Assessing the impacts of climate change on Atbara flows using bias-corrected GCM scenarios SIGMED and MEDFRIEND International Scientific Workshop Relations man / environment and sediment transport: a spatial approach Algeria 7&8 June 2011 Mohamed Elshamy
Outline 1. Uncertainty Cascade 2. The Nile Basin & Previous CC studies 3. Study Area & Methodology 4. Results 5. Conclusions
Uncertainty Cascade Emissions Concentrations Radiative Forcing Global Climate Models Regional Details (Downscaling) Impact Models (e.g. Hydrology) Observations
The Nile Basin Large area (2.9 x 10 6 km 2 ) Low specific discharge Spans several climate regions Variable topography High runoff variability High Sensitivity to Climate
Lake Nasser Flood & Drought Control Project (2008) Previous Studies (1) –6 Transient scenarios (3 GCMs x 2 Emission Scenarios) –Statistically downscaled using a spatio-temporal weather generator –Changes at Dongola from Elshamy, M.E., Sayed, M.A.-A. and Badwy, B., Impacts of climate change on Nile flows at Dongola using statistically downscaled GCM scenarios. Nile Water Science & Engineering Magazine 2: 1-14
Previous Studies (2) Elshamy et al. (2009) –17 GCMs x A1B scenario –Statistically downscaled using Bias Correction Method –Blue Nile Flow Changes: -60% to +45% Elshamy, M.E., Seierstad, I.A. and Sorteberg, A., Impacts of climate change on Blue Nile flows using bias-corrected GCM scenarios. Hydrol. Earth Syst. Sci., 13(5):
The Atbara Basin Climate: Semi-Arid/Arid Area: km 2 Mean Rainfall: 500 mm/yr Mean PET: 1730 mm/yr Mean Flow: 8.5 BCM/yr ( ) Very Sensitive to Climate
The Atbara Basin
Sensitivity to Climate
Methodology Coarse Daily GCM Rainfall Coarse Daily GCM Rainfall Bias Correction Downscaling Fine-Scale Daily Rainfall Hydrological Model Flow at Atbara 17 GCMs x 1 Scenario Flow at Atbara 17 GCMs x 1 Scenario Compatible PET Scenarios
Bias Correction Downscaling Bias correction for downscaling rainfall (based on fitting the gamma distribution to daily rainfall) Simple bias correction for PET (ratio) NFS & HBV for hydrological modeling An ensemble approach (17 GCMs – A1B) Baseline , Future , Daily rainfall data & Monthly PET data
Rain gauge Data Satellite Images Rainfall Estimation Models Rainfall Estimates Hydrological Models Hydrological Models Simulation and Extended Stream Flow Prediction (ESP) Simulation and Extended Stream Flow Prediction (ESP) Water Balance Hill Slope Routing Swamp Lake Historical Climate Historical Climate GIS Nile Forecast System (NFS)
HBV Hydrological Model Precipitation on lakes Lake evaporation EA=EPOT P From soil moisture routine dUZ UPPER ZONE LOWER ZONE KUZ1 KUZ UZ1 UZ Q11 Q10 Q2 LZ KLZ PERC Lake area in % (LA) Runoff, Q KLZ KUZ KUZ1 UZ PERC : Time constant, lower zone, 1/t : Time constant, upper zone, 1/t : Water level, upper zone : Percolation to lower zone, mm/day Q Q10 Q11 Q2 = Q10 + Q11 + Q2 = MIN (UZ, UZ1)*KUZ = MAX (0, (UZ - UZ1)*KUZ1)) = KLZ*LZ PARAMETERS IN THE RESPONSE FUNCTION : RUNOFF COMPONENTS : UZ = dUZ - PERC - Q11 - Q10 Water balance equation, upper zone: LZ = PERC + (P – EPOT)*LA/100 - Q2 Water balance equation, lower zone: LZ: Water level, lower zone UZ1: Threshold for quick flow, mm
Model Performance Monthly NSE = 0.69 & 0.83 for NFS & HBV respectively
The GLUE Framework GLUE: Generalized Likelihood Uncertainty EstimationGLUE: Generalized Likelihood Uncertainty Estimation GLUE rejects the concept of a single optimal model and parameter setGLUE rejects the concept of a single optimal model and parameter set Assumes all model structures and parameter sets have a likelihood of being acceptedAssumes all model structures and parameter sets have a likelihood of being accepted Likelyhood depends performance as measured by a selected criteriaLikelyhood depends performance as measured by a selected criteria
Results: Rainfall Changes
Results: PET Changes
Results: Flow Changes - NFS
Results: Flow Changes - HBV
Conclusions GCMs agree on Temperature rise (2-5.3 °C) leading to 3-17% increase in PET GCMs disagree on precipitation changes (-36% to +39%) High Sensitivity of Basin leads to extreme flow change ranges: -76% to +97% from both NFS & HBV Ensemble mean flow is reduced by 25% & 6% for NFS & HBV respectively Hydrological models add another uncertainty GLUE provides a framework to propagate the uncertainty from scenarios to impacts Probabilities are now attached to the uncertainty bounds Small sample size lead to small difference between GLUE bounds and max/min bounds