Detailed program. - Des prévisions hydrologiques

FROM OPERATIONAL HYDROLOGICAL FORECAST TO RESERVOIR MANAGEMENT
OPTIMIZATION
DES PRÉVISIONS HYDROLOGIQUES OPÉRATIONNELLES VERS UNE OPTIMISATION DE
LA GESTION DES RÉSERVOIRS
September 17 – 19, 2014, Québec, Québec, Canada
Program Outline
Day 1 Wednesday 17th
8h30 – 10h00 Registration
10h-10h15 Opening Session
10h15 – 11h00 Plenary session 1
Jery R. Stedinger
11h00 – 12h15 Session 1
Reservoir Optimization I
12h15 13h45 Lunch
13h45 – 15h00 Session 2
Hydrological ensemble predictions and
Inflow Forecast I
Day 2 Thursday 18th
Day 3 Friday 19th
8h30 – 9h15 Plenary session 2
John C. Shaake
9h15 – 10h05 Session 4
Hydrological ensemble predictions and
Inflow Forecast II
8h30 – 9h15 Plenary session 4
Andrea Castelletti
10h05 – 10h30 Coffee break
10h30 – 11h00 Coffee break
10h30 – 12h10 Session 5
Reservoir Management and Inflow
forecast II
12h10– 13h30 Lunch
9h15 – 10h30 Session 8
Reservoir Optimization II
11h00 – 12h00 Panel
13h30 – 14h15 Plenary session 3
François Anctil, Vincent Fortin
15h00 – 15h30 Coffee break
14h15 – 15h05 Session 6
Hydrological ensemble predictions and
Inflow Forecast III
15h30 – 16h45 Session 3
Reservoir Management and Inflow
forecast I
15h05 – 15h30 Coffee break
17h00 – 19h00 Happy Hour
15h30 – 16h45 Session 7
Reservoir Management and Inflow
forecast III
2
Location of Activities
The workshop takes places at the Château Laurier downtown Québec city, 1220, Place
George-V Ouest, Québec G1R 5B8.
All Activities take places at the RC Main Floor “Salle des Plaines A, B and C”.
3
Day #1
Plenary session 1
Chaired by André St-Hilaire
10h15 – 11h00 Value and Use of Forecasts in Hydropower Optimization Models
Jery R. Stedinger, PhD, NAE, Dist. M. ASCE
School of Civil and Environmental Engineering, Cornell University
Research has showed that the use of forecasts is of value in the optimization of hydropower
system operations. But what is not clear is the best way to represent inflow forecasts and their
uncertainty. Is stochastic programming employing ensemble streamflow forecasts in an simple
event tree sufficient, or should such trees have multiple branches, and how many branches? Is
much to be gained by a full (sampling) stochastic dynamic programming (SSDP) model
employing streamflow sequences and reflecting the evolution of streamflow, forecasts and their
uncertainty? Recent results for a small New England system and the large Federal Columbia
River system will be described. Also of interest are efforts to extend the range of problems
feasibly addressed with SSDP models by focusing on realistic sets of states within the state space
and using effective numerical approximations of the cost-to-go function.
4
Session 1: Reservoir Optimization I
Chaired by Michel Gendreau
11h00 – 11h25 Dealing with uncertainties in water management systems: from theoretical to
practical framework.
Bruno Larouche, Jean Paquin, Marco Latraverse and Pascal Côté
Québec Power operation, Rio Tinto Alcan
Rio Tinto Alcan operates two private hydro electrical systems in Canada to supply energy to its
aluminum smelters. The Quebec’s system has an installed capacity of more than 3000 MW and
the other system in British-Columbia has an installed capacity of 1000 MW. The presentation
shows how the water management group has set up its R&D project in an open innovation
mode and how the link between the operational group and the development team has been
managed. Some developments regarding inflow forecast and stochastic optimization will be
presented.
11h25 – 11h50 Short-term Reservoir Management of Hydropower Reservoirs by Multi-Stage
Stochastic Optimization.
Dirk Schwanenberg, Divas Karimanzira, Steffi Naumann and Christopher R. Allen
We present recent implementations of an open source optimization model for the short-term
management of hydropower reservoirs over forecast horizons of 15-21 days. Focus is the
generation and assessment of probabilistic ensemble forecasts, its transformation into scenario
trees and integration into a novel multi-stage stochastic optimization approach. Use cases
include reservoir systems operated by the Brazilian Companhia Energética de Minas Gerais S.A.
