SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve...

44
SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, DANUBE AND ELBE TO SUPPORT IMPROVED TRANSPORT COST PLANNING - Demonstrator -

Transcript of SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve...

Page 1: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, DANUBE AND ELBE TO SUPPORT IMPROVED TRANSPORT COST PLANNING

- Demonstrator -

Page 2: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

2

Deliverable 9.4 Semi-operational forecasting system for Rhine, Danube and Elbe to support improved transport cost planning

Related Work Package: WP 9: Sectoral Survey Transport

Deliverable lead: BfG

Author(s): Bastian Klein, Dennis Meißner

Contact for queries Bastian Klein <[email protected]>

Grant Agreement Number: n° 641811

Instrument: HORIZON 2020

Start date of the project: 01.10.2015

Duration of the project: 48 months

Website: www.IMPREX.eu

Abstract Hydrological forecasts of different lead times are needed by inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the overall transport market.

Within IMPREX several pre-operational forecast products have been developed to support decision making in this sector: 1) probabilistic 10-day waterlevel forecast for waterway Rhine,the most important inland waterway in Europe, required tooptimize transport management, short-term stock managementand waterway maintenance;2) monthly to seasonal flow forecast products for the Germanwaterways Rhine, Elbe and Danube required for medium- to long-term planning and optimization of the water-bound logistic chain(stock management, fleet composition, adjustment of theindustrial production chain, connecting waterway transport withother transport modes / modal split planning), as well as for thesediment management of harbours and tidal influencedwaterways.

Besides the forecasts products, which are the visible output for the stakeholders, deliverable D9.4 describes the underlying forecasting systems, which have been set-up in the course of IMPREX at BfG. These forecasting systems form the backbone of future forecasting services providing users with specific forecast products. All forecast products presented in D9.4 have been co-designed with the WP9 stakeholders of IMPREX.

Page 3: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

3

Dissemination level of this document

X PU Public

PP Restricted to other programme participants (including the Commission Services)

RE Restricted to a group specified by the consortium (including the European Commission Services)

CO Confidential, only for members of the consortium (including the European Commission Services)

Versioning and Contribution History

Version Date Modified by Modification reasons

v.01 01/03/2019 B. Klein, D. Meißner First version

v.02 07/03/2019 B. Klein Internal BfG review by Asta Kunkel

v.03 22/03/2019 B. Klein Review Linus Magnusson

v.04 16/04/2019 B. Klein Review Bvd Hurk

Page 4: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

4

Table of Contents

List of figures ...........................................................................................................................................5

List of tables ........................................................................................................................................... 6

Acknowledgement .................................................................................................................................. 7

Glossary ................................................................................................................................................. 8

Introduction ................................................................................................................................... 12 1

Forecast area German waterways .................................................................................................. 16 2

10-day probabilistic waterlevel forecast for waterway Rhine ......................................................... 18 3 Meteorological Forcing Data .................................................................................................. 19 3.1 Workflow .............................................................................................................................. 20 3.2 Forecast Product ................................................................................................................... 22 3.3 Verification ............................................................................................................................ 23 3.4

Monthly forecast German waterways Rhine, Elbe, Danube .......................................................... 26 4 Meteorological Forecast Data ................................................................................................ 27 4.1 Workflow .............................................................................................................................. 28 4.2 Forecast Product ................................................................................................................... 29 4.3 Verification ............................................................................................................................ 31 4.4

Seasonal forecast German waterways Rhine, Elbe, Danube .......................................................... 34 5 Meteorological Forecast Data ................................................................................................ 34 5.1 Workflow ............................................................................................................................... 34 5.2 Forecast Product .................................................................................................................... 36 5.3 Verification ............................................................................................................................ 37 5.4

Summary ...................................................................................................................................... 40 6

References ........................................................................................................................................... 42

Page 5: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

5

List of figures

Figure 1: Interaction between hydrologic conditions (water level), waterway parameters (fairway depths), specific navigation thresholds and transport costs (Meißner et al. 2017) ................................. 13

Figure 2. Map showing the German stretches of the international waterways Rhine, Danube and Elbe (left part), for forecast gauges (black dots) the long-term monthly mean flow rates (1971-2000) are visualized (right part). ............................................................................................................................ 16

Figure 3: Forecasting system of the BfG for the waterway Rhine(Meissner & Rademacher 2010) .......... 18

Figure 4: Current operational deterministic 4-day water-level forecast for gauge Kaub / Rhine initialized 22. February 2019 07:00 published via www.elwis.de. Observed waterlevels in blue, Forecast in red. ... 19

Figure 5: Numerical weather prediction products used to produce the probabilistic water level forecasts20

Figure 6: Workflow 10-day probabilistic water level forecast for the River Rhine ................................... 21

Figure 7: Time series display of forecasting system FEWS showing the raw (upper panel) and the probabilistic (lower panel) 10-day waterlevel forecast at gauge Kaub initialised at 22. February 2019 06:00UTC 22

