virginia, US, 86– 103, 2006.

4Meißner, D., Klein, B., and Ionita, M.: Development of a monthly to seasonal forecast framework tailored to inland waterway transport in central Europe, Hydrol. Earth Syst. Sci., 21, 6401–6423, https://doi.org/10.5194/hess-21-6401-2017, 2017.

5Turner, S. W. D., Bennett, J. C., Robertson, D. E., and Galelli, S.: Complex relationship between seasonal streamflow forecast skill and value in reservoir operations, Hydrol. Earth Syst. Sci., 21, 4841–4859, https://doi.org/10.5194/hess-21-4841-2017, 2017.

6Schepen, A., Zhao, T., Wang, Q. J., Zhou, S., and Feikema, P.: Optimising seasonal streamflow forecast lead time for operational decision making in Australia, Hydrol. Earth Syst. Sci., 20, 4117– 4128, https://doi.org/10.5194/hess-20-4117-2016, 2016.

7Hamlet, A. F., Huppert, D., and Lettenmaier, D. P.: Economic Value of Long-Lead Streamflow Forecasts for Columbia River Hydropower, J. Water Res. Plan. Man., 128, 91–101, https://doi.org/10.1061/(ASCE)0733-9496(2002)128:2(91), 2002.

8White, C. J. et al.: Potential applications of subseasonal-to-seasonal (S2S) predictions, Meteorol. Appl., 24, 315–325, https://doi.org/10.1002/met.1654, 2017.

 

Posted in decision making, ensemble techniques, forecast techniques, operational systems, seasonal prediction, verification | Leave a comment

Workshop Summary: Hydrological Services for Business

Posted on June 6, 2018 by Rebecca Emerton

contributed by Shaun Harrigan, ECMWF

Over 60 participants ranging from consultancy companies, hydro-meteorological services, (re)insurance, and academia were welcomed at the European Centre for Medium-Range Weather Forecasts (ECMWF) in Reading, UK from the 8th to 9th of May 2018 to meet with the Global Flood Awareness System (GloFAS) development team. They were provided the opportunity to shape the future of GloFAS forecasting products and service provision. GloFAS is part of the Copernicus Emergency Management Service.

Participants at the ‘Hydrological Services for Business’ workshop at ECMWF in Reading 8th to 9th May 2018; Photo by ECMWF

The first session introduced GloFAS and the need for a global hydrological service. Peter Salamon from the European Commission Joint Research Centre (JRC) kicked off proceedings giving an overview of the Copernicus Emergency Management Service in providing hazard information for improved disaster management in Europe. David Green from NASA put forward the case that hydrological services are critical for global businesses by reducing risk and building resilience. Sazzard Hossain, Marcio Moraes, and Anshul Agarwal presented case study applications of GloFAS for flood forecasting in Bangladesh, Brazil, and Myanmar, respectively.

In the second session Catalina Jamie from the Red Cross Red Crescent Climate Centre showed the importance of hydrological services in the humanitarian sector for providing forecast-based financing allowing action before a disaster happens. John Bevington from JBA Consulting demonstrated that innovative products can be produced by combining raw GloFAS output with flood hazard maps via their Flood Foresight service.

Participants defining what makes a good global hydrological service and discussing the barriers of service uptake. Photo by ECMWF.

Plenty of lively discussion at the poster session. Photo by Hannah Cloke.

Participants got a chance to provide input during an interactive session on defining the most important criteria for a good global hydrological service. Common feedback included provision of information in regions lacking data and scientific capacity in order to aid local decision making; information on quality of forecasts (skill, accuracy, reliability) and associated documentation; service reliability; and availability of training material. Identified key barriers of service uptake included a perceived/actual lack of skill in the service in the region of interest; lack of access to discharge observations in many parts of the globe to allow model calibration and forecast evaluation; and managing the volume and complexity of forecast data.

The interactive participation continued with 3 minute ‘ignite’ talks ranging from Jan Verkade on the Deltares GLOSSIS system for storm surge forecasting, Matteo DallAmico on ‘mysnowmaps.com’, to Andy Wood highlighting hydrological prediction developments at NCAR. There were lively discussions and networking at the evening poster session.

Plenty of lively discussion at the poster session. Photo by Hannah Cloke.

 

Christel Prudhomme from ECMWF kicking off the interactive session on shaping the future of GloFAS. Photo by Hannah Cloke.

