A downscaling and adjustment method for climate projections in mountainous regions applicable to energy balance land surface models D. Verfaillie, M. Déqué, S. Morin, M. Lafaysse Météo-France CNRS, CNRM UMR 3589, France ICRC-Cordex 2016 conference Session C3 19 May 2016
The issue Strong societal demand about climate change impacts and its adaptation/mitigation Mountain regions : - snow availability - impacts on glaciers & water resources - ecosystem vulnerability - climate-related hazards Impact models sensitive to sub-diurnal variations Downscaling and bias-adjustment method : - quantile mapping & weather regimes - vs. a meteorological reanalysis (SAFRAN) - multi-variable and hourly 2
Reanalysis & snow modelling Northern Alps Southern Alps Temperature/ Humidity Wind Solar radiation IR radiation Rainfall Snowfall SAFRAN massifs SAFRAN reanalysis used as «pseudo-observation» Spatial subdivision by massifs & 300 m Snow model ISBA-Crocus elevation bands 3
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 4
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 5
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 6
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 7
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 8
SAFRAN ( C) Downscaling & Quantile-quantile historical SAFRAN/ diagrams q 80 (SAF) q 80 (R) histo ( C) 9
SAFRAN ( C) Downscaling & Adjustment of values (histo + scenarios) vs. SAFRAN Slope DT q 80 (SAF) q 80 (SAF) q 79 (SAF) DT q 79 (R) q 80 (R) q 80 (R) histo ( C) Adjustment : For a value z(r), As soon as q x (R) z(r) : z(r) corr = q x-1 (SAF) + (z(r) q x-1 (R)) DT x 10
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 11
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 12
Temperature (K) Downscaling & Disaggregation of adjusted from daily to hourly : temperature day i-1 (i-1) (i) day i day i+1 (i+1) (i-1) (i) (i+1) T SAF h : hourly T from SAFRAN analog T AL h : hourly adjusted ALADIN T, with α = 0 α = 2 (final value) α = Tmin/max AL : daily min/max raw ALADIN T Tmin/max AL : daily min/max adjusted ALADIN T (i-1) (i-1) (i) (i) (i+1) (i+1) 0h 2h 4h 6h 8h 10h 12h 14h 16h 18h 20h 22h 24h 20h 22h 24h 0h 2h 4h
Temperature (K) Downscaling & Disaggregation of adjusted from daily to hourly : temperature day i-1 (i-1) (i) day i day i+1 (i+1) (i-1) (i) (i+1) T SAF h T SAF h : hourly T from SAFRAN analog T AL h : hourly adjusted ALADIN T, with α = 0 α = 2 (final value) α = Tmin/max AL : daily min/max raw ALADIN T Tmin/max AL : daily min/max adjusted ALADIN T (i-1) (i-1) (i) (i) (i+1) (i+1) 0h 2h 4h 6h 8h 10h 12h 14h 16h 18h 20h 22h 24h 20h 22h 24h 0h 2h 4h
Temperature (K) Downscaling & Disaggregation of adjusted from daily to hourly : temperature day i-1 (i-1) (i) day i day i+1 (i+1) (i-1) (i) (i+1) T SAF h h (i) T- ALh (24h,i) T ALh (24h,i-1) T ALh (1h,i) T ALh (1h,i+1) T SAF h : hourly T from SAFRAN analog T AL h : hourly adjusted ALADIN T, with α = 0 α = 2 (final value) α = Tmin/max AL : daily min/max raw ALADIN T Tmin/max AL : daily min/max adjusted ALADIN T (i-1) (i-1) h (i) (i) (i) (i+1) (i+1) 0h 2h 4h 6h 8h 10h 12h 14h 16h 18h 20h 22h 24h 20h 22h 24h 0h 2h 4h
SAFRAN ( C) Downscaling & 1. Neighbouring grid points 2. & ERA- Interim regimes 3. SAFRAN : 1h daily 4. Percentiles of historical & SAFRAN (variable, season, regime) 5. Quantile mapping (variable, season, regime) 6. SAFRAN analogous days, criteria: - same month - same regime - coherent Pr. - consecutive days (when possible) 7. Adjusted : daily 1h (shape from SAFRAN analogues) 8. Total precipitation separation into rain and snow (1 C threshold) + additional quantile mapping of rain and snow separately ALADIN ( C) 16
Method evaluation Downscaling and adjustment of ALADIN-Climate driven by ERA-Interim reanalysis (domain centred on France) Temperature, precipitation, snow depth 300 m elevation bands for each massif and integrated for Northern & Southern Alps 2 neighbour selection methods : simple 2-D & more complex N50 (3-D with more weight on altitude) 3 «learning» periods (percentiles) : 1980-1994, 1995-2010 & 1980-2010 Correlation, ratio of standard deviations, altitudinal gradient, interannual evolution, annual cycle, PDF, RMSE, bias, 17