(CEMIG) in the state of Minais Gerais, Brazil, and the Bonneville Power Administration (BPA) in
the Pacific Northwest of the United States.
11h50 – 12h15 Performance of Sampling Stochastic Dynamic Programming Algorithm with
Various Inflow Scenario Generation Methods.
Jennifer Schaffer and Ziad Shawwash
We present the implementation of a sampling stochastic dynamic programming (SSDP)
algorithm to maximize water value for the BC Hydro hydroelectric system in British Columbia,
Canada. This study investigates model performance and outputs with variation of inflow inputs.
Sixty years of historical data is used in various methods to generate inflow scenarios including
historical sequences, ensemble streamflow forecasts (historical and future), and an
autoregressive model with lag-1. Seasonal volume forecast is employed as a hydrologic state
variable. We present results of our implementation of the SSDP algorithm including discussion
on improved reservoir operation policy and value of water functions.
5
Session 2: Hydrological ensemble predictions and Inflow Forecast I
Chaired by Marie-Amélie Boucher
13h45 – 14h10 Ensemble Forecast of Water Levels and Flows in the Lake Ontario - St.
Lawrence River System
Jacob Bruxer, Rob Caldwell, Yin Fan, Paul Yu
The International St. Lawrence River Board of Control ensures outflows from Lake Ontario
through the St. Lawrence River meet the requirements of the International Joint Commission’s
Orders of Approval. In support of this objective, Board staff produces a bi-nationally
coordinated weekly forecast of Lake Ontario water levels and outflows. An ensemble
forecasting technique has been developed that employs a time series of historical estimates of
weekly net basin supplies derived as a residual from water level and flow data. Each of the
historical net basin supply scenarios, along with observed initial conditions, is then used as input
to a regulation and routing model, producing an ensemble of simulated water levels and flows
throughout the Lake Ontario - St. Lawrence River system. A comparison of probabilistic results
of the ensemble forecast compare favourably to the previous forecast method employed by the
Board, and the ensemble method more accurately portrays the true variability in water levels
and flows throughout the system.
14h10 – 14h35 Ensemble forecasts for the prediction of snowmelt of the Mistassibi
watershed
Qingxiao Zhou and Marie-Amélie Boucher
Snow is one of the most important parameters for hydrological forecasting in northern
catchments. The uncertainty on snow measurement and modelling is reflected on the forecasts
for the spring freshet (both timing and volume). In this work, a degree-day snow module
(CemaNeige) and an energy balance snow module (UEB) are both used in conjunction with the
hydrological model CEQUEAU and ensemble meteorological forecasts from Environment
Canada. This allows for the assessment of the uncertainty related to spring streamflow forecasts
by accounting both for the uncertainty related to meteorological conditions and to completely
different modelling schemes for snow processes.
6
14h35 – 15h00 Modeling hydraulics inflows for several plants of a valley
Jérôme Collet
EDF R&D
This study is part of a project to improve hydraulic valleys management practices, in order to
improve compliance with environmental constraints. For this, we have to model the behavior of
inflows. As years are almost independent one to another, each one is considered as a large
observation, which allows modeling seasonal and long-range dependencies. The residual noise is
modeled by a homogeneous process. To validate the model, we calculate averages on several
plants and several days, comparing their distribution to the observed one.
Session 3: Reservoir Management and Inflow forecast I
Chaired by Richard Turcotte
15h30 – 15h55 Dealing with probabilistic constraints in EDF river-chains management
Simon Fécamp
EDF R&D
The hydro-electric power plants managed by EDF undergo many constraints that don’t allow
feasibility for each water inflows scenario. Nevertheless, for issuing constraints EDF aims at
reaching a given probability of respect. To anticipate constraints, we use a “constrained
management”: if we don’t fulfill some conditions on the levels of the reservoirs, our decisions
are based on a non-economic criteria. We aim at maximizing the expected revenue of the riverchain depending on the commands and the conditions of the constrained management, subject
to the expected probabilities of respect of the constraints. We present a heuristic approach to
solve this problem. This consists in iterating strategies and associated probabilities of respect
calculus.