Day 2 turned towards how users can help shape future products and service provision of GloFAS. Arthur Essenfelder from the Euro-Mediterranean Centre on Climate Change (CMCC) talked on the importance of co-development between service providers and end users using climate services as an example. This provided fodder for the second interactive session where participants provided their input on improving current products (i.e. 30-day flood and seasonal forecasts) and shaping the development of future products (i.e. weekly forecast summary and rapid flood risk assessment).

 

The final session concentrated on how to improve service provision to end users. Carlo Buontempo from ECMWF showed how large complex datasets are made easily accessible to a variety of end users within the Copernicus Climate Change Service (C3S) Climate Data Store (CDS) together with an online interactive toolbox for retrieving, analysing and plotting data (will go live later in June). Jim Nelson from Brigham Young University (BYU) showed how tailor-made GloFAS forecasts at smaller spatial scales can be provided to the community and presented the infrastructure that enables a service to deliver local access to global forecast data. The session finished by asking users to describe their workflows in regards to obtaining forecast information through ‘user stories’. This feedback will be used to prioritise future GloFAS service developments with the aim to better accommodate user needs.

Peter Salamon closed the workshop and thanked the speakers and participants for their enthusiastic participation and valuable feedback during the interactive sessions that will help prioritise developments and shape the future of GloFAS. Workshop presentations are available online here.

An optional GloFAS training session took place after the workshop which included many hands-on exercises for participants to discover the different products and how they could be used for decision-making.

GloFAS training given by Ruth Coughlan, ECMWF. Photo by Louise Arnal.

About GloFAS

GloFAS went fully operational as a 24/7 supported service on the 23rd of April 2018. GloFAS is part of the Copernicus Emergency Management Service and provides forecasts of floods up to 30 days ahead as well as outlooks of high and low flows up to 4 month ahead in rivers across the world. GloFAS has currently over 1,600 registered users. Its forecasts are freely available to all users and can be accessed here.

 

Posted in forecast users, meetings, operational systems | Leave a comment

Risk Communication for Cyclone Early Warning – Do people get the message, and understand what it means for them?

Posted on May 30, 2018 by Rebecca Emerton

Contributed by Bapon Fakhruddin, Senior DRR and Climate Resilience Specialist, Tonkin+Taylor & Co-chair of the Risk Interpretation and Application of IRDR/ICSU

Every year, New Zealand is impacted by ex-tropical cyclones (ETCs) – typically, one ETC makes landfall per year, between the months of November and April. As tropical cyclones approach New Zealand, they begin to lose their strength and undergo extratropical transition (ETT). Some of these weather systems are as large as New Zealand’s North Island when considering the full diameter of the storm, and ETCs can still induce heavy rainfall as well as strong mean winds and wind gusts, coupled with an increased forward motion, making them just as much of a hazard as tropical systems. ETCs have impacted New Zealand in the past (Cyclone Cook, Pam, Victor, Drena, Fergus, etc.) and have the potential to cause flooding and coastal infrastructure damage, generate primary and secondary wind damage to vegetation, and higher-than-normal wave heights and coastal storm surges. The recent Cyclone Cook impacted many regions and devastated peoples’ livelihoods and properties (e.g. heavy rain and high winds in Bay of Plenty, Gisborne and Hawke’s Bay caused flooding, landslips, fallen trees and widespread power outages).

Figure 1. Cyclone Cook approaching New Zealand. Image via NZ Herald.

Producing and Communicating Coastal Inundation Forecasts

Early warning is a key element of disaster risk reduction. It has long been recognized that if society could have advanced information on weather, the adverse effects associated with it could be minimized. Coastal inundation associated with tropical and extra tropical cyclones has a long history of causing death and destruction along our coastlines—and the threat has never been greater. It is imperative that the ever-increasing coastal population understands cyclone risk, particularly related to coastal inundation and storm surge.

Advances in meteorological, hydrological and engineering sciences are fast generating a range of new methodologies for forecasting weather and flood events, including ensemble prediction systems (EPS) and new hydrological or hydrodynamic models. However, many of these advanced prediction systems have not yet been incorporated into operational forecast systems. Consequently, operational forecasts have not yet been integrated into decision making processes in order to reduce disaster risks. In the real world, it has been observed that people do not always notice warnings, or are unable to understand the meaning of probabilistic forecasts well enough to consider themselves at risk. This provides a call to action for our Met Service and Civil Defence research and operations program to develop and implement new coastal inundation mitigation strategies.