Figure 8: PDF-Report of the probabilistic 10-day water level forecast for gauge Kaub / Rhine initialised 22. February 2019 06:00 UTC ................................................................................................................ 23

Figure 9: Continuous ranked probability skill score CRPSS, mean absolute error MAE, mean error ME and relative frequency of absolute error below 20 cm rfae20cm of the probabilistic water level forecasts at gauge Kaub (verification period 1. January 2008 – 31. December 2015) as a function of the lead time. Ensemble: raw ensemble, EMOS post-processed predictive distribution, EEMOS expected value of the predictive distribution, ENSMEAN mean of the raw ensemble, HRES deterministic forecast using ECMWF-HRES as forcings ..................................................................................................................... 25

Figure 10: Forecast Methods used for long-term forecasting in IMPREX ............................................... 27

Figure 11: Workflow 6-week flow forecast for the waterways Rhine, Elbe and Danube ......................... 28

Figure 12: Time series display of forecasting system FEWS showing the 6 week observed climatological (upper panel) ESP (middle panel) and the ECMWF-ENS (lower panel) extended forecast initialised at 18. February 2019 for gauge Kaub / Rhine ................................................................................................... 29

Figure 13: PDF-Report of the 6-week forecast for gauge Kaub / Rhine initialised 18. February 2019 00:00 UTC (page 1-2) ...................................................................................................................................... 30

Figure 14: PDF-Report of the 6-week forecast for gauge Kaub / Rhine initialised 18. February 2019 00:00 UTC (page 3-4) ...................................................................................................................................... 31

Figure 15:CRPSS of the forecasted weekly mean flows at Kaub / Rhine (black), Hofkirchen / Danube (blue) and Neu-Darchau / Elbe (red) of the period 02. January 2000 – 09. March 2016 ........................... 32

Figure 16: Workflow Seasonal flow forecast for the waterways Rhine, Elbe and Danube ...................... 35

Figure 17: Time series display of forecasting system FEWS showing the observed climatological (upper panel) ESP (middle panel) and the ECMWF-SEAS forecast initialised at 01. February 2019 for gauge Kaub / Rhine .......................................................................................................................................... 36

Page 6: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

6

Figure 18: Seasonal Forecast for gauge Kaub / Rhine initialised 01. February 2019 00:00UTC. Upper figure: categorical forecast of five flow classes, lower figure: probability distribution of mean monthly flows 37

Figure 19: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Kaub / Rhine: climatological meteorological forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast period analysed 1981-2016 ..................................................................................................... 38

Figure 20: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Hofkirchen / Danube: climatological meteorological forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast period analysed 1981-2016 ....................................................................... 38

Figure 21: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Neu-Darchau / Elbe: climatological meteorological forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast period analysed 1981-2016 ....................................................................... 38

List of tables

Table 1: User needs of the different waterway users, corresponding decisions requiring additional forecast lead-time ................................................................................................................................. 14

Table 2. Catchment size, annual mean, mean low flow and mean high flow at selected forecast gauges at Rhine, Danube and Elbe. Characterising statistics from undine.bafg.de. ........................................... 17

Table 3: Forecasting gauges, gauge location at the river, catchment area and waterlevel thresholds relevant for navigation: Gleichwertiger Wasserstand GlW, water-level relevant for waterway maintenance, undershot on 10-20 days per year in the long-term mean, TuGlW water depth below GlW, HSW highest navigable water-level and HHW highest observed waterlevel .......................................... 19

Table 4: Features of the numerical weather prediction products used to produce the probabilistic water level forecasts ....................................................................................................................................... 20

Page 7: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

7

Acknowledgement

IMPREX has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement N°641811.

Page 8: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

8

Glossary

Bias correspondence between the average forecast values and the average observed values

Categorical forecast forecast which predicts a discrete number of (discharge / water level) categories (e.g. dry, normal, wet). Categorical forecasts could be based on a deterministic or a probabilistic prediction

Climatological forecast a forecast based upon the observed climatological statistics of a forecast predictand. The climatological forecast is often used as reference forecast to calculate the skill of a forecast system

Draught Draught total of a vessel in motion = draught loaded + squat. Draught loaded – the distance between the lowest point of the bottom of a loaded, stationary vessel and the water surface

EMOS Ensemble Model Output Statistics. Statistical Post-processing method to estimate the predictive distribution of / to calibrate raw ensemble forecasts

Fairway the part of a waterway in which specific widths and depths are maintained to enable continuous navigation

Fairway depth available depth between the water surface and the riverbed determined for a “minimal cross section” of a defined fairway width

Forecast frequency recurrence interval for the calculation / publication of a forecast

Forecast horizon is the length of time into the future for which forecasts are calculated / published

Forecast initialization is the time at which the forecast calculation starts. Initial values of model states and fluxes at the beginning of the forecast are required. These could be e.g. derived from a model run up to the forecast initialization date driven by observations.