15h55 – 16h20 An Operational System for Water Supply Forecasts and Reservoir
Management
Karel Heynert, Matthijs Lemans and Edwin Welles
Deltares
We present a software system used around the world for operational forecasting and
regulations. The Delft Flood Early Warning System (Delft-FEWS) grew up as a platform to
support flood forecasting, but has now been implemented for purposes well beyond that,
including operational reservoir optimization. The popularity of the Delft-FEWS platform is
derived from the flexible architecture it implements. Reservoir optimization and hydrological
models are linked synchronously through time series. Parameter storage, model connectivity,
basin descriptions, operational work flows and predefined display configurations are all stored
by FEWS. Delft-FEWS supports uncertainty estimates via ensemble model runs. It facilitates an
integrated forecast and optimization environment.
7
16h20 – 16h45 Reservoir Inflow Forecasting for the Winnipeg River using integrated
Distributed Hydrological Modelling and Reservoir Simulation
W. Jenkinson, J. Bomhof, M. Serrer, M. DeWolfe
The Lake of the Woods Control Board (LWCB) is a Canadian board, formed in 1919 as required
under Canadian federal and provincial legislation and a Canada-United States Treaty. The LWCB
regulates the water levels of Lake of the Woods and Lac Seul, and the flows in the Winnipeg and
English Rivers downstream of these lakes to their junction. Lake levels and river flows are
regulated by operating dams at the outlets of Lake of the Woods and Lac Seul, and a number of
other dams in the watershed play critical roles in flow and lake level regulation throughout the
150,000 km² watershed. The LWCB seeks to consider all stakeholder interests in the basin, and
using the best available measuring, forecasting and information systems, the LWCB aims for
solutions that accommodate all interests. The LWCB continually seeks to improve their
capabilities in water level predictions and flow forecasting to better serve the stakeholders living
and working in the watershed. To this end, the LWCB is developing new forecasting and
modelling tools to assist in deterministic prediction to improve dam and hydraulic control
operation procedures. As part of a development project lead by the National Research Council
of Canada (NRC) and involving Environment Canada and Manitoba Hydro, the LWCB is
implementing an integrated inflow and lake/reservoir forecasting and decision support system
using a combination of HEC-ResSim (Reservoir Simulation) model to provide real-time decision
support of basin regulation and WATFLOOD for hydrological inflow forecasting. The WATFLOOD
model is calibrated using the new Environment Canada (EC) precipitation reanalysis product
CaPA (Canadian Precipitation Analysis), meteorological station temperature data, EC streamflow
and lake level data and integrated into a custom-built flow forecasting system. This novel system
of integrated models will provide real-time evaluations of the hydrological state of the
watershed and the impacts to stakeholders across the basin. The forecasting framework was
designed to run the WATFLOOD model operationally and automatically provide flow forecast
estimates to HEC-ResSim operational model. The framework provides users the capability to
apply a number of state variable updates including channel storage, lake levels, accumulated
snow pack, and precipitation rate adjustments, or any combination of the above, to calibrate
flow forecast simulations. Additionally the forecasting framework automatically downloads,
transforms and incorporates the Regional and Global Deterministic Prediction System (RDPS and
GDPS) forecast model data to provide sub-basin 10-day forecasted inflows. This is the first
known operational deployment of WATFLOOD for flow forecasting. Preliminary results show
that using CaPA precipitation data, instead of gauge data, improves the temporal and spatial
resolution of modelled streamflows. Additionally, the standardized options of state variable
updates ensure consistent optimization and thorough documentation of each forecast. This
forecast system implementation will provide flows directly to HEC-ResSim to optimize regulation
in the basin on a weekly basis, and will also be used to predict flow peaks during high water
events.
8
Day #2
Plenary session 2
Chaired by Luc Perreault
8h30 – 9h15
Hydrologic Ensemble Prediction: Past, Present and Future
John C. Shaake
Consulting Hydrologist, NOAA/NWS ret.
