Flood Forecasting in large basins

Transcription

1 Flood Forecasting in large basins Prof. Dr. Carlos E. M. Tucci IPH -Institute of Hydraulic Research UFRGS - Federal University of Rio Grande do Sul

2 Contents Main Issues on Floods Flow forecast : basic concepts Short term Long term Case studies : Uruguay River Basin and São Francisco River Basin

3 Main Issues Urban developments: occupation of flood plains and local urbanization; Planning and operation of hydraulic works such as reservoirs Energy planning: for countries highly dependent on hydropower. Water uses such as navigation and agriculture Environment conservation.

4 Flood Plain and Hazard areas Without urban planning for the flood plains, population moves to it after some years of low floods. When a high level flood occurs the cost of damage cost is sky high and creates numerous economic and social problems.

5 Climate and variability issues Change in the hydrology series trends and the impact of water uses and society Pantanal: Paraguay River : society impact Change of 2,5 m in the mean annual flood level in Pantanal; change in the mean flood areas from km2 to km2 Annuall Flood levels, m Years A property flooded during 20% of the time in the 60 s is now flooded 97% of the time

6 WET SEASON DRY SEASON

7

8 Paraná River and Brazilian Energy system Mean Flow increase after 70 s in Paraná River and most of neighbors basins 60% of Brazilian energy is produced in the basin 90% of Brazilian energy production is based on hydropower ; Thermal plants has been constructed in order to support the lack of water or storage of the system The system is highly dependent of the next semester forecasting (mainly wet months) River Section Before Increase % Parana River at Jupiá 5,852 (+) 6,969 19,1 R. Paranapanema at Rosana 1,057 (+) 1,545 46,2 R. Paraná at São José 6,900 (+) 8,520 23,3 R. Paraná at Guaira 8,620 (+) 11,560 34,1 R. Paraná r at Posadas 11,600 (*) 14,255 22,9 R. Paraná at Corrientes 15,265 19,510 27,8

9 Flood Management Flood Evaluation : Prediction = statistical risk based on stationary series Flood forecasting = short and long term based on known conditions Flood measures: structural and nonstructural where flood forecasting is one of the measures.

10 Benefit of flow forecasting water resource development and environment conservation has been planned based on historical and stationary (assuming) flow series; the increase of the uncertainty of the projects due to climate change (natural and antrophogenic (non-stationary conditions); Flow forecasting can decrease the uncertainty of these projects

11 Needs of forecasting in large basin Usually hydropowers are in large basins > 2000 km 2 and mainly > km2 Systems highly depend on climate variability of long term Its complemented by thermal energy with higher cost Forecasting is the base for optimization of such system

12 Types of Flow forecast Lead Time: Short term : hours to a few days; long -term: up to 9 months. Operational: continuous - when the flow is forecast all year long for navigation, hydropower, supply, etc event - the forecast is done after the warning condition until the risk is below the warning situation

13 Input of the forecast 1. QPF Quantitative Precipitation forecast + Rainfall - runoff model; Evaluation: Regional meteorological model, Satellite, Radar and Telemetric network 2. Upstream level or discharge 3. Combined information

14 Rainfall - Runoff input = Rainfall short peak time Flood routing input = flow or level slow flow in large basins Rainfall - Runoff and routing inputs = rainfall and upstream flow

15 Hydrological forecasting models Rainfall- runoff hydrological models: the input is rainfall and the models simulate the basin system routing models: flow or level upstream are the inputs and the models simulate only the river system The models can be empirical or conceptual Conceptual can be lumped or distributed;

16 Forecasting without modeling updates Historical events Input Pt, Qt Pt or Qt θi Model Fitting tools Fitting off - line Model Forecast Parameters θi Flow Q(t+dt) Forecasting without parameters or updated variables

17 Forecasting with modeling updates Inputs Pt, Qt θi Model F[ Pt,Qt, θi] Simulation mode Output Q(t+dt) Time step change Update of parameters or state variables Adaptive mode Record Q(t+dt)

18 Empirical models Usually tools such as: ARMA, ARIMA, multiple regressions, neural waves, etc they do not have a conceptual relationship with the simulated system; the model parameters or state variable can easily be updated; good results for short lead time They use the knowledge from the past to forecast the future

19 Conceptual model Lumped Distributed sub- basins Distributed by grid

20 Lumped models do not take into account the space variability of rainfall, parameters and basin characteristics; easy to fit and update due to the small number of parameters; used mainly for small basins with uniform rainfall and physical characteristics