Communication of storm coastal inundation is closely tied to how such forecasts are generated and the accuracy of the scientific data they are based on. For example, tropical cyclone forecasts issued by the Met  Service have lots of uncertainty. To generate coastal inundation information requires the integration of a cyclone model, storm surge model, wave model and hydrodynamic or river model. It is obvious that when linking a number of models, the level of uncertainty will be very high. At the same time, cyclones change path very frequently and the model needs to have the capacity to generate information in short intervals so that forecast information can reach people living within a few kilometers of the coast. While cyclone track forecasting continues to improve, the mean position errors for tropical cyclones in the area near the South Pacific (160ᵒ East to 120ᵒ West, 25ᵒ South to 40ᵒ South – this area is covered by the Tropical Cyclone Warning Center (TCWC) Wellington, operating within the Meteorological Service of New Zealand Ltd, or MetService) during 2014/15 and 2015/16 cyclone season were 73 km and 93 km respectively, for 24 hours lead time. The longer the lead time, the larger the uncertainties in the track forecasts. Figure 2 shows the cyclone forecast track error (based on ECMWF forecasts 2010-2016) with lead time.

Figure 2: Tropical Cyclone track errors in the ECMWF forecasts (2010-2016). The bars at the top (bottom) of the lines signal the 95th (5th) percentile of track errors. The upper (lower) bounds of the quadrate boxes signal the 75th (25th) percentile of track errors. The bars inside the boxes are the median track errors.

Figure 3. Cyclone COOK Cone of Uncertainty forecast by MetService

A major conundrum in weather messaging is how to communicate forecast uncertainty. While opinions are mixed, there is consensus within the weather enterprise that the level of certainty should be communicated as part of forecasts. For example, the cone of uncertainty released by TCWC based on past track error is probably the most recognized uncertainty graphic, certainly in TC-prone areas (See Figure 3), but often, critiqued, actual testing of its interpretation and use by stakeholders is scarce.

Geographers at the University of Alabama have been testing alternative TC warning graphics with the public, and results indicate preference for a Color Probability Cone that is a revised version of one issued by the Australia Bureau of Meteorology (See Figure 4).

Figure 4. Example of color probability cone and inland hazards for community understanding. The inland hazards are outlined in color and expected times for experiencing those specific hazards are included in the colored zone (Radford, 2012, thesis document)

Significant research efforts have focused on understanding how people make evacuation decisions including the important effects of past experience. Research shows that before deciding to take a disruptive and often expensive action such as evacuation, people must understand the forecast, believe it applies to them and, most importantly, feel that they and/or their loved ones are at risk. However, common practice has been to prepare and release forecast messages without adequately understanding how they are received, understood, and interpreted.

For any new forecasts product (e.g. coastal inundation), there will at first be a lack of communication of the warning to the affected people, and interpretation or internalization of the information for decision-making and response. In order to make a good decision, the capacity to generate coastal inundation forecasts with sufficient lead-time and an acceptable degree of accuracy is essential based on end to end early warning framework.

Research and development (RD) advancements in tropical cyclone (TC) forecasts using ensemble methods have been widely used for operational TC track forecasting. Either simple, weighted, or selective ensemble mean TC track forecasts tend to have smaller position errors than single-model-based (deterministic) forecasts. It’s clear that early warning is not helpful unless it reaches the people who need to act, and provides information about impacts (Figure 5). To respond to the early warning, the information needs to be understood and internalized by the public. Thus an interpretation and translation of the scientific information is essential. The new system needs to incorporate users’ needs to enable people to visualize the possible scenarios with probabilities of risk to reduce their vulnerabilities.

Figure 5: End-to-end impact-based early warning system

 

Further Reading:

Lorrey, A. M., Griffiths, G., Fauchereau, N., Diamond, H. J., Chappell, P. R. and Renwick, J. (2014), An ex-tropical cyclone climatology for Auckland, New Zealand. Int. J. Climatol., 34: 1157–1168.

S.H.M. Fakhruddin, Akiyuki Kawasaki, Mukand S. Babel, Community responses to flood early warning system: Case study in Kaijuri Union, Bangladesh, International Journal of Disaster Risk Reduction, Volume 14, Part 4, December 2015, Pages 323-331, ISSN 2212-4209, https://doi.org/10.1016/j.ijdrr.2015.08.004.

TCC 16 report of TCWC Wellington. World Meteorological Organization, RA V meeting, Honiara, Solomon Island

Morrow, B. H., and Lazo, 2015: Effective tropical cyclone forecast and warning communication: Recent social science contributions. Tropical Cyclone Research and Review, 4, 38-48

Fakhruddin SHM (2015) Risk Communications for Coastal Inundation Forecasting to the Community. J Psychol Psychother 5:203. doi: 10.4172/2161-0487.1000203 ↑

Posted in decision making, disaster risk reduction, forecast communication | Leave a comment

Help build Ireland’s new Flood Forecast Centre!