The first hydrologic ensemble forecasts were made beginning about 40 years ago. These were
based only on hydrologic initial conditions, did not use weather nor climate forecasts and were
most valuable for medium and longer range predictions. Since then there have been great
advances in weather and climate forecasting as well as development of techniques for using
them in hydrologic ensemble prediction. We will reflect on how this process has evolved and
look at some examples of current operational ensemble forecast systems. This will include
discussion of why ensemble prediction (as well as a general manifestation of probabilistic
prediction) is important for hydrologic applications. It will include some examples of
operational user applications of hydrologic ensemble forecasts. Some of the future
opportunities and challenges that remain will be discussed. Many of these are being addressed
by participants in the international Hydrologic Ensemble Prediction Project (HEPEX) that held its
10th Anniversary Workshop in June, 2014 at the U.S. National Centers for Environmental
Prediction. Some of the results of that workshop will be reviewed.
9
Session 4: Hydrological ensemble predictions and Inflow Forecast II
Chaired by François Brissette
9h15 – 9h40
Prévisions d'apports probabilistes à Hydro-Québec Production
Claude Gignac, Éric Crobeddu, Louise Rémillard
Hydro-Québec Production
Hydro-Québec est le plus grand producteur d'électricité au Canada. Il possède 60 centrales pour
une puissance installée de 35 900 MW, 26 grands réservoirs ayant une capacité de stockage de
175 TWh, 656 barrages et 97 ouvrages de régulation. Hydro-Québec Production (HQP) est en
charge de l'exploitation du parc hydroélectrique et de la planification de la production d'énergie.
Les prévisions d'apports sont produites à HQP et servent à (i) optimiser la production d'énergie;
(ii) protéger le public contre les inondations; (iii) assurer la sécurité des ouvrages; (iv) garantir
l'approvisionnement énergétique de la province et (v) satisfaire les demandes du milieu pour
l'usage des rivières et des réservoirs. Le processus opérationnel quotidien à HQP pour générer
les prévisions d'apports se déroule comme suit: (1) une prévision météo court terme (variant de
4 à 9 jours) est produite par un météorologue; (2) l'équipe de contrôle de la qualité des données
hydroélectriques effectue la validation des données des stations de mesure d'Hydro-Québec et
de ses partenaires; (3) enfin, les prévisionnistes d'apports génèrent une prévision d'apports
probabilistes pour le court terme (1 à 10 jours), moyen terme (200 jours) et long terme (108
semaines). La qualité des prévisions d'apports probabilistes à HQP est principalement limitée
par (i) la densité du réseau de stations hydrométéorologiques; (ii) la qualité de l'historique
météo-apports; (iii) l'incertitude liée à la modélisation hydrologique; (iv) la prise en compte de
l'incertitude de la prévision court terme du météorologue et (v) l'impact des changements
climatiques. Les projets en cours à HQP visant à améliorer les prévisions d'apports probabilistes
sont (i) l'ajout de stations météorologiques et hydrométriques; (ii) la révision de l'historique
météo-apports; (iii) l'amélioration du modèle hydrologique HSAMI et l'intégration du modèle
hydrologique HYDROTEL; (iv) l'intégration de l'incertitude des modèles de prévisions
météorologiques et (v) l'intégration de l'impact des changements climatiques.
10
9h40 – 10h05 Modeling rainfall-runoff models uncertainty to improve probabilistic forecasts
quality
M. Courbariaux, E. Parent, A.-C. Favre, P. Barbillon
Probabilistic forecasts aim at accounting for uncertainty by producing a predictive distribution of
the quantity of interest to be forecasted instead of a single best guess estimate. With regard to
river flow forecasts, uncertainty is mainly due (a) to the unknown future rainfalls and
temperatures, (b) to the possible inadequacy of the deterministic model mimicking the rainfallrunoff transformation. The first source of uncertainty can nowadays be taken into account using
ensemble forecasts as inputs to the rainfall-runoff model (RRM). However, the second source of
uncertainty due to the possible RRM misrepresentation remains. A simple way to integrate it
consists in adjusting the forecast's density as much as necessary to get a prediction consistent
with the observations. This step is called post-processing. Our work focuses on series of river
flow forecasts routinely issued at EDF (Electricity of France). We aim at reducing the sharpness
loss in the post-processing step while guaranteeing point-wise and temporal consistency. To do
so, we write a joint model on the RRM errors along the whole trajectory to be predicted. Pointwise and temporal consistency are then obtained relying on a Bayesian approach. We also
establish a classification of time periods according to their uncertainty level. The set of
predictors includes inputs, state variables, outputs' sensibility to parameters choice, and
observed errors of the RRM in simulation mode. Conditioning on that classification should result
in simpler RRM models of errors and should help reduce sharpness loss.