21 Distributed models take into account non-uniform rainfall and physical conditions; difficulties in parameters fitting and updating; require flow routing and floodplain evaluations for large basins usually can simulate long lead time as compared to empirical models

22 Short - term forecasts Flash floods - usually are events which occur inside of 6 hours of the causative rainfall; Streamflow forecast: forecast lead time from 6 hours to a few days. It can be done for floods events or for continuous flow forecast

23 Flash Floods Short time of basin concentration which requires a small lead time = use of a rainfall - runoff model; basin characteristics : - urban drainage in large cities - uphill rural basins highly dependent of the spatial and temporal rainfall forecast (QPF - quantitative precipitation forecast)

24 Streamflow Forecasting it can use a rainfall-runoff model or a routing model depending on the lead time required; medium sized rivers; less dependent of QPF; main uses: flood alert; operation of a dam for hydropower, flood control and navigation; hydraulics works security,etc.

25 Long Term Forecasting reasonable results for systems with a long memory and well known seasonal behavior; Is highly dependent of rainfall forecast by climate model or other variable such as ocean temperature; main uses: power systems market (the forecast is a commodity); agriculture production (soil moisture ); navigation in river systems; environment; water resource conflicts; and flood management.

26 Warning Systems Institutional aspects : for most developing countries, understanding institutional limitations, such as capacity building, investments limitations and operations maintenance conditions, is the main issue; evaluation of what can be done with human and technical resources in each situation; understand the limits of forecasting

27 Case Studies Uruguay River Basin (Brazilian area of ~ km 2 ) Hydrologic characteristics: wet region of about 1,600 mm annual mean rainfall; small basin memory (small groundwater storage), it does not shows climate seasonality. The floods and drought could be in any month of the year; small time of concentration. São Francisco River Basin: km 2 with wet and semi-arid regions, mean annual rainfall from 1,500 mm to 400 mm; large time of concentration, long memory and seasonality well define The outputs were developed for a group of researchers from IPH, USP and CPTEC/INPE

28 Localization

29 Models Flow forecasting long - term Uruguay flow statistics based on the historical flow series ; forecasting based on empirical relationship with Pacific and Atlantic temperature; climate model + hydrological model

30 Uruguay River Basin Modelo Numérico do Terreno from 1400 to 150 m < >=1449 RIO IBICUI G RIO URUGUAI U R U G U A I 30 Km 0 RIO URUGUAI RIO PEPERI-GUACU RIO CHAPECO RIO IJUI RIO PIRATINI RIO ICAMAQUA A R G E N T I N A E S C A L A Km D RIO DO PEIXE S A N T A C A T A R I N A E C B RIO DA V'ARZEA RIO CANOAS RIO PELOTAS F A R I O G R A N D E D O S U L

31 Hydrologic model Distributed model with grid of 0.1 x 0.1 deg. Model fitted through multi-objective optimization method Uruguay River = with low memory and lack of seasonal variability and high interannual variability about km2

32 Grid of 0.1 degree Modelo Numérico do Terreno < >=1449

33 Model Blocks Água Floresta + Pastagem Agricultura + Floresta Floresta Floresta em solo raso Pastagem Pastagem em solos rasos Agricultura + Pastagem

34 Rainfall gages raingages

35 Fitting To start, we used the same parameters from the neighboring basin 5 flow gages were used for fitting and 12 gages for verification the fitting was done on hydrographs and flow duration curves

36 Fitting Sample calculado observado Passo Caxambu km /01/88 01/02/88 01/03/88 01/04/88 01/05/88 01/06/88 01/07/88 01/08/88 01/09/88 01/10/88 01/11/88 01/12/88 Vazão (m3/s)

37 Fitting the flow duration curve calculado observado Vazão (m3/s) Tempo de permanência (%)

38 Fitting period RNS = 0,77 RNSL = 0,74 DV = -4,1% RNS = 0,88 RNSL = 0,86 DV = 0,8% RNS = 0,70 RNSL = 0,75 DV = -19,1% RNS = 0,76 RNSL = 0,74 DV = -6,2% RNS = 0,86 RNSL = 0,80 DV = 2,3% Sem dados

