Development and implementation of a probabilistic medium-range forecasting service for waterway transport on the River Rhine

Size: px

Start display at page:

Download "Development and implementation of a probabilistic medium-range forecasting service for waterway transport on the River Rhine"

Mabel Fitzgerald
6 years ago
Views:

(HITS) Department M2 - Water Balance, Forecasting and Predictions Federal Institute of

1 Development and implementation of a probabilistic medium-range forecasting service for waterway transport on the River Rhine Bastian Klein, Dennis Meissner, Stephan Hemri (HITS) Department M2 - Water Balance, Forecasting and Predictions Federal Institute of Hydrology (BfG) HEPEX Workshop Université Laval, Québec, Canada, June 6 to 8, 2016 Seite 1

The Waterway Rhine Amongst the German waterways the River Rhine takes an extraordinary position The River Rhine is one of the world s most frequented inland

The average number of vessels using the Rhine each day is approx. 600 at the Dutch-German border (~ 400 at Cologne).

2 The Waterway Rhine Amongst the German waterways the River Rhine takes an extraordinary position The River Rhine is one of the world s most frequented inland waterways. The volume of goods transported on the Rhine can be estimated at 310 million tons. The average number of vessels using the Rhine each day is approx. 600 at the Dutch-German border (~ 400 at Cologne). The fleet travelling on the Rhine waterway can be estimated at about 6,900 vessels (1,200 are pushed barges, 4,400 motor cargo vessels and 1,300 tankers). Duisburg is the world s largest inland port (1350 ha area, 21 port basins, 200 km railway track). Seite 2

3 Waterlevel Forecast River Rhine Model Chain Meteorological model P, T Hydrological model Q River Forecast station Lateral Input Hydrodynamic Model Stretch Hydrological Model districts Hydrodynamic model W Seite 3

4 Waterlevel Forecast River Rhine Model Chain Meteorological model P, T Hydrological model Q River Forecast station Lateral Input Hydrodynamic Model Stretch Hydrological Model districts Hydrodynamic model W Seite 4

5 Waterlevel Forecast River Rhine Model Chain Meteorological model P, T Hydrological model Q River Forecast station Lateral Input Hydrodynamic Model Stretch Hydrological Model districts Hydrodynamic model W Seite 5

Probabilistic Forecasts Quantification of the meteorological forecast uncertainty by using meteorological ensemble forecasts Problem: meteorological ensemble forecasts are generally underdispersiv

6 Probabilistic Forecasts Quantification of the meteorological forecast uncertainty by using meteorological ensemble forecasts Problem: meteorological ensemble forecasts are generally underdispersiv and biased Application of the statistical post-processing method Ensemble Model Output Statistics EMOS (Gneiting et al. 2005) to estimate the predictive uncertainty of water level forecasts Hindcast: daily forecasts for the period to ICON-EU ICON HRES COSMO-LEPS ENS GEFS Reforecast 2 11 Lead Time (days) Seite 6

7 Probabilistic Forecasts Quantification of the meteorological forecast uncertainty by using meteorological Aim of ensemble statistical forecasts post-processing Problem: (Gneiting meteorological al. ensemble 2007) forecasts are generally underdispersiv and biased Maximizing the sharpness of the predictive Application of the statistical post-processing method Ensemble distributions subject to calibration Model Output Statistics EMOS (Gneiting et al. 2005) to estimate the predictive uncertainty of water level forecasts Hindcast: daily forecasts for the period to ICON-EU ICON HRES COSMO-LEPS ENS Lead Time (days) GEFS Reforecast Seite 7

8 Current Workflow Deterministisic Forecasts Real Time Observation Meteorology Observation Runoff Observation Waterlevel Meteorological Forecasts HBV AR- Correction SOBEK ARIMA (0,1,0) Correction Runoff Forecast Waterlevel Forecast Seite 8

9 Future Workflow Probabilistic Forecasts Real Time New Worflow Components Observation Meteorology Observation Runoff Meteorological Forecasts HBV AR- Correction EMOS ECC-T Runoff Forecast Waterlevel Forecast Seite 9

10 Probabilistic Hydrographs Seite 10

11 Probabilistic Hydrographs Hydrodynamic model needs hydrographs and no probability distributions as input Ensemble Copula Coupling ECC-T is applied to calculate runoff trajectories for the input gauges Space-time dependency of the raw ensemble is retained Seite 11

