GENERATING STREAMFLOW FORECASTS FOR THE SOUTHEASTERN/SOUTHERN
BRAZILIAN HYDROPOWER PRODUCTION USING EUROBRISA´S INTEGRATED
RAINFALL CLIMATE FORECASTS
Alexandre K. GuetterUniversity of Parana Brasil
EUROBRISA Final Workshop, Barcelona, 13-16/Dec/2010
PROBLEM DESCRIPTIONStreamflow Monthly ForecastingFor 3-month Forecasting Horizon
At 68 specific basins (average basin scale ~ 40.000 km2)
• Brazilian Energy Supply:Market Size: EU$100 billion/yr
Hydropower: 75%
Thermal: 15%
Nuclear: 2%
Import: 8%Brazil
NO- North
NE-Northeasten
SE – Southeast/ Midwest
SU - Southern
• Different hydrologic regimes grouped on a continental scale
• Installed Capacity
Southeastern/Midwest: 63%
Southern: 17%
Northeastern: 14%
Northern: 6%
• Assumptions About Seasonal Precipitation Forecasting Skill
High: Southern, Northern, Northeastern
Low: Southeastern
% of Brazil´s Energy Production
Southeastern (Parana): 63%
EU$50 billion/yr market share
Southern (Uruguai): 17%
EU$13 billion/yr market share
STUDY AREA: PARANA BASIN + URUGUAI BASIN
HYDROPOWER ENERGY: PHYSICAL CONCEPTS
Q = FLOW THROUGH
TURBINES (m3/s)
H = HEAD (m)DAM AND
INTAKE
RESERVOIR WATER LEVEL
POWER GRID CONNECTION
TURBINE
GENERATOR
CONDUIT
POWERHOUSE
DOWNSTREAM
WATER LEVEL
DRAFT TUBE
3
* [KWh]
time [h]
Q = flow [m /s] large variation
H = head [m] assumed constant
ENERGY t gQH
t
IF WE COULD PREDICT STREAMFLOW => THEN WE CAN PREDICT THE ENERGY PRODUCTION FOR EACH POWERPLANT
HYDROPOWER SYSTEM COMPLEXITY
Interconnected Hydropower System
Cascade – Centered in the Southeast51 reservoirs17 throughflowEquivalent Reservoir Concept
• Current operational energy programming tools:– Equal probability streamflow scenarios based on
synthetic time series generation for each basin preserving spatial correlation through the statistics for 68 locations
• We propose:– Monthly streamflow forecasting using EUROBRISA
´s integrated rainfall forecasting as input data for a basin calibrated rainfall-runoff-routing model + raingauge data for basin mean-areal precipitation
OBJECTIVE
• Check whether the state of the science for seasonal climate forecasting is already useful for hydropower optimization programming
ACTIVITIES• Hydrologic model calibration for 8 large basins
within the Southern/Southeastern areas;• Calibration of the hydrologic model updating
parameters (we neglected updating);• 1987-2001 EUROBRISA´s rainfall hindcast as
input data for the hydrologic model;• Evaluation of Indices for Streamflow
Forecasting Accuracy ;
1987-2001 EUROBRISA´s HINDCAST
• Four Climate Dynamic Models– System 3 (ECMWF)– GloSea (UKMO)– Méteo-France– CPTEC
• One Empirical Model (CPTEC)• Integrated product – 5 Model Bayesian
Combination
STUDY AREA: Southern/Southeasten Brazil
2.000.000 km2
Study Area
• 5 large basins in Southern/Southeastern Brazil• 6 reservoirs sampling each one of the large basins + 2 on the
Parana River
─ Furnas (Grande River)─ Emborcação (Paranaiba River)─ Foz do Areia (Iguacu River)─ Ilha Solteira (Parana River)─ Itaipu (Parana River─ Itá (Uruguai River)
Study Area
Study Area - Southeastern
Study Area - Southern
Data
• Raingauge at selected locations (monthly), instead of GPCP gridded rainfall;
• Naturalized streamflow series (monthly);• Potential Evapotranspiration;• EUROBRISA´s integrated rainfall forecasting
for 1987-2001;
