ݺߣ

ݺߣShare a Scribd company logo

Waterjade ml hydro_forecast_eng

•
•
Matteo Dall'Amico
Matteo Dall'Amico

This document discusses Waterjade's machine learning algorithms for forecasting water discharge in hydropower plants. It provides forecasts at various timescales, from very short term (0-4 hours) to long term (over 6 months). Short term forecasts (4 hours to 5 days) use numerical weather prediction data and machine learning, while seasonal forecasts (6 months) use global model data downscaled with machine learning to local scales. Performance is better for temperature than precipitation forecasts, especially in winter. The system aims to improve forecasts by incorporating snow data from satellites and sensors.

1 of 32
Download to read offline
Waterjade® – by MobyGIS S.r.l Operative seat: Viale Dante 300 | 38157 Pergine Valsugana, Trento (Italy) +39.0461.1560037 | info@waterjade.com | www.waterjade.com Machine learning algorithms for the forecast of water discharge in complex hydropower plants Ing. Stefano Tasin, Ing. Matteo Dall’Amico Hydromatters 4.0, Padova (Italy) 22­09­2020
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Discharge and Production Forecast 2050 Trading Safety Budgeting and  water level  management plant design or  revamp Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h Maneuvers  and plant  optimization +24 h Each prediction lead­time addresses a particular need  Now
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com NWP Numeric Weather Prediction  (WRF, seasonal forecast,  CMIP5 projection) Geomorphology Hydrometer Snow­Meteo ground data Satellite data Machine Learning Physical model Downscaling Weather Generator Solid precipitation  estimation Reservoir Mass Balance  Input Data Technology Production data
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Lead time Up to 5 hours Temporal resolution Hourly, sub­hourly Target Discharge Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast ● It exploits the discharge or  level data measured by  the hydrometers  upstream of the target  section; ● The predictive ability  depends on the distance  of the hydrometer and  the speed of propagation  wave in the river.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Row 1 Row 2 Row 3 Row 4 0 2 4 6 8 10 12 Column 1 Column 2 Column 3 1 week Discharge[m3 /s] Time With the hydrometric signals of S. Michele all'Adige (orange) and Mezzolombardo (green) it is possible to predict the  discharge of Ponte S. Lorenzo in Trento (blue).
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Forecast and data availability Now Observation data Gap Hydrological forecast Time Useless Useful ● GAP: time between the acquisition of the data and the actual  availability of the model prediction; ● The lower the GAP, the more useful and accurate the forecast
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Multi­Model Optimization Training set Prediction +3H Training set Prediction +2H Model +2H Model +3H +1 H +3 H+2 H Last available value Now GAP Model Hydrometer +2H A, B +3H A Each model can only exploit  hydrometers with a predictive capacity  greater than that its lead time. A B
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Performance Forecast Model Discharge variation Discharge variation R2 RMSE +1H 0.7244 5.295 +2H 0.6692 5.834 +3H -0.1143 11.020 1 day Discharge[m3 /s] Time [hours] 500 400 With sub-hourly models better performance can be achieved. Model validation in Ponte S.Lorenzo (Trento). Performances are good for models up to + 2H, while at + 3H they degrade quickly  because the predictive capacity of the hydrometers involved is exceeded.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Lead time Up to 3­5 days  Temporal resolution Hourly Target Production Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short term forecast Guatemala Target Guatemala
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short term forecast ● Hybrid approach (machine learning + semi­distributed physical model) for  flow generation; ● In hydrologically complex systems, physical models "help" machine learning,  providing more meaningful predictors. NWP Historical production Production forecast
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short­term forecasts vs baseline  RMSE 0-24 h RMSE 24-48 h Baseline prediction 21.5 22.0 ML prediction 15.8 16.0 Production[Mw] It is difficult to predict signals of a discontinuous nature. The “baseline” is defined as the state­of­the­art  predictor. For example, we could use the observation  of yesterday (latest data available) as a prediction for  tomorrow or the day after tomorrow. The Machine Learning (ML) prediction model  performs better than the baseline, despite the  difficulty of modeling a discontinuous variable.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Forecast Lead time Up to 6 months Temporal resolution Decadal, Monthly Target Discharge, Production Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal products ● Products from various European and non­European centers are  available thanks to the European Copernicus project (C3S).
