Climate change scenarios downscaling to bridge the gap between dynamical models and the end user: application for hydrometeorological impact studies in Spain Marco Turco 1, Maria del Carmen Llasat 1, Pere Quintana Seguí 2 http://gama.am.ub.es/ mturco@am.ub.es 1 GAMA (Meteorological Hazards Analysis Team), University of Barcelona, Spain. 2 Observatori de l Ebre (Universitat Ramon Llull - CSIC), Roquetes, Spain. http://www.obsebre.es/
Project type: National project Funding institution: Ministerio de Medio Ambiente y Medio Rural y Marino, Oficina Española del Cambio Climático Program: Spanish R+D Program 2008-2011 Objective: regional (20 km) and local (stations) scenarios for temperature (maximum and minimum) and precipitation for the XXI century, at a daily scale in Spain. Start data: September 2009
Our ultimate objectives (1)to analyze the impact of climate change on water resources in Spain. (2)to generate regional scenarios with statistical downscaling (3) to focus on the uncertainty and on the robustness of the regional scenarios
Motivation (1/3) Climate influences processes in natural and human system (e.g. Calmanti et al. 2007) Snout position variation of 5 big Alpine glaciar in northwestern Italy. Arbitrarily the origin of the distance axis at deviation from the glaciers position has been set in 1930. All the glaciers shown in this figure experienced a significant retreat during the measurement period. (Calmanti, Motta, Turco, Provenzale, Int. J. Climatology, 2007).
Motivation (2/3) The GCM resolution (few hundred kilometres, Solomon et al. 2007) is still much coarser than the driving processes of many impacts (Giorgi et al. 2001). GCM REAL
Motivation (3/3) The climate projections are affected by uncertainties (McAvaney et al. 2001, Giorgi 2005) Source: Giorgi (2005)
STRATEGY (follow IPCC guidelines, Wilby et al. 2004) Project aims and objectives DATA quality and quantity min(e1) min(e2) Downscaling scheme under optimal condition (Reanalysis) Calibration and verification Downscaling scheme under suboptimal condition Calibration and verification To generate scenario Impact assessment HYDROLOGICAL MODEL Focusing on: Extremes E.g. heavy rainfall proportion (R95p /PRCPTOT) or the longest dry period (CDD) Uncertainty Using different scenarios, different GCM models, different downscaling methods (eventually apply RCM scenarios) Impact on hydrology Using SURFEX to obtain future scenarios of soil wetness and runoff generation
STRATEGY (follow IPCC guidelines, Wilby et al. 2004) Project aims and objectives DATA quality and quantity Climatology of past Extremes (Turco and Llasat, submitted) min(e1) Downscaling scheme under optimal condition (Reanalysis) Calibration and verification To build a new gridded dataset using the SAFRAN analysis system (Durand et al, 1993, Quintana-Seguí et al., 2008). min(e2) Downscaling scheme under suboptimal condition Calibration and verification To generate scenario HYDROLOGICAL MODEL More details on the poster: 9. Simulation of the water balance of the NE Iberian Peninsula
Climatology of past extremes (EXTREME) INDICES ETCCDI: Expert Team on Climate Change Detection and Indices, http://cccma.seos.uvic.ca/etccdi Name PRCPTOT (mm) SDII (mm/d) R20 (days) RX1DAY (mm/d) RX5DAY (mm/5d) CDD (days) Definition Total precipitation in wet days (> 1 mm) Mean precipitation amount on a wet day (Simple Daily Intensity Index) Number of Days with precipitation over 20 mm/day Highest precipitation amount in one day Highest precipitation amount in five-day period Consecutive Dry Days (< 1 mm) TREND SIGNIFICANCE: (Circular) Block Bootstrap (Kiktev et al. 2003) The code will be freely available in short time http://gama.am.ub.es TX90p: Warm Days - Percentage of days with Tx>prctile(Tx,90) TEMPERATURE: E-OBS (25x25 km), 1950 2008 (Haylock et al. 2008) Our analysis of the temperatures in the Northeast of Spain indicates a clear signal of increase, coherent to the observed global warming (IPCC, 2007).