12 Ensemble Copula Coupling Seite 12

13 Ensemble Copula Coupling Seite 13

14 Ensemble Copula Coupling Seite 14

Public Future Workflow Probabilistic Forecasts Real Time New Worflow Components

Forecasts HBV AR- Correction ARIMA (0,1,0) EMOS ECC-T SOBEK EMOS Correction Archiv

Forecasts Hindcast HBV AR- Correction Parameter Estimation EMOS ECC-T Hindcast SOBEK

15 Public Future Workflow Probabilistic Forecasts Real Time New Worflow Components Observation Observation Observation Waterlevel Meteorology Runoff Meteorological Forecasts HBV AR- Correction ARIMA (0,1,0) EMOS ECC-T SOBEK EMOS Correction Archiv meteorol. Forecasts Hindcast HBV AR- Correction Parameter Estimation EMOS ECC-T Hindcast SOBEK ARIMA (0,1,0) Correction Parameter Estimation EMOS Observation Meteorology Observation Runoff Observation Waterlevel Offline Runoff Forecast Waterlevel Forecast Seite 15

16 Verification Kaub - Rhine Seite 16

17 Verification Kaub - Rhine Seite 17

waterway transport on longer forecast lead times. Publication of daily mean values due to the large uncertainties.

18 Potential Probabilistic Forecast Products Product 1: deterministic 4-day forecast ( best guess ) Product 2: probabilistic forecast with lead time 0 to 10 days 0 5 days publication of instantaneous values 6 10 days outlook to cover the high demand of inland waterway transport on longer forecast lead times. Publication of daily mean values due to the large uncertainties. Publication of expected value and selected quantiles of the predictive distribution, exceedance / non-exceedance probabilities of defined water levels Seite 18

19 Simulation-based Cost Model Costs of IWT are mainly affected by waterway-characteristics, vessel-sizes and operating conditions Model Chain Meteorological model P, T Hydrological model economy of scale Simulation-based cost model (developed by DST) takes into account representative vessel types with their specific draughts, cargo capacities etc. Aim: demonstrating the potential economic benefit by using probabilistic forecasts Cost structure model Q Hydrodynamic model W, v Seite 19

Simulation-based Cost Model Water-level forecast Costs of IWT are mainly affected by waterway-characteristics, vessel-sizes and operating conditions Hydrodynamic simulations economy of scale

20 Simulation-based Cost Model Water-level forecast Costs of IWT are mainly affected by waterway-characteristics, vessel-sizes and operating conditions Hydrodynamic simulations economy of scale Simulation-based cost model (developed by DST) takes into account representative vessel types with their specific draughts, cargo capacities etc. The main cost components are taken into account: fixed costs (labour, insurance, investment), variable costs (fuel, lubrication, lighterage, waiting time) Seite 20

Simulation-based Cost Model reference run ( perfect forecast ) deterministic forecast risk-averse 1) deterministic forecast risk neutral deterministic forecast risk-seeking 1) Ensemble forecast (0.

21 Simulation-based Cost Model reference run ( perfect forecast ) deterministic forecast risk-averse 1) deterministic forecast risk neutral deterministic forecast risk-seeking 1) Ensemble forecast (0.25 Quant.) risk-averse Ensemble forecast (median) risk neutral Ensemble forecast (0.75 Quant) risk-seeking Gustav Königs Johann Welker GMS- 110 GMS- 135 GMS- Koppelv. Jowi Schubv. 2 * 2 7 ship types Rotterdam Port Duisburg Rotterdam Rhine-Moselle region Rotterdam Rhine-Neckar-region 3 routes upstream 1 direction The model causes virtual traffic Every day (within 4.5 years) one ship of each vessel type for each destination leaves the port of Rotterdam 1530 trips x 7 vessel types x 3 destinations = trips The load of each trip depends on - forecast variant used - willingness to take a risk - destination Seite 21

22 Difference in transport costs in relation to current deterministic forecast [ /ton] Simulation-based Cost Model Demonstrating the added (economic) value of probabilistic forecasts for waterway transport Vessel type: JOWI (max capacity 6,100 tons) GlW MW HSW Water-level at Kaub [cm] Seite 22