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Global Model to locals scale ● Time resolution: monthly or  daily ● Spatial resolution: 1° (about 110 km) ● Forecasts updated once  a month. ● In South Europe a cell is  about 9.000 km2 but the  basins upstream of the  plants are often less than  100 km2. Great caution is needed to use the raw data without downscaling
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Temperature Downscaling Weather station Location: Italian Alps Altitude: 274 m Downscaled vs Ground data R2 Score: 0.91 The Machine  Learning (ML) is used  for downscaling the  2 m temperature. This  allows to eliminate  the bias due to the  altitude and the  local micro­climate. However, seasonal  forecasts do not  capture the  dynamics on a  synoptic scale.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Precipitation Downscaling Weather station Location: Italian Alps Altitude: 2100 m The downscaling of the  precipitation is  conducted on the  various members of the  ensembles, with the  objective to force the  median member to  comply with monthly  observed accumulations.  Very often global models  tend to underestimate  the precipitation at high  elevation. 
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Performance ­ Temperature Values above 0.5  denote useful skill  compared to  climatology.  Relative operating characteristics (ROC; Stanski et al. , 1989) for near‐surface temperature for the GloSea5 system in the periods JJA and DJF. ECMWF: Global Seasonal  forecast system version 5  (GloSea5)
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Performance ­ Precipitation Relative operating characteristics (ROC; Stanski et al. , 1989) for precipitation for the GloSea5 system in the periods JJA and DJF. Values above 0.5  denote useful skill  compared to  climatology.  ECMWF: Global Seasonal  forecast system version 5  (GloSea5)
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Performance in Europe  ● Seasonal vs climatology Temperature Precipitation Summer Good Discrete Winter Discrete Poor ● It is difficult to predict  the accumulation of  snow in the winter  season; ● but it is possible to  estimate its melting in  the spring, knowing  the current state of  snow.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Solid Precipitation estimation Reference point Location: Italian Alps Elevation: 2000 m ● Filter snow data ● Compute fresh­snow ● Compute fresh­snow density ● Compute virtual precipitation The procedure helps not to underestimate  the solid precipitation at high elevation.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Initial Condition Correction ● From SnowMaps1 (res. 250 m) ● From Sentinel 2 “Optical” ● From Sentinel 1 “SAR”  Forecast 1. Dall’Amico, M., Endrizzi S. and Tasin S. (2018): Mysnowmaps: operative high-resolution real-time snow mapping. Proceedings of the International Snow Science Workshop, Innsbruck, 328-332
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Hybrid approach: Physical Model + Machine Learning - SWE - Temperature - Precipitation - Temperature Forecast
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Model Performance MASE (Mean Absolute Scaled Error) <1  indicates better performance of the forecast  compared to the baseline (climatic average)  in predicting the observed data. Alpine basin, area 55 km2, between 2000 and 3000 m altitude. Forecasts from November 2017 to April 2020. Jan Feb Mar Apr May JunDic Jul MASE = MAEforecast MAEbaseline 0.85 0.72 0.61 The performance of the forecast model is always better than the baseline. The best performance is achieved in  the Spring months, thanks to the information on the state of the snow, as the winter season progresses.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Long Term Projection Lead time Up to 30­50 years Temporal resolution Daily, Monthly Target Discharge Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Climate Projection input meteo data: CMIP5 ● Different global models ● Spatial resolution: from  0.125°x0.125° to 5°x5° ● Time coverage:  1850­2300 ● Different scenarios  (RCP 2.6, 4.5, 6.0, 8.5)
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com CMIP5 Climate Projection: 2 m Temperature ● RCP 8.5  (business as  usual) ● RCP 2.6 (best) Alpine basin between  1000 and 1500 m altitude
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Weather Generator Probabilistic approach: 50­100 seeds to  generate daily time  series
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Discharge Projection Model Weather Generator Downscaling Observed meteo data Meteo projection Hydrological model Discharge projection CMIP5 The weather generator provides a precipitation  ensemble (50­100 members) and thus the hydrologic  model releases an ensemble of flow rates.
®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Separation of discharge contributions High­altitude alpine basin (2500 m ­ 3200 m) ● The physical  approach allows  separate the  contributions from  the various sources ● The spring peak of  snow melting is  noted (red),  followed by the  glacial one  (green).  ● The contribution of  rain is purely in the  summer period  (blue). Rain contribution Snow melt Glacial melt Discharge (m3/s)
Waterjade® – by MobyGIS S.r.l Operative seat: Viale Dante 300 | 38157 Pergine Valsugana, Trento (Italy) +39.0461.1560037 | info@waterjade.com | www.waterjade.com Thanks for the attention!