Posted on May 23, 2018 by Rebecca Emerton

Contributed by Sinéad Duffy, Met Éireann

Here at Met Éireann, the Irish national meteorological service, we are working on building an operational Flood Forecast Centre (FFC) for fluvial and coastal floods. After widespread flooding across Ireland in December 2015/January 2016, the Government of Ireland took the decision to establish a National Flood Forecast and Warning Service (NFFWS). Met Éireann are working with the Office of Public Works (OPW) to make this a reality. The NFFWS will incorporate the operational FFC in Met Éireann with guidance for standards and performance overseen by the OPW.

I’m writing this blog post as we have two invitations to make to the HEPEX community and beyond. The first: take a look at the Hydrometeorologist jobs we’ve advertised and apply if you’d like to work in the new FFC at our Dublin headquarters. The second: let us know if you have hydrological models and integrator systems suitable for real-time flood forecasting. We are doing a study to review, develop and trial models suitable for operational forecasting for five representative catchments around Ireland and integrator systems suitable for real-time flood forecasting.

Hydrometeorologists Wanted!

We recently recruited a Chief Hydrometeorologist and we are now recruiting hydrometeorologists to work in a position which will be a mix of operational flood forecasting, development of hydrological and coastal flood forecasting models, and managing dissemination of Met Éireann Flood Forecast Centre products. The jobs will be varied and interesting, and you will be involved in building a national flood forecasting system from the early stages.

We need people with:

  • Experience of the development, calibration and operation of flood forecasting models and services;
  • A qualification of at least Level 8 (B.Sc. Hons) on the National Framework of Qualifications in one or more of the following: Hydrology, Oceanography, Meteorology or where Hydrology was taken as a major component e.g. Civil Engineering, Earth and Environmental Sciences.

The job information booklet and link to the application system are available here. The closing date for applications is 7th June 2018. This is an established (permanent) post subject to successful completion of a probationary contract of one year from appointment date (further details in the information booklet). On completion of three years satisfactory service, the Hydrometeorologist will receive three additional salary increments if they take or have taken exams to show that they are capable of reading and understanding with ease technical publications in any two of the following languages: French, German and Russian.

Other departments in Met Éireann are looking for meteorologists to work in Numerical Weather Prediction, Regional Downscaling, Operational Forecasting and Digital Communications. If you or anybody that you know would like a permanent post in those fields, please have a look at the information booklet and link to the application system here. The closing date for meteorologist post applications is 31st May 2018.

Hydrological Models and Integrator Systems for Operational Fluvial Flood Forecasting in Ireland Wanted!

To help us equip the Flood Forecast Centre for its fluvial forecasting task, we have decided to undertake a study of a range of existing available hydrological models and integrator systems. They will be reviewed, developed and trialed for operational fluvial flood forecasting use in Ireland.

The study was awarded to IMDC, an engineering company experienced in hydrological modelling and forecasting systems. The study started April 2018 and will last until October 2019. The study contains following stages:

  • Comprehensive literature review of existing hydrological models
  • Comprehensive literature review of existing integrator systems. An integrator system combines a hydrological model with real-time observed and forecasted data, starts new simulations, facilitates the visualisation of model input and output, triggers alerts and disseminates information to relevant stakeholders
  • Model development for five representative catchments: Shannon, Barrow, Nore, Slaney and Moy & Killalla
  • Trialling of hydrological models for fluvial flood forecasting. Models are tested on five representative catchments
  • Trialling of integrator systems for fluvial flood forecasting, including a 2 month pre-operational test

Based on the literature review three hydrological models and three integrator systems will be selected for further development and trialling on the five representative catchments. Each model is first calibrated and validated, and then tested both with historic data and with real-time forecast data from a number of meteorological data-sets.

The target forecast accuracy is expressed in terms of advance warning time and flow magnitude. A model should predict the actual peak flow to within +/- 10% and 6 hours of the actual peak at the gauged forecast points, and with a lead time of greater than 24 hours.

More details can be found on the Met Éireann website.

Submissions from Providers

Providers of hydrological models and integrators systems suitable for real-time flood forecasting are kindly invited to get in contact with Met Éireann. Suitable models and integrator systems will then be included in the literature review. Your submission should reach us by 15th of June 2018.