Forecast length see: Forecast horizon

Forecast quality description of how well the forecasts are in comparison to the observation. Forecast quality is defined by many different forecast attributes, such as e.g. Accuracy, Bias, Discrimination, Reliability, Sharpness, and Resolution

(Economic) Forecast value a forecast has value if it helps improving decisions. The forecast value is linked to forecast skill but it is not the same! The relative economic value plot is often used as verification diagnostic of forecast value.

GlW equivalent low water-level (GlW = “Gleichwertiger Wasserstand”) low flow reference water-level used for waterway maintenance on the River Rhine and Elbe, which is not exceeded on 20 ice-free days on average

GlQ Flow rate associated with GlW

Hindcast see: Re-Forecast

IWT Inland Waterway Transport

Page 9: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

9

Deliverable n° 9.4

HSW I / HSW II Highest Navigable Water-Level (HSW = Höchster Schifffahrtswasserstand) leading to restrictions (HSW I) or suspension (HSW II) of navigation on a specific waterway stretch.

Lead time period of time between the date of initialization of the forecast (the forecast issue date) and the date at which the verification of the forecast applies

Low-water surcharge long periods of low water increase transport prices because as water levels go down so does the load capacity of inland vessels. More tonnage is required to carry the same amount of cargo, which increases costs substantially. To compensate the disproportionate increase in costs low-water surcharges are payed to the skipper or the transport company. These are defined for several low water-levels for several gauges at the waterway Rhine and Danube.

Means of transportation conveyance to perform transport, usually self-propelled; the most dominant means of transportation are road transport, railway, ship and aviation

Model bias systematic error / bias of a forecast model. See: Bias

Model drift the tendency of seasonal forecasts to drift towards the climate model from which they are issued with increasing lead time, giving rise to model bias

Modal split the partition of freight transport volume amongst the different modes of transportation: road, rail, air and water; dependant on the type of cargo, the transportation route, the external influences etc. the optimal mode of transportation could vary (synonym: modal share)

Modes of transportation see: Means of transportation

NWP Numerical weather prediction

Multi-modal transport transport where complete transport units (e.g. containers) are carried by at least two different means of transportation along the route

Probabilistic forecast forecast of the future probability of one or more events occurring. The events can be discrete or continuous. In case of continuous variables the probabilistic forecast is described by a predictive distribution.

Re-(trospective) forecast also called hindcast: a forecast made for a period in the past using only information available before the beginning of the forecast. Usually the current configuration of the forecast system is used. Re-Forecasts can be used to estimate the parameters of statistical post-processing methods to calibrate the forecast, estimate bias-/drift correction parameters and/or estimate the skill of the current forecast system. In hydrological re-forecasting meteorological forecasts are needed as forcings. These could be in the best case also re-forecasts or alternatively archived operational forecasts.

Page 10: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

10

Transport movements of goods in one or more trips using one or more means of transportation

Transport chain technical and organisational linkage of transportation activities

Water depth distance between the water surface and the riverbed

Water level is the height of the water surface above or below a reference level, e.g. gauge zero

Page 11: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the
Page 12: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

12

Introduction 1

Inland waterway transport is vulnerable to long-lasting low flow periods, demonstrated by the recent low flow events 2015, 2016 and especially 2018. Such long-lasting and extreme drought events have a harmful effect on the competitiveness of the inland navigation sector within the overall transport market. Although extreme events play a major and well documented role, also hydrological effects of non-extreme low-flow conditions have an impact on waterway transport and respectively its transport when the water-level drops below mean water-level conditions. The maximum draught is limited and therefore the transport volume of the ships is reduced leading to higher transport prices for companies relying on Inland Waterway Transport (IWT). The additional charges (Low-water surcharge payed to the skipper or the transport company) paid by the dispatching company, to compensate the disproportionate increase in costs, are defined for several low water-levels for several gauges at the waterway Rhine and Danube. Below a certain low water threshold there is no longer an obligation to transport by the shipper on contracted transport prices, i.e. price frameworks are not longer valid in extreme low flow conditions, and transport prices could be negotiated freely (see Figure 1).

High water-levels also affect inland waterway transport. In addition to the protection of the infrastructure the security of navigation is the main motivation, because high flow velocities occurring during floods reduce the manoeuvrability of the vessels traveling downstream. Restrictions are related to defined water-levels of the waterway. The most relevant ones are the Highest Navigable Water-Levels HSW I and HSW II (HSW = “Höchster Schifffahrtswasserstand”) (see Figure 1). The exceedance of HSW I (the lower navigational flood level) means that vessels have to reduce their speed (leading to longer travel times) and they are forced to travel within the fairway. If HSW II is reached or exceeded shipping along the waterway section concerned is prohibited. Additionally the guaranteed clearance below bridges might become too low and limits the possible layer of containers. Therefore, water-levels are not solely relevant for the flood protection community but also for navigational user. Although the duration and frequency of occurrence of floods is significantly lower than low flows, floods could cause relevant costs with regard to inland waterway transport.