TRENDS IN PRECIPITATION SPAIN02 (20x20 km), 1950-2003 (Herrera et al. 2010; www.meteo.unican.es/es/datasets/spain02) Since a rather controversial picture appears in the studies on the precipitation trend in the NE of Spain, a greater effort has been done for this variable In order to analyze the influence of the length of the series as well as the departure point, 24 series have been built, shifting the starting year of the series. Temporal variability of the number of available stations used to build SPAIN02 with at least 75, 90, and 99% of valid daily data for every year. TREND PERIOD = 53 YEARS TREND PERIOD = 52 YEARS TREND PERIOD =.YEARS TREND PERIOD = 30 YEARS
TRENDS IN PRECIPITATION Annual total precipitation Trend analysis: no general trends *gray bands: sampling error Local increase in Consecutive Dry Days (CDD) index * 0.6 Dipolar trend pattern of the summer series of Prec. Intensity (SDII index)* 0.4 0.2 0-0.2-0.4-0.6 *These trend patterns have 99% of field significance; these trends persist for all the time windows
STRATEGY (follow IPCC guidelines, Wilby et al. 2004) Project aims and objectives DATA quality and quantity min(e1) min(e2) Downscaling scheme under optimal condition (Reanalysis) Calibration and verification Downscaling scheme under suboptimal condition Calibration and verification To implement the downscaling scheme To quantify the error 1 (assessing the strength and weakness of the SD ) (Turco et al., in preparation) To generate scenario HYDROLOGICAL MODEL
Analog Method This method assumes that analogue weather patterns (predictors) should cause analogue local effects (predictands). Predictor(s) Predictand B A 1961 1990 1991 2000 calibration validation adapted from Fernández and Sáenz (2003)
First test Temperature Domain Most common predictors in literature: slp+t850 (Brands et al. 2010) Obs. Data: E-OBS Predictor: ERA40, slp+t850 1979 1993 TX90p: Warm Days - Percentage of days with Tx>prctile(Tx,90)
First test Precipitation Domain Most common predictors in literature: MSLP, humidity, temperature, winds, geopotential (Turco et al., in preparation) Obs. Data: SPAIN02 Predictor: ERA40 fields Testing with the same standard predictors the downscaled rain occurrences shows some amount of skill but an extra effort it is necessary to solve the underestimated rainfall amounts.
Traditional approach Dynamical downscaling GCM OR Statistical downscaling GCM RCM Statistical downscaling Downscaled data Downscaled data Source: UC, http://grupos.unican.es/ai/meteo/ensembles/downscaling/index.html
Hybrid approach Dynamical AND Statistical downscaling GCM RCM Statistical downscaling Downscaled data Maraun et al 2010, Themebl et al 2010, Piani et al 2010
The simulations are the ERA40-driven RCMs provided by the EUfunded project ENSEMBLES. Period: 1961-2000 Acronym Model Institution Reference ETHZ CLM Swiss Institute of Technology Jaeger et al. 2008 KNMI RACMO Koninklijk Nederlands Meteorologish Instituut Van Meijgaard et al. 2008 METO-HC1 HadRM3 Q0 Hadley center / UK MetOffice Collins et al. 2006 MPI REMO Max Planck Institute of Meteorology Jacob D. 2001 UCLM PROMES Universidad de Castilla La Mancha Sanchez et al. 2004 The 5 RCMs used in this study were identified as the best performing models in Spain by Herrera et al. (2010) annual precipitation climatology of Spain02 and of the RCMs (mm)
Validation (1/4) 30 years for calibration, 10 for testing 2 experiments, test on driest/wettest period Standardized time series year Period Years Wettest 1996, 1969, 1997, 1979, 1963, 1972, 1977, 1989, 1971 and 1987 Driest 1964, 1998, 1994, 1990, 1970, 1967, 1983, 1973, 1980 and 1981 The wettest (driest) years have been identified in this way: the annual total precipitation in wet days for each point has been standardized, spatial averaged and finally sorted
Validation (2/4) Test period: Wettest CDD (day) SDII (mm/d)
Test period: wet Validation (3/4) RCMs Test period: dry ETCDDI index 1. The AM show bigger correlations than the RCMs 2. substantially invariance of the results testing the method in the wettest or driest years 3. Among the RCMs, KNMI, UCLM and ETHZ have the best correlation 4. The RX1DAY index (maximum precipitation in one day) has the lowest correlation The significance bars are calculated with a bootstrap method.
Dry test period Validation (4/4) Wet test period MAEr (mm/d) CORR ME (mm/d) MAEr= MAE / MEAN(OBS) AM tends to underestimate and it is able to reduce the peaks of the RCM bias ETHZ model shows a noisier pattern, with an overestimation in the Northeast and in the central part of Spain and an underestimation along the mountain range in the Northwest and in the South of Spain Although the correlation of the AM decreases in the test period wet, is still superior to the RCM The MAE r highlights the greater difficulties of the RCMs and of the AM to reproduce the precipitation in the Mediterranean area
NEXT STEP (1/3): SUB-OPTIMAL CONDITION Project aims and objectives DATA quality and quantity Downscaling scheme under optimal condition (Reanalysis) min(e1) Calibration and verification min(e2) Downscaling scheme under suboptimal condition Calibration and verification To implement the downscaling scheme with RCM driven by GCM To generate scenario HYDROLOGICAL MODEL
NEXT STEP (2/3): ASSESS THE UNCERTAINTIES Project aims and objectives DATA quality and quantity Downscaling scheme under optimal condition (Reanalysis) min(e1) Calibration and verification min(e2) Downscaling scheme under suboptimal condition Calibration and verification To generate scenario HYDROLOGICAL MODEL Using different scenarios, different GCM (or) RCM models, different downscaling method
NEXT STEP (3/3): HYDROLOGY Project aims and objectives DATA quality and quantity More details on the poster: min(e1) min(e2) Downscaling scheme under optimal condition (Reanalysis) Calibration and verification Downscaling scheme under suboptimal condition Calibration and verification To generate scenario HYDROLOGICAL MODEL 9. Simulation of the water balance of the NE Iberian Peninsula Using SAFRAN analysis system as observational database to apply the downscaling methods developed to force the SURFEX land-surface model in the future to obtain future scenarios of soil wetness and runoff generation.