Please contact:

Email: hydroreview@met.ie

C/O: Eoin Sherlock, Met Éireann, Glasnevin Hill, Dublin 9, D09 Y921, Ireland
Tel: +353-1-8064200

cid:image001.png@01D3EEC1.B8B45080 cid:image012.png@01D3EEC7.6010C250 cid:image013.jpg@01D3EEC7.6010C250

Further reading

  1. “Update on the development of a National flood forecasting and warning service for Ireland” by Jim Casey and Oliver Nicholson, OPW in the EFAS Bulletin for Feb/Mar 2018.
  2. Hydrometeorologist job details and application. The closing date is 7th June 2018.
  3. Meteorologist job details and application. The closing date is 31st May 2018.
Posted in announcements-events, jobs | 1 Comment

Recent development of post-processing methods in short-term hydrometeorological ensemble forecasting

Posted on May 9, 2018 by Rebecca Emerton

Contributed by Wentao Li and Qingyun Duan 

Due to various uncertainties in model inputs and outputs, initial and boundary conditions, model structures and parameters, raw forecasts from meteorological or hydrological models suffer from systematic bias and under/overdispersion errors and they need to be corrected before being used in applications. Various statistical post-processing methods have been developed to correct these errors and achieve “sharp” forecasts subject to “reliability”. As in the book “Statistical methods in the atmospheric sciences” by Wilks, statistical post-processing methods can be generally divided to two categories from the view of statistics, namely regression-based methods (e.g., ensemble MOS and logistic regression) and kernel density-based methods (e.g., BMA and ensemble dressing). An example of the flow of a regression-based statistical post-processing method is shown in Figure 1. As there is already a review of post-processing in a previous blog in 2013,  in this blog post we discuss several newly developed post-processing methods for short- to medium-term hydrometeorological forecasting.

Figure 1. An example of a regression-based statistical post-processing method for hydrometeorological ensemble forecasting, modified from Dr. John Schaake’s presentation of Ensemble Pre-Processor (EPP).

As described in several papers (e.g., Scheuerer et al., 2015), there are several difficulties in post-processing variables such as precipitation and streamflow/river stage: (1) these variables follow a mixed distribution of a positive probability at zero value and a skewed continuous distribution for non-zero amounts; (2) the heteroscedasticity problem, namely that the forecast uncertainty increases with the magnitude of forecast variables; and (3) the representation of spatio-temporal and inter-variable dependency, which is important for applications such as hydrological forecasting.

To model the hydrometeorological variables with skewed distribution and non-homogeneous variance, one common treatment is to apply transformations to normalize the variables and stabilize the variance. After the transformation, traditional statistical models under assumptions of Normal distribution and homogeneity can be applied. Examples of the transformations include the Box-Cox or power transformation in heteroscedastic censored logistic regression (HCLR) and the log-sinh transformation in Bayesian joint probability (BJP). Moreover, there are also post-processing models that directly use non-Gaussian distributions without any transformations, such as the censored, shifted Gamma (CSG) distribution-based EMOS by Scheuerer and Hamill (2015). To deal with the heteroscedasticity problem, EMOS model includes the ensemble spread of raw forecasts as predictor to adjust the non-homogenous forecast uncertainty. R packages such as “ensembleMOS” and “crch” have made it easy to apply these methods.

How to model spatio-temporal and inter-variable dependency of hydrometeorological variables has gained much attention in recent years. To solve this problem, several “shuffling techniques” have been developed, namely to “shuffle” the ensemble members generated from the post-processed probability distributions according to some “rank structures” which represent the spatio-temporal and inter-variable dependency. Among these methods, the Schaake shuffle are mostly applied, in which the ensemble members are reordered according to the “rank structures” obtained from historical observation archives. However, the drawback of Schaake shuffle is that the templates from past observations may not represent the current synoptic situation.

Recently, two types of Schaake shuffle variants have been developed. One type is the ensemble copula coupling (ECC) scheme developed by Schefzic et al. (2013). ECC reorders ensemble members according to the “rank structures” of raw ensemble forecasts, thus accounts for the multivariate rank structure information of the current synoptic situation. The other type of variants select the “rank structures” from a subset of historical observations under “similar” situations using synoptic analogs or other similarity criterions. This type of method includes the “SimSchaake” by Schefzic et al. (2016), the minimum divergence Schaake shuffle (MDSS) by Scheuerer et al. (2017) and the meteorological analogues-based Schaake shuffle by Bellier et al. (2017). Wu et al. conducted a comparison experiment of the three schemes of shuffling techniques, namely the original Schaake shuffle and its two types of variants.