More Related Content

Waterjade ml hydro_forecast_eng

  • 1. Waterjade® – by MobyGIS S.r.l Operative seat: Viale Dante 300 | 38157 Pergine Valsugana, Trento (Italy) +39.0461.1560037 | info@waterjade.com | www.waterjade.com Machine learning algorithms for the forecast of water discharge in complex hydropower plants Ing. Stefano Tasin, Ing. Matteo Dall’Amico Hydromatters 4.0, Padova (Italy) 22­09­2020
  • 2. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Discharge and Production Forecast 2050 Trading Safety Budgeting and  water level  management plant design or  revamp Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h Maneuvers  and plant  optimization +24 h Each prediction lead­time addresses a particular need  Now
  • 3. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com NWP Numeric Weather Prediction  (WRF, seasonal forecast,  CMIP5 projection) Geomorphology Hydrometer Snow­Meteo ground data Satellite data Machine Learning Physical model Downscaling Weather Generator Solid precipitation  estimation Reservoir Mass Balance  Input Data Technology Production data
  • 4. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Lead time Up to 5 hours Temporal resolution Hourly, sub­hourly Target Discharge Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h
  • 5. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast ● It exploits the discharge or  level data measured by  the hydrometers  upstream of the target  section; ● The predictive ability  depends on the distance  of the hydrometer and  the speed of propagation  wave in the river.
  • 6. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Row 1 Row 2 Row 3 Row 4 0 2 4 6 8 10 12 Column 1 Column 2 Column 3 1 week Discharge[m3 /s] Time With the hydrometric signals of S. Michele all'Adige (orange) and Mezzolombardo (green) it is possible to predict the  discharge of Ponte S. Lorenzo in Trento (blue).
  • 7. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Forecast and data availability Now Observation data Gap Hydrological forecast Time Useless Useful ● GAP: time between the acquisition of the data and the actual  availability of the model prediction; ● The lower the GAP, the more useful and accurate the forecast
  • 8. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Multi­Model Optimization Training set Prediction +3H Training set Prediction +2H Model +2H Model +3H +1 H +3 H+2 H Last available value Now GAP Model Hydrometer +2H A, B +3H A Each model can only exploit  hydrometers with a predictive capacity  greater than that its lead time. A B
  • 9. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Performance Forecast Model Discharge variation Discharge variation R2 RMSE +1H 0.7244 5.295 +2H 0.6692 5.834 +3H -0.1143 11.020 1 day Discharge[m3 /s] Time [hours] 500 400 With sub-hourly models better performance can be achieved. Model validation in Ponte S.Lorenzo (Trento). Performances are good for models up to + 2H, while at + 3H they degrade quickly  because the predictive capacity of the hydrometers involved is exceeded.
  • 10. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Very Short Term Forecast Lead time Up to 3­5 days  Temporal resolution Hourly Target Production Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
  • 11. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short term forecast Guatemala Target Guatemala
  • 12. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short term forecast ● Hybrid approach (machine learning + semi­distributed physical model) for  flow generation; ● In hydrologically complex systems, physical models "help" machine learning,  providing more meaningful predictors. NWP Historical production Production forecast
  • 13. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Short­term forecasts vs baseline  RMSE 0-24 h RMSE 24-48 h Baseline prediction 21.5 22.0 ML prediction 15.8 16.0 Production[Mw] It is difficult to predict signals of a discontinuous nature. The “baseline” is defined as the state­of­the­art  predictor. For example, we could use the observation  of yesterday (latest data available) as a prediction for  tomorrow or the day after tomorrow. The Machine Learning (ML) prediction model  performs better than the baseline, despite the  difficulty of modeling a discontinuous variable.
  • 14. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Forecast Lead time Up to 6 months Temporal resolution Decadal, Monthly Target Discharge, Production Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
  • 15. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal products ● Products from various European and non­European centers are  available thanks to the European Copernicus project (C3S).
  • 16. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Global Model to locals scale ● Time resolution: monthly or  daily ● Spatial resolution: 1° (about 110 km) ● Forecasts updated once  a month. ● In South Europe a cell is  about 9.000 km2 but the  basins upstream of the  plants are often less than  100 km2. Great caution is needed to use the raw data without downscaling