39 CPTEC Climate model grid

40 Sistematic errors in the Rainfall prediction Recorded mean rainfall Simulated mean rainfall

41 Error of the rainfall forecast

42 Statistical Correction : correction forecasted Correção da previsão

43 FORECAST WITH 6 MONTHS OF LEAD TIME a a Ensemble flow forecast mean Statistical mean recorded

44 Mean of three month forecast

45 Forecasting

46 Reduction of the uncertainties

47 Standard error for monthly flow forecasts Climate model Climate model with Recorded With known without correction correction standard error rainfall Climate + hydrological model with rainfall correction decreased the error by 34%

48 Standard error for three months forecasts Climate model Climate model with Recorded With known without correction correction standard error rainfall Climate + hydrological model with rainfall correction decreased the error by 54%

49 Comments For a few events the downscale climate model was evaluated, but it did not improve the forecasts ; we could not find any serial correlation on the residuals of the forecasts and recorded flow; Basin Characteristics and climate forecast for the region were the main factor of the positive results Is it possible to forecast soil humidity from the hydrologic model?

50 Long Term São Francisco Semi-arid

51 Characteristics hydropower: main source of energy of Northeast of the country; Great hydrologic and climate change along the basin: upstream basin supply most of downstream flow; In the semi-arid it is lowest social-economic development of the country.

52 Sobradinho Paulo Afonso Basin discretization and sub-basins Três Marias

53 Basin Characteristics

54 Fitting the hydrologic model Três Marias: Upstream dam at wet region Sobradinho

55 GCM Grid Model CPTEC/COLA; 200km grid

56 The model underestimate the regional rainfall in absolute values; It is not bad in relative values Rainfall Forecast m1 m2 m3 m4 m5 obs M OND 1997 NDJ DJF JFM 1998 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 1999 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2000 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2001 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2002 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2003 FMA Precipitação (mm/dia) m1 m2 m3 m4 m5 OBS M OND 1997 NDJ DJF JFM 1998 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 1999 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2000 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2001 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2002 FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF JFM 2003 FMA mm/dia

57 Evaluation of 9 years in a specific area 800 Chuva total mensal (mm) Observado Climatologia do modelo (9 membros) Climatologia média do modelo jan/82 jan/83 jan/84 jan/85 jan/86 jan/87 jan/88 jan/89 jan/90 jan/91

58 Statistic correction Observado Climatologia do modelo Chuva mensal (mm) Valor previsto Valor corrigido Probabilidade

59 Rainfall forecast and recorded after Sample of a month correction

60 Rainfall forecast for the sub-basins Três Marias lead time: 1 month Sobradinho lead time 3 months 600 Três Marias - 1 mês 800 Sobradinho - 3 meses 500 Observado "Ensemble" (5 membros) Previsão média Observado Previsão média Chuva mensal (mm) Chuva acumulada (mm) out/97 abr/98 out/98 abr/99 out/99 abr/00 out/00 abr/01 out/01 abr/02 out/02 0 out/97 abr/98 out/98 abr/99 out/99 abr/00 out/00 abr/01 out/01 abr/02 out/02 abr/03

61 Flow forecast Três Marias 1 month 2 months

62 Models Comparison: TRÊS MARIAS RED: Hydrologic model and record rainfall GREEN: Climatic + hydrologic model YELLOW: stochastic model used by Eletric System; BLUE: persistence PURPLE: Empírico basedontsm

63 Models Comparison: Sobradinho RED: Hydrologic model and record rainfall GREEN: Climatic + hydrologic model YELLOW: stochastic model used by Electric System; BLUE: persistence PURPLE: Empiric based on TSM

64 Downscale models: ETA and BRAMS Três Marias Sobradinho Part of the total series: wet periods

65 Short term Forecast in São Francisco ETA 40 + Hydrologic Model Lead time : two until four weeks

66 Rainfall forecasted lead of 1 week Três Marias Sobradinho Accumulated values

67 Forecast errors TRÊS MARIAS SOBRADINHO

68 Conclusions Long Term or seasonal forecasting is an important area of development as decision support tool on water management. Its is highly dependent on interdisciplinary research and development of climate, hydrology, system control and economic evaluations. The improvement of the forecast is highly dependent in the skill of the climate models and other tools; There is a need for improvement in rainfall forecast for long term based on downscale models; Forecast services requires an understanding of institutional issues such as capacity building, integration of services among entities, etc.; The lack of data and hydrometeorological real time measurement in developing countries will need better satellite rainfall forecast;