Besides this progress, what challenges remain for post-processing? A 2013 blog post by Nathalie Voisin, Jan Verkade and Maria-Helena Ramos (here) included a list of challenges for post-processing, many of which still need to be worked on. We also recommend this chapter by Dr. Thomas M. Hamill, which also emphasizes several challenges in post-processing, such as developing post-processors suitable for limited training data, and sharing post-processing software and data together “to build a postprocessing community”.

What other challenges do you think exist in post-processing? We welcome your comments on your experiences and opinions of post-processing below.

You can find the full review on statistical postprocessing methods for hydrometeorological ensemble forecasting in the authors’ recently published review paper:

Li W, Duan Q, Miao C, et al (2017). A review on statistical postprocessing methods for hydrometeorological ensemble forecasting. WIREs Water, e1246.

Posted in postprocessing | Leave a comment

23 (+1) unsolved questions in hydrology

Posted on May 2, 2018 by Rebecca Emerton

Contributed by Bettina Schaefli

Have you heard about the “23 unsolved problems” initiative of IAHS, in reference to the 23 unsolved problems of Hilbert? After the launch of this initiative in November 2017, the web-based discussion culminated in a round of brainstorming at EGU 2018 and in a voting process at the Vienna Catchment Science Symposium.

Personally, I did not contribute any question at the discussion stage. I simply could not think of a question that relates to observed phenomena, is universal and specific (the three original requirements to formulate the problems). Are all questions that I work on not related to a particular climate region or to modeling rather than observed phenomena?

The proposed questions and the brainstorming session clearly showed that most colleagues did not self-censor their ideas and simply proposed anything what hydrologists currently work on.

At this stage, I really asked myself how we could possibly bring the hundreds of questions and problems (around 260) down to a reasonable number. And what would the added value of such a process be? Asking this question around me during the brainstorming session, I got an interesting answer: at the very least, we can learn something about what our research community is concerned about.

Of course! Why did I not consider this aspect before? With renewed enthusiasm and with a complete perspective change, I went to the Vienna Catchment Science Symposium following EGU; not to decisively influence the final choice of “the unsolved problems” but to observe a scientific experiment of a new kind: put 40 scientists in a room, together with a moderator. Give them a list of 60+ questions and roughly 1.5 h to reformulate, rank or delete them. Of course the process starts slowly. People do not know each other, some hesitate to openly say what they think about questions that were obviously formulated by some of the most famous hydrologists. Should I really vote to delete the favorite question of the moderator?

The great thing about a direct-democratic process, with hand-voting, is that it creates its own dynamic. Everyone can see what you vote or that you don’t vote; and there is no time for deep thinking. Over the course of the exercise, it becomes more and more fun: the moderator announces a problem number, the audience yells “delete” or votes for gold, silver or bronze. Hands go up and down, and even the most intriguing problems are voted within a few seconds.

Incredibly enough, after three such rounds in three parallel rooms, the list was brought down to 16 gold questions and 29 silver questions. The final outcome is now in the hands of a paper drafting team and will be published in a paper with a giant author list in the Hydrological Sciences Journal. Once you read the questions/problems, you will no longer be able to decipher what complex processes have led to these specific questions. But every reader might find a few unexpected questions that trigger new thinking; and together, the selected problems nicely reflect what the hot topics are in hydrology at the moment. With some gaps however. The question of how to make hydrology more open and replicable is not reflected in the retained questions, for example.

And: the single most important problem has been completely forgotten: why is hydrology not more gender-balanced?

Posted in activities, meetings, projects, social participation | Tagged , , | 1 Comment

Take a Breath of the Wild: geoscience in computer games

Posted on April 25, 2018 by Rebecca Emerton

contributed by Rolf Hut

During the recent “Games for Geoscience” session at the EGU General Assembly in Vienna, researchers, many from HEPEX among them, presented how they use games to actively communicate scientific results and/or to engage with stakeholders.

The success of using games for science communication sparked a new question in us (Sam Illingworth, Chris Skinner, Casper Albers and Rolf Hut). We wonder how people in general, and earth scientists in particular, view computer-generated landscape images that are now common in (video) games. To study this, we have designed a survey in which people are asked to judge images of landscapes from both the real world and from fictional worlds. We would greatly appreciate any input for this (short) survey, which you can access here if you’d like to take part: “Into the wild: how realistic are artistic renderings?”