  • 17. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Temperature Downscaling Weather station Location: Italian Alps Altitude: 274 m Downscaled vs Ground data R2 Score: 0.91 The Machine  Learning (ML) is used  for downscaling the  2 m temperature. This  allows to eliminate  the bias due to the  altitude and the  local micro­climate. However, seasonal  forecasts do not  capture the  dynamics on a  synoptic scale.
  • 18. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Precipitation Downscaling Weather station Location: Italian Alps Altitude: 2100 m The downscaling of the  precipitation is  conducted on the  various members of the  ensembles, with the  objective to force the  median member to  comply with monthly  observed accumulations.  Very often global models  tend to underestimate  the precipitation at high  elevation. 
  • 19. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Performance ­ Temperature Values above 0.5  denote useful skill  compared to  climatology.  Relative operating characteristics (ROC; Stanski et al. , 1989) for near‐surface temperature for the GloSea5 system in the periods JJA and DJF. ECMWF: Global Seasonal  forecast system version 5  (GloSea5)
  • 20. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Performance ­ Precipitation Relative operating characteristics (ROC; Stanski et al. , 1989) for precipitation for the GloSea5 system in the periods JJA and DJF. Values above 0.5  denote useful skill  compared to  climatology.  ECMWF: Global Seasonal  forecast system version 5  (GloSea5)
  • 21. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Performance in Europe  ● Seasonal vs climatology Temperature Precipitation Summer Good Discrete Winter Discrete Poor ● It is difficult to predict  the accumulation of  snow in the winter  season; ● but it is possible to  estimate its melting in  the spring, knowing  the current state of  snow.
  • 22. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Solid Precipitation estimation Reference point Location: Italian Alps Elevation: 2000 m ● Filter snow data ● Compute fresh­snow ● Compute fresh­snow density ● Compute virtual precipitation The procedure helps not to underestimate  the solid precipitation at high elevation.
  • 23. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Initial Condition Correction ● From SnowMaps1 (res. 250 m) ● From Sentinel 2 “Optical” ● From Sentinel 1 “SAR”  Forecast 1. Dall’Amico, M., Endrizzi S. and Tasin S. (2018): Mysnowmaps: operative high-resolution real-time snow mapping. Proceedings of the International Snow Science Workshop, Innsbruck, 328-332
  • 24. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Hybrid approach: Physical Model + Machine Learning - SWE - Temperature - Precipitation - Temperature Forecast
  • 25. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Seasonal Model Performance MASE (Mean Absolute Scaled Error) <1  indicates better performance of the forecast  compared to the baseline (climatic average)  in predicting the observed data. Alpine basin, area 55 km2, between 2000 and 3000 m altitude. Forecasts from November 2017 to April 2020. Jan Feb Mar Apr May JunDic Jul MASE = MAEforecast MAEbaseline 0.85 0.72 0.61 The performance of the forecast model is always better than the baseline. The best performance is achieved in  the Spring months, thanks to the information on the state of the snow, as the winter season progresses.
  • 26. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Long Term Projection Lead time Up to 30­50 years Temporal resolution Daily, Monthly Target Discharge Input data Technology ScopoProblem 2050 Short term Forecast Seasonal Forecast Long Term Projection +4 h +5 days-2 year +6 month Very Short Term Forecast 0 h +24 h NWP
  • 27. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Climate Projection input meteo data: CMIP5 ● Different global models ● Spatial resolution: from  0.125°x0.125° to 5°x5° ● Time coverage:  1850­2300 ● Different scenarios  (RCP 2.6, 4.5, 6.0, 8.5)
  • 28. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com CMIP5 Climate Projection: 2 m Temperature ● RCP 8.5  (business as  usual) ● RCP 2.6 (best) Alpine basin between  1000 and 1500 m altitude
  • 29. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Weather Generator Probabilistic approach: 50­100 seeds to  generate daily time  series
  • 30. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Discharge Projection Model Weather Generator Downscaling Observed meteo data Meteo projection Hydrological model Discharge projection CMIP5 The weather generator provides a precipitation  ensemble (50­100 members) and thus the hydrologic  model releases an ensemble of flow rates.
  • 31. ®Copyright Waterjade® Any reproduction without written permission is prohibited. Waterjade® – by MobyGIS S.r.l www.waterjade.com Separation of discharge contributions High­altitude alpine basin (2500 m ­ 3200 m) ● The physical  approach allows  separate the  contributions from  the various sources ● The spring peak of  snow melting is  noted (red),  followed by the  glacial one  (green).  ● The contribution of  rain is purely in the  summer period  (blue). Rain contribution Snow melt Glacial melt Discharge (m3/s)
  • 32. Waterjade® – by MobyGIS S.r.l Operative seat: Viale Dante 300 | 38157 Pergine Valsugana, Trento (Italy) +39.0461.1560037 | info@waterjade.com | www.waterjade.com Thanks for the attention!