Conclusion 1/4 Trend analysis Temperatures a clear signal of increase, coherent to the observed global warming (IPCC, 2007). Precipitation no general trends were found. Two local trend patterns: local increase in CDD index (around 30% of the area has an increase of around 2 days/decade) 0.6 0.4 0.2 0-0.2-0.4-0.6 dipolar trend pattern of the summer series of precipitation intensity (in the inland part, around -0.5 mm/decade, along the coast, around 1 mm/decade).
Conclusion 2/4 Analogs method under optimal conditions (reanalysis): skill in downscaling the mean and the extreme temperatures (using RCM) skill in downscaling the mean and the extreme precipitation SDII (mm/d)
Conclusion 3/4 Next steps: 1. To calibrate the AM under sub-optimal conditions 2. To generate the multi-scenarios 3. To use SAFRAN as an observation database and to force SURFEX with the downscaled scenarios on the NE of Spain
Conclusion 4/4 Purposes of our strategy to bridge the gap between dynamical models and the end user: 1. To supply the downscaled scenarios 2. To provide consideration of how the scenarios can be best used to plan and manage the strategies of adaptation and mitigation so that they are robust to the many uncertainties about the future.
Thank you for your attention mturco@am.ub.es http://gama.am.ub.es Picture by Eduardo Llasat
Extra slides
RX1DAY the monthly maximum calculated for each grid point has been averaged over each of these basins
The annual cycle of RX1DAY is reproduced quite properly by the AMs and the RCMs a slight improvement by the AM, with reduced spread The autumn months (mainly September) in the Mediterranean basins present the maximum observed values, the largest RCM spread and the AMs underestimate these values
difference among simulated (AM as well the respective RCM) and observed RX1DAY value for the month of September weighted on observed values Segura Levante Ebro Catalana Baleares WET DRY WET DRY WET DRY WET DRY WET DRY ETHZ KNMI METO- MPI UCLM ENS2 HC1 RCM 0.07-0.13-0.74-0.42 0.01-0.53 AM -0.36-0.42-0.73-0.62-0.68-0.37 RCM 0.43 0.51 0.25 0.14 0.93-0.08 AM -0.20 0.22 0.25-0.16 0.25 0.02 RCM 0.22-0.10-0.70-0.45 0.00-0.5 AM -0.28-0.33-0.60-0.60-0.54-0.43 RCM -0.01-0.03-0.33-0.12 0.66-0.35 AM -0.30 0.05 0.11-0.07 0.09-0.23 RCM 0.25 0.32-0.22-0.11 0.39-0.28 AM -0.18-0.06-0.22-0.18-0.20-0.02 RCM -0.13-0.19-0.21-0.26 0.12-0.43 AM -0.25-0.29-0.08-0.39-0.20-0.27 RCM 0.28 0.23-0.50-0.30-0.11-0.47 AM -0.3 0.02-0.37-0.40-0.26-0.03 RCM 0.02-0.37-0.24-0.38-0.06-0.47 AM -0.10-0.40-0.19-0.27-0.17-0.06 RCM -0.08-0.40-0.74-0.35-0.33-0.60 AM -0.29-0.39-0.65-0.68-0.54-0.15 RCM 0.03-0.03-0.52-0.27-0.24-0.43 AM -0.17-0.29-0.06-0.16 0.30-0.07 Errors greater than 75% (>0.75) are red coloured, between 50% and 75% in orange, between 25% and 50% in yellow, without colour for errors less than 25% biggest errors appear for the Segura, Levante and Baleares river basins AM aids to reduce the error of the respective RCM considering the test period dry instead AM has similar or worst performance of the respective RCM during the wet periods Best RCM: ETHZ
Dry test period Wet test period MAEr (mm/d) MAE (mm/d) CORR MAEr: MAE / MEAN(OBS) although the correlation of the AM decreases in the test period wet, is still superior to the RCM The MAE r highlights the greater difficulties of the RCMs and of the AM to reproduce the precipitation in the Mediterranean area
DSTD (mm/d) ME (mm/d) AM tends to underestimate, mostly on the Mediterranean coast ETHZ model shows a noisier pattern, with an overestimation in the northeast and in the central part of Spain and an underestimation along the mountain range in the northwest and in the south of Spain AM is able to reduce the peaks of the RCM bias The standard deviation is in general underestimated by the AM, especially along the Mediterranean coast. The ETHZ model shows again a noisier pattern, very similar to the ME pattern