Besides the increasing transport costs the overall reduced transport capacities of the inland waterway during extreme and long-lasting low flow situations, as e.g. in 2018, could lead to relevant supply shortages of raw materials for producing companies and as a consequence to a reduction or a stoppage of the production. Therefore extreme low flow events also have large impacts on the national economy. To be prepared for extreme low flow events, to optimize transport costs and efficiency, as well as to avoid supply shortfall for major industries as long as it is possible, there is a strong need for flow and water-level forecasts of different lead times with special focus on the waterway transportation sector.

Page 13: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

13

Deliverable n° 9.4

Figure 1: Interaction between hydrologic conditions (water level), waterway parameters (fairway depths), specific navigation thresholds and transport costs (Meißner et al. 2017)

At the first IMPREX WP9 transport user workshop in Koblenz 11th April 2016 user needs for navigation related forecast products have been identified (Klein & Meissner 2016) (see Table 1). Navigational users need short-to-medium term water-level forecasts to optimize the load capacity of their vessels as well as to take into account in time that waterways might be blocked due to floods, to plan complete transport cycles, multi-modal transport planning, stock and supply management, and optimized timing of transports. Monthly to seasonal forecasts are required for the medium- to long-term planning and optimization of the water bound logistic chain (stock management, fleet composition, adjustment of the industrial production chain, modal split planning).

Within IMPREX pre-operational forecast products for the River Rhine – Europe’s most important waterway – and the German parts of the waterways Elbe and Danube, covering different lead times have been developed by the Federal Institute of Hydrology (BfG) and provided to the stakeholders. In the following chapters of this report the pre-operational forecast systems and the related forecast products are presented:

• the probabilistic 10-day water-level forecast system developed for the waterway Rhine in section 3,

• the monthly and seasonal flow forecast systems developed for the German waterways Rhine, Elbe and Danube in section 4 and 5.

Page 14: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

14

Table 1: User needs of the different waterway users, corresponding decisions requiring additional forecast lead-time

Page 15: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the
Page 16: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

16

Forecast area German waterways 2

The European inland waterways offer more than 40,000 km network of canals, rivers and lakes connecting cities and industrial regions across the continent. The German inland waterway network – an integral part of the trans-European waterway system – comprises about 7,350 kilometres, of which approximately 75 percent are rivers and 25 percent canals. The major inland waterways with regard to freight transport are the Rhine (with its tributaries Neckar, Main, Moselle and Saar) and the Danube, as well as parts of Elbe and some canals interconnecting the natural waterways. About two third of the German waterways are of international relevance, whereupon the importance of River Rhine is outstanding: with almost 200 million tons transported along the Rhine per year (approximately 2/3 of the European IWT volume) the Rhine isn’t solely Germany’s, but also Europe’s most important inland waterway (CCNR 2016). Approximately one third of the rivers in Germany used as waterways are free-flowing, so they are particularly affected by low flows as the dominating hydro-meteorological impact on IWT. Therefore this study focusses on the free flowing stretches of the international waterways Rhine, Danube and Elbe (see Figure 2).

Figure 2. Map showing the German stretches of the international waterways Rhine, Danube and Elbe (left part), for forecast gauges (black dots) the long-term monthly mean flow rates (1971-2000) are visualized (right part).

The River Rhine, with a total length of 1,230 km, drains an area of approx. 200,000 km2 with a mean flow rate of approx. 2,500 m³/s at its mouth in the North Sea. It is shippable for large vessels between Rotterdam and Basel on a length of about 800 km. While the main shippable tributaries of the River Rhine are impounded offering a guaranteed fairway depth, the Rhine itself is a free flowing waterway between Iffezheim / Karlsruhe and the beginning of the delta near Pannerdensche Kop in the

Page 17: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

17

Deliverable n° 9.4

Netherlands (approximately 500 km). The flow regime of the River Rhine shifts in downstream direction from a snow-dominated regime induced by the Alps to a complex flow regime in the Middle (gauge Kaub) and Lower Rhine stretch (gauge Cologne and Ruhrort) due to the increasing influence of the rainfall dominated (pluvial) flow regimes of the major tributaries (Neckar, Main, Moselle). Low flows, leading to restrictions for waterway transport, typically occur in the River Rhine in late summer and autumn due to high evaporation and low melt water input from the Alpine region. The River Danube, with a total length of 2,826 km, drains an area of 817,000 km² with a mean flow rate of approximately 6,500 m³/s. It is shippable on a length of 2,415 km between Kelheim and the Black Sea. The German part of the waterway (approx. 220 km) is impounded offering a minimum fairway depth of 2.70 m up to 2.90 m, except for a 70 km section between Straubing and Vilshofen. The flow regime in this critical stretch for waterway transport is pluvio-nival with a complex broad-peaked flow shape resulting from an overlapping of rainfall and snowmelt influences. Autumn is the typical low flow season, often extended to the winter months. The River Elbe, with a total length 1,090 km, drains an area of approx. 150,000 km² with a mean flow rate of approximately 860 m³/s. About 930 km are shippable between Pardubice in the Czech Republic and the mouth in the North Sea at Cuxhaven / Germany. The stretch upstream of Geesthacht up to the German - Czech border (nearly 600 km) is free-flowing. The Elbe between Dresden and Neu-Darchau shows pronounced pluvial flow regimes with maximum flows in late winter / spring and lowest flow values in summer and autumn. Compared to Rhine and Danube, the low flow season relevant for waterway transport already starts in early summer and lasts until autumn. Figure 2 shows the location of selected forecast gauges along the three considered rivers and visualizes (right part) the flow regime of the three waterways represented by the long-term monthly mean flow rate at the forecast gauges (period 1971 – 2000). Table 2 gives an overview about characterising statistics of selected forecast gauges.