11
Session 5: Reservoir Management and Inflow forecast II
Chaired by Annie Poulin
10h30 – 10h55 État d'avancement des activités de prévision hydrologique en support à la
gestion des barrages publics du Québec
Richard Turcotte, Julie Lafleur
Centre d'expertise hydrique du Québec
Le Centre d'expertise hydrique du Québec (CEHQ) exploite 761 barrages dont une centaine
pouvant nécessiter des opérations. Parmi les barrages opérables, 40 font l’objet d’une gestion
en temps réel. La gestion des barrages des systèmes hydriques du lac Kénogami, du Haut-SaintFrançois et de la rivière du Lièvre présente des exemples décrivant les principaux défis de
gestion du CEHQ. Depuis la fin des années 90, le CEHQ opère un système de prévision pour
l’aide à la gestion des barrages qui s'appui sur les observations du réseau météorologique
coopératif du Québec, du réseau nivométrique et du réseau hydrométrique du Québec, sur des
prévisions météorologiques déterministes d'Environnement Canada et sur le modèle
hydrologique Hydrotel. Des outils de simulation des impacts de décision de gestion sur les
niveaux et les débits sont aussi disponibles. Depuis 2013, le mandat du CEHQ s’étend aussi à la
réalisation et à la diffusion sur l’Internet de prévisions pour des stations d’intérêt en sécurité
publique non influencées par l’opération des barrages. La synergie entre les besoins de la
prévision en sécurité publique et en gestion des barrages a mené à l’accroissement du nombre
et de la formation des prévisionnistes et a enclenché un cycle d’amélioration de
l’environnement informatique. Le CEHQ s'intéresse à la prévision d'ensemble et à des outils
d'optimisation stochastiques de la gestion bien qu’aucun produit opérationnel ne soit disponible
actuellement. Pour la prévision d'ensemble, le CEHQ s'est associé à différents partenaires
(Université Laval et Environnement Canada) pour obtenir des réponses à des questions de
recherche préalables au passage aux opérations. Les principaux obstacles à l’utilisation de la
prévision d’ensemble sont dorénavant liés à des défis technologiques et informatiques. En
parallèle, des outils d'optimisation complémentaires au système de prévision ont émergé au
cours des années suite à des projets d’étude ad hoc. Une revue de ces outils sera présentée
brièvement en mettant en évidence les défis à relever pour en augmenter la pertinence et
l'usage.
12
10h55 – 11h20 Des outils d'aide à la gestion de réservoirs qui valorisent l'information
hydrométéorologique historique
Charles Poirier et Martin-Pierre Lavigne
Pour la gestion des barrages publics, le CEHQ produit des prévisions hydrologiques à court
terme avec le modèle HYDROTEL. En hiver, l’évolution du couvert nival simulée est corrigée à
posteriori par une interpolation de données nivométriques à une fréquence généralement
bihebdomadaire (Turcotte et al., 2007). Cette méthodologie a été reprise en mode étude pour
produire une banque historique d’ÉEN aux 24h sur la majeure partie du territoire au sud 50ième
parallèle (Poirier et al., 2014 – à préciser).Parallèlement à ces travaux, le CEHQ produit
également sur demande des prévisions hydrologiques à moyen terme, en recourant à des
scénarios probabilistes historiques (ESP – extended streamflow predictions). Il est classique pour
ce faire d’utiliser des scénarios météorologiques, constitués généralement de la précipitation P
et la température T. Sur un territoire enneigé, cette pratique revient à laisser le modèle
hydrologique simuler lui-même l’évolution du couvert nival à partir d’un état initial donné. Or,
les résultats d’analyses récentes indiquent que l’évolution de l’ÉEN simulé par HYDROTEL
seulement à partir de P et T historiques peut différer significativement de l’évolution de l’ÉEN
simulé avec P, T et les données nivométriques historiques, lorsque l’horizon de simulation
s’étend au-delà d’une période relativement courte au printemps. Ces résultats mènent à
quelques constats. Premièrement, le caractère quasi-indispensable des données nivométriques
ressort nettement. Ensuite, la période relativement courte au-delà de laquelle HYDROTEL dérive
significativement dans ses estimations d’ÉEN au printemps incite à la prudence en regard de
l’usage de ce modèle pour réaliser des ESP. Enfin, pour le CEHQ, il s’avère pertinent d’amorcer
une recherche visant des solutions alternatives au modèle de fonte d’HYDROTEL dans sa forme
actuelle. Dans ce contexte, le CEHQ a amorcé le développement d’outils d’aide à la décision
valorisant l’information hydrométéorologique historique, et particulièrement la neige. La
présentation exposera un aperçu de la méthodologie de recherche d’états de neige voisins et la
déclinaison des outils d’analyse qui en découlent.