tQPtW
PEWt
W
PEW
iSii
ii
i
)1(
21
21
001
Data Joint Consistency Analysis• Monthly surface water balance
• Input data should be stationary• Soil-water variability estimates
3R Hydrologic Model
RESULTS – Southeastern Brazil FURNAS
Naturalized Streamflow Annual CycleFurnas – wet: October-April ; dry: May-September
FURNAS – Interannual Variability
Observed Monthly Streamflow Variability
• Intercomparison between observed-forecasted basin mean-areal precipitation (1987 – 2001)
ERROR Furnas
Average (mm/mês) 0,0Standard Deviation (mm/mês) 33,6
Correlation Coefficient 0,92
Percentil Raingauge Forecasting ∆ (%)0.1 15.5 19.6 26.5
0.33 56.7 57 0.50.5 97.2 110.7 13.9
0.67 158.2 161.7 2.20.9 245.5 237 -3.5
Data Joint Consistency Analysis• Soil Water Intercomparison
Hydrologic Model Calibration
Hydrologic Model Calibration
Soil Water Variability
Fluxo Observado (mm mes-1)
Simulado (mm mes-1)
Chuva 1412
Evaporação Potencial 1049
Evaporação Real 857
Vazão 539 542
Escoamento Base 234
Escoamento Superficial 308
Recarga do Aqüífero 12
Ave Error <1%
Obs Vs. Forecasted Streamflow: ρ=0.93
1-Month Streamflow Forecasting
Statistic Qprev (1 month) Qprev (2 months) Qprev (3 months)
Average Observed Streamflow
45
Standard DevObs Streamflow
29
Average ForecastedStreamflow
42 43 45
Standard DeviationObs Streamflow
19 20 24
Ave (Pred-Obs) -2.1 -1.0 1.6
Sdev (Pred-Obs) 19.3 19.4 20.0
Correlation(Pred-Obs)
0.76 0.76 0.74
* Dados de vazão e desvio padrão em mm/mês
Streamflow Forecasting Statistics
2-mon forecasting
3-mon forecasting
Intercomparison perfect rainfall forecasting
Statistics – Perfect rainfall forecasting
Statistic Qprev (1 month) Qprev (2 months) Qprev (3 months)
Average Observed Streamflow
45
Standard DevObs Streamflow
27
Average ForecastedStreamflow
42 43 45
Standard DeviationObs Streamflow
19 20 24
Ave (Pred-Obs) -2.5 -1.3 1.4
Sdev (Pred-Obs) 14.2 15.7 16.6
Correlation(Pred-Obs)
0.86 0.81 0.80
• Correlation Conditioned on Oct-Apr Streamflow for the <20% and >80% of the empirical distribution => 60% hit rate
CONCLUSIONS:For Southeastern Brazil
(which is generally regarded as having low predictability)
• Streamflow forecasting was surprisingly accurate with regard to hydrograph phase and intensity;
• Streamflow forecasting identified whether the rainy season started at the expected month;
• Streamflow forecasting identified both wet and dry spells;
IN REGARD TO MODELLING:
• Raingauge rainfall (local data) should be used for model calibration and both simulation of past events and climatology for rainfall forecasting;
• Basin model calibration is necessary to achieve streamflow forecasting accuracy required for hydropower programming;
WHAT WE HAVE ALREADY ACHIEVED:
• Seasonal forecasting is very useful for the Southeasten Region (60% of Brazilian hydropower generation) – strong annual cycle;
• Seasonal forecasting is somewhat useful for the Southern Region (15% of Brazilian hydropower generation) – almost uniform annual cycle
WHAT WE PLAN TO ACHIEVE IN 2011
• Analysis for the Northeastern Region (14% of Brazilian hydropower generation) – strong annual cycle;
• Analysis for the Northern Region (6% of Brazilian hydropower generation) – stong annual cycle;
• Develop an aggregate energy model (Method of Natural Energy) to estimate economic value;