Table 2. Catchment size, annual mean, mean low flow and mean high flow at selected forecast gauges at Rhine, Danube and Elbe. Characterising statistics from undine.bafg.de.

Gauge River Area (km2) Mean Flow (m3/s) Mean Low Flow (m3/s) Mean High flow

[m³/s]

Kaub Rhine 103 729 1650 779 4270

Cologne Rhine 144 232 2090 913 6200

Ruhrort Rhine 152 895 2230 1040 6630

Pfelling Danube 37 775 456 202 1490

Hofkirchen Danube 47 518 637 305 1870

Dresden Elbe 53 096 332 111 1700

Magdeburg Elbe 94 942 554 231 1850

Neu Darchau Elbe 131 950 705 271 2020

Page 18: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

18

10-day probabilistic waterlevel forecast for waterway Rhine 3

The probabilistic 10-day waterlevel forecast is based on the forecast system of BfG used to produce the operational 4-day water-level forecasts published on the River Information System ELWIS (www.elwis.de). Originally navigation-related forecasts have been developed in order to primarily support the individual skipper who aims at maximizing the load of an upcoming trip (Meissner & Rademacher 2010). Therefore the current lead-times of one to several days usually comply with the travel time of the vessels to pass the main bottlenecks of a waterway leaving the loading port. The operational forecasts published on ELWIS are still deterministic. To account for the uncertainty of the forecast, lead-times published are dependent on the quality of the forecast. Two types of forecast qualities are distinguished:

1. “Vorhersage” (Forecast): 80% of the forecast errors of the deterministic forecast have to be in the interval -10cm to +10 cm

2. “Abschätzung” (Trend): 80% of the forecast errors of the deterministic forecast have to be in the interval -20cm to +20 cm

For the waterway Rhine forecasts are published for all gauges with a lead time of 4 days (day 1 to 2 “Vorhersage”, day 3 to 4 “Abschätzung”).

Navigation related forecasts are published for 7 gauges along the river Rhine, relevant for navigation (see Figure 3, Table 3).

Figure 3: Forecasting system of the BfG for the waterway Rhine(Meissner & Rademacher 2010)

For illustration, Figure 4 shows the deterministic 4-day forecast initialized 22. February 2019 07:00 for gauge Kaub located at the river Rhine published via ELWIS.

Page 19: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

19

Deliverable n° 9.4

To extend the forecast horizon in the light of the increasing need to integrate IWT into multi-modal transport chains and the overall tendency to increasing vessel sizes, the existing forecast system was extended.

Additional meteorological forecast products, including ensemble systems, are integrated in the system to consider the meteorological forecasting uncertainty and the statistical post-processing method Ensemble Model Output Statistics EMOS was used to estimate the total uncertainty of the waterlevel forecast.

Table 3: Forecasting gauges, gauge location at the river, catchment area and waterlevel thresholds relevant for navigation: Gleichwertiger Wasserstand GlW, water-level relevant for waterway maintenance, undershot on 10-20 days per year in the long-term mean, TuGlW water depth below GlW, HSW highest navigable wa-ter-level and HHW highest observed waterlevel

Gauge River-

Location [km]

Area [km²]

GlW [cm]

TuGlW [cm]

HSW [cm]

HHW [cm]

Oestrich 518.08 87 190

Kaub 546.3 103 488 78 190 640 911 (02.02.1893)

Koblenz 591.49 109 806 78 210 650 949 (23.12.1993)

Köln 688 144 232 139 250 830 1069 (01.01.1926)

Düsseldorf 744.2 147 680 97 250 880 1110 (02.01.1926)

Ruhrort 780.8 152 895 233 280 1130 1300 (02.01.1926)

Emmerich 851.9 159 555 84 280 870 986 (03.01.1926)

Figure 4: Current operational deterministic 4-day water-level forecast for gauge Kaub / Rhine initialized 22. February 2019 07:00 published via www.elwis.de. Observed waterlevels in blue, Forecast in red.