13
11h20 – 11h45 Post-processing of multi-model hydrological forecasts for the Baskatong
catchment
Fabian-Tito Arandia Martinez, Marie-Amélie Boucher, Jocelyn Gaudet and
Maria-Helena Ramos
This study focuses on the Baskatong catchment exploited by Hydro-Québec for hydro-power
generation. Operational streamflow forecasts used by Hydro-Québec are compared to a new
mul-ti-model ensemble forecasting system. This new system involves meteorological ensemble
fore-casts from three meteorological agencies, obtained though the TIGGE database. The new
multi-model ensemble forecasting system explicitly accounts for uncertainties related to the
choice of an atmospheric model and also to those related to the initial condition of the
atmosphere. The uncertainty linked to the choice of the hydrologic model and to the
parameterization of those models are still uncultured and therefore, the raw forecasts are
initially under-dispersive. Conse-quently, there is a need for statistical correction of the bias and
spread in the raw forecasts. Here we present and compare the results for four possible
methods for post-processing: a regression method, the best member method, the weighted
kernel dressing method of Fortin et al. (2006) and the Bayesian Model Averaging method of
Raftery et al. (2005).
11h45 – 12h10 Caractérisation de l'influence des conditions initiales sur un ensemble
multimodèle
Antoine Thiboult, François Anctil
La prévision hydrologique d'ensemble offre la possibilité d'apprécier dynamiquement
l'incertitude liée à la prévision. Cependant, bien qu'une approche multimodèle permette de
quantifier et réduire l'incertitude structurelle et paramétrique, l'incertitude sur les conditions
initiales est fréquemment sous-estimée. L'utilisation de techniques d'assimilation telle que le
filtre d'ensemble de Kalman (EnKF) permet de coupler les observations et la prévision afin
d'estimer plus précisément les conditions initiales. Un ensemble multimodèle est composé à
partir de 20 modèles choisis pour leur diversité structurelle et conceptuelle. Grâce à une étude
approfondie, une implémentation optimale d'EnKF est identifiée pour chaque modèle ce qui
permet la constitution d'un ensemble multimodèle présentant une résolution et une fiabilité
accrue, facilitant ainsi une potentielle prise de décision.
14
Plenary session 3
Chaired by Marco Latraverse
13h30 – 14h15 Ensemble forecasting at Environment Canada: living with uncertain weather
and enjoying it
F. Anctil1, V. Fortin2, N. Gagnon3 and S. Bélair2
1
Département de génie civil et génie des eaux, Université Laval, Québec, Canada
2
Meteorological Research Division, Environment Canada, Dorval, Québec, Canada
3
Meteorological Service of Canada, Environment Canada, Dorval, Québec, Canada
Over the last few years, ensemble predictions systems (EPS) have become more and more
important at Environment Canada. From one hour to the seasonal time scale, the Canadian
Meteorological Centre (CMC) is now issuing ensembles of atmospheric forecasts that consist
typically twenty scenarios. The horizontal resolution ranges from 15 km (regional EPS) to 300 km
(seasonal coupled atmospheric-ocean systems CANSIPS). Operational ensemble data
assimilation is performed using the ensemble Kalman filter (EnKF) approach. EnKF is used for
both atmospheric data assimilation and land data assimilation (soil moisture, soil temperature
and snow on the ground). Recently, the range of the global EPS (GEPS) was extended to reach 32
days. An exciting new feature of the latest GEPS is the fact that an operational reforecast is
produced routinely with the forecast. This procedure is generating historical forecasts (from
1995) with the current operational GEPS but with a simpler initialization and a smaller ensemble
size. The objective is to allow CMC as well as end users to improve the post-processing of model
forecasts by having access to a comprehensive database of past forecasts in order to assess the
skill and reliability of the current operational system, which can lead to more optimal decisions.