23 Difference in transport costs in relation to current deterministic forecast [ /ton] Simulation-based Cost Model Demonstrating the added (economic) value of probabilistic forecasts for waterway transport Vessel type: JOWI (max capacity 6,100 tons) GlW MW HSW probabilistic forecast synthetic deterministic forecast (overall improved by 5 cm) Water-level at Kaub [cm] Seite 23

24 Time of travel 3-4 days Simulation-based Cost Model Demonstrating the added (economic) value of an extended 10-day-forecast for waterway transport Reference run: each vessel has to leave the port of Rotterdam each day (particular load is based on forecast day 3) Modification: each vessel could decide within an additional time slot of 7 days (10 day forecast 3 day travel time) when to start the trip to the Rhine-Neckar metropolitan region Here: forecasts are assumed to be perfect Next step: test different forecast variants (meteo inputs, post-processing) Seite 24

25 Additional load [tons] water-level station Kaub [cm] Time of travel 3-4 days Simulation-based Cost Model 4,000 Demonstrating the added (economic) value of an extended 10-day-forecast for waterway transport 800 3,500 3,000 2,500 2,000 1,500 1, Reference run: each vessel has to leave the 700 port of Rotterdam each day (particular load is based on forecast day 3) 600 Modification: each vessel could decide within an additional time slot of 7 days (10 day forecast 3 day travel time) when to start the trip to the Rhine-Neckar metropolitan region Here: forecasts are assumed to be perfect Next step: test different forecast variants (meteo inouts, post-processing) Seite 25

26 Additional load [tons] water-level station Kaub [cm] Time of travel 3-4 days Simulation-based Cost Model 4,000 Demonstrating the added (economic) value of an extended 10-day-forecast for waterway transport 800 3,500 3,000 2,500 2,000 1,500 1, Reference run: each vessel has to leave the port of Rotterdam each day (particular load is based on forecast day 3) Modification: each vessel could decide within an additional time slot of 7 days (10 day forecast 3 day travel time) when to start the trip to the Rhine-Neckar metropolitan region Here: forecasts are assumed to be perfect Next step: test different forecast variants (meteo inouts, post-processing) Seite 26 0

27 Time of travel 3-4 days Simulation-based Cost Model Demonstrating the added (economic) value of an extended 10-day-forecast for waterway transport Reference run: each vessel has to leave the port of Rotterdam each day (particular load is based on forecast day 3) Modification: each vessel could decide within an additional time slot of 7 days (10 day forecast 3 day travel time) when to start the trip to the Rhine-Neckar metropolitan region Here: forecasts are assumed to be perfect Next step: test different forecast variants (meteo inouts, post-processing) Seite 27

Conclusion & Outlook Technical implementation of probabilistic waterlevel forecasts in the operational forecasting system of BfG finished Currently discussions about the publication platform /

28 Conclusion & Outlook Technical implementation of probabilistic waterlevel forecasts in the operational forecasting system of BfG finished Currently discussions about the publication platform / forecast products with the Federal Ministry of Transport and Digital Infrastructure and the Waterway Shipping Administration Further-development of statistical post-processing methods in collaboration with Heidelberg Institute for Theoretical Studies HITS and in the context of ongoing research projects Further work on demonstrating the potential economic benefit by using probabilistic forecasts in the context of the EU-Project IMPREX Seite 28

Vielen Dank für Ihre Aufmerksamkeit! Dr.-Ing.

Forecasting and Predictions Federal Institute

29 Merci beaucoup de votre intérêt! Thank you very much for your attention! Vielen Dank für Ihre Aufmerksamkeit! Dr.-Ing. Bastian Klein Department M2 - Water Balance, Forecasting and Predictions Federal Institute of Hydrology (BfG) Am Mainzer Tor 1, Koblenz, Germany Tel.: / klein@bafg.de Project Partners: Funding Seite 29

Impact of hydrological model uncertainty on predictability of seasonal streamflow forecasting in the River Rhine Basin

Impact of hydrological model uncertainty on predictability of seasonal streamflow forecasting in the River Rhine Basin Bastian Klein, Dennis Meißner Department M2 - Water Balance, Forecasting and Predictions