Bias (kg m -2 month -1 ) Bias (kg m -2 month -1 ) Method evaluation Ex1 : Biases for precipitation in the Southern Alps (1980-2010) DJF JJA Elevation (m) Elevation (m) L. 1980-1994 L. 1980-1994 N50 L. 1980 1994 no corr L. 1995-2010 L. 1995-2010 N50 L. 1995 2010 no corr L. 1980-2010 L. 1980-2010 N50 L. 1980 2010 no corr 18
Mean Monthly Snow Depth (m) Mean Monthly Snow Depth (m) Method evaluation Ex2 : Annual cycle of snow height in the Northern Alps at 1200 m (1980-1994 & 1995-2010) 1980-1994 1995-2010 SAFRAN reanalysis L. 1980-1994 L. 1980-1994 N50 L. 1995-2010 L. 1995-2010 N50 L. 1980-2010 L. 1980-2010 N50 19
Method evaluation Good reproduction of the statistical characteristics of SAFRAN variables Need for the ultimate quantile mapping on rain and snow separately (coherency between T and Pr phase) Good inter-variable consistency (snow) Limit in the temporal transferability? 20
First results Meteorological variables Temperature Only ARPEGE/ALADIN 21
First results Meteorological variables Precipitation Only ARPEGE/ALADIN 22
First results Snow Snow height from ALADIN Crocus at Col de Porte Only ARPEGE/ALADIN 23
Conclusions & Perspectives Generic downscaling & adjustment method for climate projections Respects the chronology Yields continuous hourly time series of adjusted meteorological variables (not only Pr & T) energy balance land surface models Numerous possible applications: natural & artificial snow production, agricultural & forest management, mountain ecosystems, glaciology, glacial hydrology,... Work in progress : - article in prep. on the evaluation of the method - application to all EUROCORDEX EUR-11 simulations (97) Future work : - production of new climate and snow scenarios - analysis of the uncertainties 24
Thanks for your attention Any question?
Downscaling & Forcing/ Scenarios EUROCORDEX simulation = 4 forcing/scenarios 13 s 10 s Hist RCP 2.6 Histo RCP2.6 RCP4.5 RCP8.5 Total 35 8 20 34 97 RCP 4.5 RCP 8.5 26
Downscaling & Forcing/ Scenarios Hist EUROCORDEX simulations Development of a downscaling and as generic as possible RCP 2.6 Scenarios of T, Q, Pr, Wind, Rad RCP 4.5 Quantile mapping vs. SAFRAN Northern Alps Southern Alps SAFRAN massifs Spatial subdivision by massifs, 300 m elevation bands RCP 8.5 27
Downscaling & Forcing/ Scenarios EUROCORDEX simulations Hist RCP 2.6 Scenarios of T, Q, Pr, Wind, Rad RCP 4.5 Quantile mapping vs. SAFRAN ISBA- Crocus Alps & Pyrenees Snow & Uncertainties for 1950-2005 & 2006-2100 RCP 8.5 28
The ADAMONT project (2015-2017) Impacts of climate change and adaptation in mountain regions (IRSTEA Grenoble, CNRM/CEN) My objectives: Climate scenarios with downscaling on the mountain regions (Alps & Pyrenees). Simulations of the snowpack driven by these scenarios. Based on : Historical forcing + RCPs scenarios (IPCC AR5, 2013) EUROCORDEX s/s EUR-11 simulations (12 km) Quantile mapping vs. SAFRAN 29
Precipitation phase First results Meteorological variables Only ARPEGE/ALADIN 30
Bias (K) Bias (K) RMSE (K) RMSE (K) Bias (K) Bias (K) RMSE (K) RMSE (K) Method evaluation Ex1 : Bias and RMSE for temperature in the Northern Alps Northern Alps DJF Northern Alps MAM Northern Alps DJF Northern Alps MAM Elevation (m) Elevation (m) Elevation (m) Elevation (m) Northern Alps JJA Northern Alps SON Northern Alps JJA Northern Alps SON L. 1980-1994 L. 1980-1994 N50 L. 1995-2010 L. 1995-2010 N50 L. 1980-2010 L. 1980-2010 N50 31 Elevation (m) Elevation (m) Elevation (m) Elevation (m)
Method evaluation Good reproduction of the statistical characteristics of SAFRAN variables Impact of the spatial selection technique 32
Downscaling & Percentiles of the historical (1974-2005) & SAFRAN (1980-2011) distributions for each variable, each season & each regime Example : q 80 = value for which 80% of the distribution has a lower value (q 50 = median) 33