The seasonal system CANSIPS is also provided with a hindcast of 30 years. A summary of the
various configurations of Environment Canada’s ensemble prediction systems is given, as well as
an evaluation of the forecast skill against observations of temperature, precipitation, snowpack
and streamflow.
15
Session 6: Hydrological ensemble predictions and Inflow Forecast III
Chaired by François Anctil
14h15 – 14h40 Evaluating the Efficacy of Enhancements to BC Hydro's Operational Hydrologic
Ensemble Forecast System
Adam Gobena
BC Hydro
Since its first implementation in 1980 for operations planning of the Kinbasket Reservoir on the
Columbia River system, BC Hydro's ESP based long-range water supply forecast system has
undergone several enhancements with the aim of improving forecast accuracy and reliability.
These enhancements include snow data assimilation, simulation bias correction, considerations
of parameter uncertainty (via multi-parameter set modeling) and input forcing uncertainty (via
augmentation with synthetic weather sequences). However, little has been done to date to
quantify the value added by these enhancements due to data and resource limitations. This
presentation will discuss the efficacy of some of the enhancements by comparing the
performance of archived operational forecasts and limited hindcast experiments.
14h40 – 15h05 Development of a Flood Forecasting Model for the St. John River Watershed
using the Raven Modelling Framework
Hossein Babaei, Wayne Jenkinson, Alex Mineault-Guitard
The Inland Water Resources group at the National Research Council of Canada has recently
developed a flood forecasting model for the St. John River watershed using the Raven
hydrological modelling approach. Raven is a semi-distributed, physically-based model, which is
being used by Lake of the Woods Control Board, BC Hydro, and TransAlta for streamflow
forecasting purposes. The new model provides an alternative tool for streamflow forecasting for
the New Brunswick (NB) Department of the Environment, which is currently using Streamflow
Synthesis and Reservoir Regulation model. This presentation will focus on the details of the
modelling approach including identification of hydrological response units, flow routing
algorithms, model calibration, and performance evaluation. The new model is currently being
evaluated in both on- and off-line modes at the NB flood forecasting center located in
Fredericton.
16
Session 7: Reservoir Management and Inflow forecast III
Chaired by Robert Leconte
15h30 – 15h55 Utilisation des prévisions d’ensemble pour améliorer les règles de gestion du
réservoir de la rivière Nechako, C.-B., établies à l’aide de la programmation
dynamique stochastique
Alexandre Martin1, Pascal Côté2 et Robert Leconte1
1
Département de génie civil, Université de Sherbrooke
Énergie Électrique, Rio Tinto Alcan
2
Pour maximiser la production électrique de la centrale Kemano tout en respectant les
contraintes de gestion de la rivière Nechako, Rio Tinto Alcan a développé un algorithme de
programmation dynamique stochastique (PDS) qui optimise les règles de gestion du réservoir en
fonction d’une séquence historique d’apports et d’une variable hydrologique. Afin d’améliorer la
performance des règles, ce projet étudie l’utilisation de prévisions d’ensemble pour caractériser
la variable hydrologique et compare les résultats à ceux d’un banc d’essais en horizon roulant
qui effectue une mise à jour des règles à chaque trois jours en utilisant des prévisions
d’ensemble sur les apports.
15h55 – 16h20 Un outil de prévisions et d'optimisation de la gestion des eaux de sites miniers
Simon Dagher, Zoubir Bouazza
AMEC
Amec, en collaboration avec CliffsNR ont développé un outil dédié à la gestion des eaux de sites
miniers. L’outil de simulation en temps différé et d’aide à la décision intègre des modules
capables de générer des scénarios météorologiques et de prévoir les apports hydriques aux
retenues d’eau. Des conditions météorologiques normales, sèches, humides ou complètement
stochastiques peuvent être générées moyennant des algorithmes simples. Quelques scénarios
de gestion seront présentés pour démontrer l’utilité de l’outil pour le choix des consignes
d’opération et pour la planification des travaux de construction au site minier étudié. Les
scénarios couvrent la période de crue printanière.