Modern video games like Zelda: Breath of the Wild present players with a fake, but seemingly realistic, goescientific environment.

Thanks in advance!

Posted in activities, social participation | Tagged , , | Leave a comment

HEPEX Highlights from EGU 2018

Posted on April 18, 2018 by Rebecca Emerton

Last week, 15,075 scientists from 106 countries attended the European Geosciences Union (EGU) General Assembly in Vienna. The week was jam-packed with more than 17,000 talks, posters and PICO presentations in 666 different sessions. The HEPEX community was represented across 5 oral/poster sessions and 1 PICO session, plus many other related sessions and events! Attendees could attend talks covering a huge range of research, from very local-scale all the way up to global scale studies, from flash flood forecasting to seasonal hydrological predictions, and on the history of hydrology and science communication. The week was a great showcase of the excellent work around the world related to ensemble hydrological forecasting.

The Austria Centre, Vienna

This year, we wanted to hear from the HEPEX community about their experience of #EGU18, so we sent out a “call for highlights”.

These were your highlights from the week in Vienna: Continue reading →

Posted in activities, meetings | Tagged , | Leave a comment

Using ensemble forecasts to inform risk-based operations of a reservoir in Northern California

Posted on April 11, 2018 by Andy Wood

Contributed by:  Chris Delaney, Sonoma County Water Agency; John Mendoza, Sonoma County Water Agency; Brett Whitin, California Nevada River Forecast Center; Rob Hartman, Consultant

Lake Mendocino is a reservoir located in Mendocino County, California, about 110 miles north of the City of San Francisco. This small 144 million cubic meter reservoir (Figure 1) releases water into the Russian River and provides both flood protection and water supply to downstream communities. Lake Mendocino is cooperatively managed by 2 government offices: the federal U.S. Army Corps of Engineers manages flood operations and the Sonoma County Water Agency manages water supply operations. The lake receives inflow from natural runoff as well from an adjacent river system, the Eel River, through a hydroelectric facility (Potter Valley Project) a short distance upstream of the lake. Recent changes in the operations of this hydroelectric facility have drastically reduced the average annual inflow of Lake Mendocino by 45%, contributing to a water supply crisis for the region. This crisis, among other reasons, made Lake Mendocino an ideal location to evaluate forecast based operations in an effort to recover lost water supply reliability, thus Lake Mendocino was selected as a pilot location for the Forecast Informed Reservoir Operations (FIRO) program. FIRO is led by a number of federal, state and local agencies including the U.S. Army Corps of Engineers San Francisco District (‘Corps’), Sonoma County Water Agency (SCWA), National Oceanic and Atmospheric Association (NOAA), and the Center for Western Weather and Water Extremes (CW3E) at Scripps Institute.

The primary goal of this pilot is to evaluate whether forecasts can be used to inform flood operations to improve the water supply capture of Lake Mendocino without increasing flood risk to downstream communities. So why would changing reservoir flood operations improve water supply? The primary reason relates to a currently used operational rule called the storage guide curve (Figure 2), which determines the maximum water supply storage level of the reservoir. This seasonally varying guide provides increased flood capacity in the wet months (November – February) and increased water supply capacity during the dry months (May – September).

This design works well during years with sufficient springtime (March – May) rainfall to fill the reservoir as the guide curve increases, but dry spring years can be challenging for water supply because the region typically receives very little precipitation during the summer and fall. Under FIRO, the objective is to detain wet season runoff above the guide curve level until forecasts indicate water should be evacuated to provide adequate volume for predicted flood events. Stored water released in advance of a forecasted flood event would be recovered by inflows from the flood event and held in the reservoir for water supply.

Another reason that Lake Mendocino is an ideal location for the FIRO pilot is that the NOAA California Nevada River Forecast Center (CNRFC) currently prepares hydrologic forecasts of reservoir inflows and for points downstream. A key forecast product is an ensemble forecast produced with the Hydrologic Ensemble Forecast System (HEFS), which generates a 59-member ensemble with an hourly timestep up to the 15-day forecast horizon, transitioning to a daily timestep for the 16 to 365 day forecast horizon.

A central challenge for this project is to develop new operational methodologies that incorporate forecast information to make reservoir release decisions to meet the project goals. One such methodology that is being evaluated for Lake Mendocino is called Ensemble Forecast Operations (EFO), which was developed by an engineer at the Sonoma County Water Agency, Chris Delaney (that’s me). I originally developed a simple proof-of-concept model, which I have refined with the help of my colleague and fellow engineer, John Mendoza. This methodology uses the HEFS ensemble forecast prepared by the CNRFC to evaluate the risk of reaching the maximum reservoir storage level. We want to avoid going above this maximum level because this would increase the risk of uncontrolled spillway releases and could flood downstream communities.