Meteorological Forcing Data 3.1

Meteorological forcing data used to calculate water-level forecasts with an extended forecast horizon are based on a 72 member multi-model ensemble: 51 ensemble members from ECMWF ENS (1 control forecast and 50 perturbed members) as well as the control forecast ECMWF HRES with a higher spatial resolution (Leutbecher & Palmer 2008, Owens & Hewson 2018), and the 20 members of the limited-

Page 20: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

20

area ensemble prediction system by the consortium for small-scale modelling COSMO LEPS (Montani et al. 2011, Marsigli et al. 2014). Relevant features (lead time, resolution, initialisation time used) are summarized in Figure 5 and Table 4.

Archived forecasts of the period January 2008 to December 2015 have been used to produce a comprehensive water level re-forecast data set.

Table 4: Features of the numerical weather prediction products used to produce the probabilistic water level forecasts

Model Initialisation Lead time resolution Ensemble member

HRES 0:00 UTC 240 h ~9 km 1

ENS 0:00 UTC 360 h ~18 km 51

COSMO-LEPS 12:00 UTC day before

132 h ~ 7 km 20

Figure 5: Numerical weather prediction products used to produce the probabilistic water level forecasts

Workflow 3.2

In order to transfer precipitation and temperature information from meteorological measurement and forecasts into flow and subsequently into water-levels at relevant gauges, hydrological and hydrodynamic models are required. At BfG the conceptual, semi-distributed rainfall-runoff model HBV-96 (Bergström 1995, Lindstrom et al. 1997) is applied to calculate the flow forecasts used as boundary conditions and lateral inflows of the hydrodynamic model SOBEK (Deltares 2012) used to calculate water level forecasts along the river Rhine. The river Rhine basin is divided into 134 subbasins which are further subdivided into hydrological response units (HRU) according to land use and elevation classes. The flow formation processes are calculated on those HRUs. The model calculates flow with a temporal resolution of 1 h using temperature and precipitation fields that have been interpolated over the subbasins as meteorological input.

The hydrodynamic model suite SOBEK is used as one-dimensional model, which uses cross-section information of the River Rhine as well as its main tributaries. The distance between the cross-sections, which cover the river bathymetry as well as its floodplains, is non-equidistant and ranges between 100 m and 800 m. As the main tributaries of the River Rhine are impounded rivers (e.g. Moselle, Main) the SOBEK-model includes several weirs with their specific control rules in order to simulate the real behaviour of these elements, too.

Figure 6 shows the workflow of the probabilistic 10-day waterlevel forecast for the waterway Rhine. The flow and water level forecasts are initialized each day at 06:00 UTC, which means that observed real-time meteorological data, interpolated to the subbasins of the hydrological model, up to the

Page 21: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

21

Deliverable n° 9.4

forecast date is used as forcings of the hydrological model and observed flow is used as input for the hydrodynamic model to initialize the model states. For the forecast period meteorological ensemble runs from the different Numerical Weather Prediction (NWP) models (see chapter 3.1) are used as forcings of the hydrological model. Flow forecasts of the large tributaries of the river Rhine (see Figure 3) simulated with HBV are then used as input for the hydrodynamic model. To reduce the error of the input to the hydrodynamic model autoregressive error correction models (Broersen & Weerts 2005) are applied using the differences between the simulation of the model using meteorological observations as forcings and the actually past flow observations as training data. This error correction reduces the error of the hydrological model at the forecast initialization time to zero. To reduce the error of the waterlevel forecasts obtained by running the hydrodynamic model, again autoregressive error correction models are applied using the differences between the water level simulation using observed flow as input and the water-level observations of the past.

Probabilistic NWP ensemble forecasts are biased and underdispersed, i.e., they exhibit too narrow prediction intervals for surface variables like precipitation (Park et al. 2008, Bougeault et al. 2010, Gascon et al. 2019). Typically, bias and underdispersion propagate to the NWP driven hydrological ensemble forecasts (Hemri et al. 2015). Additionally, in low flow conditions (no rain) forecast errors are dominated by hydrological model uncertainty and the initial conditions like soil moisture and snowpack.

Figure 6: Workflow 10-day probabilistic water level forecast for the River Rhine

The statistical post-processing method Ensemble Model Output Statistics EMOS (Gneiting et al. 2005) is applied to estimate the predictive uncertainty of the waterlevel ensemble forecasts produced by the forecasting system, in order to provide probabilistic water level forecasts to the end users. A normal distribution is used to estimate the predictive uncertainty. To avoid physically unrealistic quantiles from the distribution we use a normal distribution truncated on both sides (Hemri et al. 2015, Hemri & Klein

Page 22: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

22

2017), which means that the distribution has a lower and upper water level boundary. In order to achieve approximate normality, the raw ensemble forecasts and the corresponding observations are Box-Cox transformed (Box & Cox 1964, Hemri & Klein 2017).