17
16h20 – 16h45 The effects of Ensemble streamflow prediction methods on hydropower
generation
Richard Arsenault1, Jie Chen1; François Brissette1 and Pascal Côté2
1
École de technologie Supérieure, Montréal
Quebec Power Operation, Rio Tinto Alcan
2
A testbed was created to simulate the hydroelectric power generation on the SaguenayLac-St-Jean water resources system under various Ensemble Streamflow Prediction
(ESP) scenarios. Different methods to generate ESPs were compared and fed to the
testbed, which uses stochastic dynamic programming to generate reservoir
management rules based on the predicted and historic inflows. The simulator then uses
the rules to estimate the hydroelectric output based on the observed data. The results
show that the selection of an ESP method can significantly increase power generation
and can assimilate trends in non-stationary conditions to further increase output.
18
Day #3
Plenary session 4
Chaired by Amaury Tilmant
8h30 – 9h15
On the direct use of hydroclimatic information to better inform water
reservoir operation
Andrea Castelletti
Dipartimento di Elettronica e Informazione, Politecnico di Milano, Milan, Italy
It is generally agreed that more information translates into better decisions. For
instance, the availability of inflow predictions can improve reservoir operation; soil
moisture data can be exploited to increase irrigation efficiency; etc. However, beyond
this general statement, many theoretical and practical questions remain open. Provided
that not all information sources are equally relevant, how does their value depend on
the physical features of the water system and on the purposes of the system operation?
What is the minimum lead time needed for anticipatory management to be effective? Is
the data-predictions-decision paradigm truly effective or would it be better to directly
use hydroclimatic data to take optimal decisions, skipping the intermediate step of
hydrological forecasting? A procedural approach addressing these issues was recently
developed and applied in different hydrological contexts. Numerical results on the
development of the procedure for re-designing the operation of a regulated lake in a
snow-dominated catchment in northern Italy will be presented.
19
Session 7: Reservoir Optimization II
Chaired by Pascal Côté
9h15 – 9h40
Use of hydrologic data to optimize Canadian dams on the Columbia River
Tim Blair
BC Hydro
The Columbia River originates in Canada, where most of the storage resides, and has
most of the power generation downstream in the U.S. The operation of Columbia basin
storage reservoirs is coordinated under the Columbia River Treaty, with the dual
purpose of providing flood control while maximizing hydropower generation. There is
flexibility for each country to re-optimize its own hydroelectric generation beyond the
requirements dictated by the treaty. This talk will cover how hydrologic data is
incorporated into Treaty planning and implementation, and how BC Hydro incorporates
hydrologic forecast uncertainty into the optimization of the Mica/Arrow reservoir
operations in Canada.
9h40 – 10h15 Gestion de réservoir par programmation dynamique
Michel Denault, Olivier-Meunier, Simonato
HEC
We investigate the optimum control of a hydro storage power plant, in the presence of
both exogenous (control-independent) stochastic state variables and endogenous
(control-dependent) state variables. Our solution approach relies on simulations and
regressions with respect to the state variables, but also grafts the endogenous state
variable into the simulation paths. That is, unlike most other simulation approaches
found in the literature, no discretization of the endogenous variable is required. The
approach is meant to handle several stochastic variables, offers a high level of ﬂexibility
in their modeling, and should be at its best in non time-homogenous cases, when the
optimal policy structure changes with time. We provide numerical results for a storagebased hydropower application, where the exogenous variable is the stochastic spot
price of power, and the endogenous variable is the water level in the reservoir.
20
10h15 – 10h30 The Role of Hydrologic Information in Stochastic Dynamic Programming: a
Case Study of Kemano Hydropower system in British Columbia
Quentin Desreumaux1, Pascal Côté2, Robert Leconte1
1
Department of civil engineeing, Sherbrooke University
Quebec Power Operation, Rio Tinto Alcan
2
We present the value of various hydrological variables in stochastic dynamic
programming (SDP) to solve the optimization problem of managing a hydropower
system with a study case. The system studied here, namely the Kemano hydroelecric
system located in British Columbia, Canada, is subject to large streamflow volumes due
to significant snow cover during winter. Results indicate that for the system herein,
maximum snow water equivalent (SWE), which is the highest levels of SWE observed
from the start of winter to the current decision period, prove to be best for effective,
safe management, compared to a Markov or order p autoregressive models .
11h00 – 12h00 Panel
21

Download Report