The way the EFO methodology works is quite straightforward. Each hydrologic ensemble member is independently modeled to forecast reservoir storage assuming no water is released. Forecasted risk is evaluated for each timestep in the forecast horizon as the percentage of ensemble member that exceed the maximum storage level (137 million cubic meters). The top panel of Figure 3 provides an example of a storage forecast with a 15-day forecast horizon and the maximum storage level shown as the black dashed line. The bottom panel of Figure 3 provides an example of the risk forecast shown as the red line. A key component of the EFO methodology is something called the risk tolerance curve, which is shown in the bottom panel as the blue dashed line. This curve defines the maximum allowable risk for each forecast timestep.

Figure 3 – (top) Storage forecast ensemble assuming zero releases; (bottom) risk of exceeding storage threshold, and operational risk tolerance level.

If forecasted risk exceeds the risk tolerance curve, as in our example, a release schedule is developed that mitigates the forecasted risk at or below the curve. For this example, the model simulated a release of 56 cubic meters per second to reduce the forecasted risk to the risk tolerance level. This is illustrated in the Figure 4 showing forecasted risk and storage levels after the release schedule has been applied. The model completes this process, updating release schedules each day as a new forecast is issued by the CNRFC.

Figure 4 – (top) Storage forecast ensemble after reservoir release schedules are calculated; (bottom) the resulting risk of exceeding the storage threshold matches the EFO risk tolerance policy.

It is important to note that a thorough analysis of the EFO methodology was made possible by a ensemble inflow hindcast dataset from the HEFS prepared by CNRFC for the Russian River Basin. The hindcast study allowed estimating what the forecast would have been for each day from 1985 to 2010, given their current forecasting skill. This allowed us to simulate reservoir operations and flows for points downstream under a variety of hydrologic conditions including the 1986 flood of record. Results show significant increases in storage levels (Figure 5) for almost the entire simulation period, with a 35% increase in median end of water year (September 30) storage levels compared to simulated existing operations.

Figure 5 – Storage levels with and without ensemble forecast operations (EFO)

Despite the generally increased storages, hindcast simulation results do not demonstrate any increased flood risk, with no instances of increased flow levels above flood stage. Figure 6 shows flows plotted as percent exceedance for the most flood prone region immediately downstream of Lake Mendocino.

Figure 6 – The distribution of downstream flows with current and EFO operations, showing that no significant increase in the risk of downstream flood-level flows with EFO.

The results of this study demonstrate that the implementation of the EFO methodology will likely achieve the goals of the FIRO Lake Mendocino pilot project – to improve water supply reliability without increasing flood risk to downstream communities. A more complete description of this study included in the Preliminary Viability Assessment Lake Mendocino Forecast Informed Reservoir Operations prepared by the FIRO Steering Committee in 2017. Following these positive outcomes, the Steering Committee is also working with the Corps to implement a revised version of the EFO methodology on an interim trial basis. For more information on FIRO and progress of the Lake Mendocino pilot project please visit the CW3E FIRO website.

We welcome comments and questions from the HEPEX community!

Additional author info: Jay Jasperse of the Sonoma County Water Agency and Marty Ralph of CW3E serve as co-chairs for the Lake Mendocino FIRO program. Rob Hartman is the former chief of the NWS California Nevada River Forecast Center, and has been a long-time contributor to HEPEX. Chris Delaney can be reached with questions and feedback at: Chris.Delaney /at/ scwa.ca.gov

Posted in case-studies, decision making, ensemble techniques, operational systems, risk management, water management | 3 Comments

EGU2018 #hepex Twitter feed – stay updated of events as they unfold

Posted on April 6, 2018 by Jan Verkade

The week of April 9 sees the 2018 edition of the annual convention of the European Geosciences Union. “EGU2018” will be visited by approx 15,000 geoscientists including many working in the hydrometeorological arena. Last week, Fredrik provided us with an overview of some of the sessions and events that will be of interest to us HEPEX-ers.

Many of the convention-visiting-hydrometeorologists will carry a Twitter enabled phone which they will use to brief the world about the most salient insights gained. These 280 character gems will be displayed below.

Enjoy!

Note to Tweeps: don’t forget to include the #hepex in your Tweets!


Posted in Unclassified | Leave a comment