To estimate the parameters of the statistical post-processing method, a water-level re-forecast data set using archived meteorological forecast data of the period 01.01.2008 – 31.12.2015 as forcings was created using the current water-level forecasting system (“Offline” mode in Figure 6). Waterlevel forecasts were initialized each day at 06:00 UTC. For each meteorological season separate EMOS models have been estimated to account for differences in the respective seasons. In operational forecasting (“Real-Time” mode in Figure 6), the estimated EMOS models are applied to estimate the predictive uncertainty of the operational waterlevel forecasts. For data pre-processing, model runs, post-processing and report generation the Delft-FEWS forecasting system (Werner et al. 2013) is used (see Figure 7).

Figure 7: Time series display of forecasting system FEWS showing the raw (upper panel) and the probabilistic (lower panel) 10-day waterlevel forecast at gauge Kaub initialised at 22. February 2019 06:00UTC

Forecast Product 3.3

Since 18th April 2017 probabilistic 10-day water-level forecasts for the River Rhine are provided to the WP9 stakeholders of IMPREX. Work-daily the new IMPREX-forecasts are available as PDF-report accessible via FTP-Server. The language of the report is German. Figure 8 shows exemplarily the forecast product of the forecast initialised on 22nd February 2019 06:00 UTC for gauge Kaub located near the most important bottleneck along the waterway Rhine.

The IMPREX forecast information is displayed as probabilistic hydrograph, as shown in Figure 8 (left part). Considering the increase in uncertainty with lead-time, we opted to publish the first five days as instantaneous forecast values with hourly time step and to publish the forecasts from day 6 to day 10 as

Page 23: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

23

Deliverable n° 9.4

daily mean values. This temporal aggregation tends to smooth the high uncertainties that may be associated with forecasts issued for longer lead times. Additionally, exceedance probabilities for relevant flood levels (highest shippable water level HSW I and II) and the non-exceedance probabilities for defined low to medium flow water levels, as well as the hydrographs of the raw water-level ensembles are published. Based on user requests, the forecast quantiles are additionally provided as *.csv files since 20th December 2017. This numerical output is used to include the forecast information in the individual decision support systems of the some users.

Figure 8: PDF-Report of the probabilistic 10-day water level forecast for gauge Kaub / Rhine initialised 22. February 2019 06:00 UTC

Verification 3.4

In order to assess the quality of the forecast the assessment framework developed in Deliverable 9.2: “Framework for the assessment of forecast quality and value in the navigation sector” (Klein & Meissner 2017)was applied to the probabilistic water level re-forecast dataset. As an example Figure 9 shows several verification results for gauge Kaub:

− the continuous ranked probability skill score CRPSS (a), − the mean absolute error MAE (b), − the mean error ME (c) and − the relative frequency of absolute error below 20 cm rfae20cm of the probabilistic water level

forecasts as a function of the lead time (d).

Page 24: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

24

MAE, ME and rfae20cm are analysed for the ensemble mean of the raw ensemble (ENSMEAN), the HRES of the ECMWF with a higher spatial resolution and the expected value EEMOS of the predictive distribution post-processed with EMOS. For the verification of rfae20cm, the best single estimator, i.e., the value that would lead to the maximum value of rfae was derived from the probabilistic forecast by maximizing the probability mass of x ± 20 cm (see chapter 4.1 in Deliverable 9.2). The rfae20cm is only analysed for the first five forecast days.

It’s clearly visible that the post-processed ensemble forecast has a smaller bias (analysis of ME) and a higher accuracy (analysis of MAE, rfae20cm) compared to the raw ensemble. This effect is consistently observed along all whole forecast horizon. Especially the bias is significantly reduced by post-processing. This overall improvement of forecast quality is also reflected in the combined measure of forecast quality CRPSS with observed climatology as baseline. The small jumps in the graphs of the raw ensemble at lead time 5 days arise from the drop out of the members of the COSMO-LEPS Ensemble (formerly 16 members now 20 members, forecast length 5 days).

Page 25: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

25

Deliverable n° 9.4

a) Continuous ranked probability score CRPSS b) Mean absolute error MAE

c) Mean error ME d) Rel. frequency of absolute error below 20cm rfae20cm

Figure 9: Continuous ranked probability skill score CRPSS, mean absolute error MAE, mean error ME and relative frequency of absolute error below 20 cm rfae20cm of the probabilistic water level forecasts at gauge Kaub (verification period 1. January 2008 – 31. December 2015) as a function of the lead time. Ensemble: raw ensemble, EMOS post-processed predictive distribution, EEMOS expected value of the predictive distri-bution, ENSMEAN mean of the raw ensemble, HRES deterministic forecast using ECMWF-HRES as forcings

Page 26: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

26

Monthly forecast German waterways Rhine, Elbe, Danube 4

In order to provide forecast information for several weeks ahead, which is needed for long-term planning of transport, logistic and industrial companies (see chapter 1), monthly forecast products are considered in IMPREX, too. Common practice before IMPREX was using observed water-level / flow climatology to support long-term decisions. Therefore climatology is the relevant reference and each new forecast product showing a better skill for the required lead-time than the climatology has an added value. The long-term forecasts are developed for gauges relevant for navigation along the waterway Rhine, Danube and Elbe River (see Figure 2, Table 2).

Different forecast methods are applied for monthly forecasting (see Figure 10). As mentioned before, climatology of observed flow is used as reference to calculate probabilities of categorical forecasts as well as to compare the forecast with the long-term observations. Two different meteorological data sets are used as input to the hydrological model:

1. Observed historical meteorological observations („Climatological“ meteorological forecast): weather trajectories of the past starting at the same day of year as the forecast date are used as a meteorological forecast. The period 1964-2014 is used to create an ensemble of 51 future weather dynamics. The hydrological forecast based on this forcing indicates how the flow situation will evolve based on climatological meteorological conditions. Predictability only arises from initial conditions of the hydrological model. In the literature this type of forecast is well-known as Ensemble Streamflow Prediction (ESP) approach (Wood et al. 2002). This approach could also been used as scenario approach to answer frequently asked questions from users like: “What will happen if we start from current already low conditions and will have meteorological conditions like the extreme year 2003 in the next months?”

2. Numerical weather prediction: hydrological forecast forced by the meteorological ensemble forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) indicates how the flow situation will evolve within the next weeks based on the currently forecasted meteorological conditions.

Page 27: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

27

Deliverable n° 9.4

Figure 10: Forecast Methods used for long-term forecasting in IMPREX

Meteorological Forecast Data 4.1

As meteorological forcing data to calculate monthly flow forecasts the extended-range forecasts from ECMWF-ENS are applied. Twice a week (Monday and Thursday), the ENS model is extended to a lead time up to 46 days by ECMWF. The horizontal resolution for the first 15 days is 0.2°x0.2° (approx. 18 km) and from day 15 to day 46 it is 0.4°x0.4° (approx. 36 km). The ensemble consists of 1 control forecasts and 50 perturbed members, made from slightly different initial atmospheric and oceanic conditions (Owens & Hewson 2018). Re-Forecasts are generated with the same model as used for the operational forecasts for the past 20 years, starting on the same day and month as each real time forecast. The ensemble size is 11-member, which means that in total 20 years x 11 members = 220 forecasts are available for each real time forecast date. The re-forecasts are also created twice a week (Mondays and Thursdays) and are available a week in advance. This re-forecast dataset could be used to correct systematic errors of the real-time forecasts or to estimate parameters of statistical post-processing methods. To assess the skill of the monthly forecasts re-forecasts for the hindcast dates 10th March 2016 – 09th March 2017 generating re-forecasts of the last 20 years have been provided by IMPREX WP 3. In total 2100 forecast dates with 11 ensemble members have been considered.

As observed meteorology used to create the climatological meteorological forecast (ESP approach), precipitation, air temperature and global radiation from the HYRAS data set (Rauthe et al. 2013) available on a 5 km x 5 km grid for the period 1951-2015 is used. This dataset was also used to calibrate the hydrological model LARSIM used to calculate long-term forecasts for waterways Rhine, Danube and Elbe.

Page 28: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

28

Workflow 4.2

The hydrological model used for long-term flow forecasting at BfG is based on the model software LARSIM (Large Area Runoff SImulation Model) (Ludwig & Bremicker 2006). The LARSIM model used in this context is called LARSIM-ME (ME – MittelEuropa = Central Europe). The model domain, the catchments of the rivers Rhine, Elbe, Weser/Ems, Odra and Danube up to gauge Nagymaros in Hungary, is divided in grid based subareas of 5 km x 5 km. Hydrological processes are modelled for each single land use category in the subareas. Due to the strong altitude dependence of temperature subareas are further subdivided in elevation zones for the simulation of the snow processes. The total catchment size, simulated by the model, is approximately 800 000 km². The model calculates flow with a temporal resolution of 1 day using temperature, precipitation and global radiation fields that have been interpolated over the subareas as meteorological input.

The long-term flow forecasts are initialized once a week (Monday) at 00:00 UTC. Real-time meteorological station data (precipitation, temperature and global radiation) is interpolated to the 5 km x 5 km model grid and used as meteorological forcing to initialize LARSIM-ME at the forecast date. For the forecast period two meteorological products are processed (see chapter 4.1): The 51 ensemble members of resampled observations from the past already available on the 5 km x 5 km model grid, whereas the 51 ensemble member of ECMWF-ENS extended have to be interpolated to the model grid.

Furthermore it is planned to use the re-forecast data set of ECMWF-ENS extended in order to drift correct the operational meteorological forecasts, because meteorological long-term forecasts tend to drift towards the model climate with increasing lead-time, giving rise to model bias. Re-forecasts of the last 20 years of the forecast date before, the forecast date and the forecast date after the current forecast date will be used operationally, in total 3 hindcast dates x 20 years x 11 ensemble members = 660 individual forecasts. Additionally the flow re-forecasts driven by the meteorological re-forecasts will be used to estimate the parameters of the statistical post-processing method EMOS (Gneiting et al. 2005) to derive the total uncertainty of the operational flow forecast.

An additional module, which is currently under development, is a statistical error correction of the flow forecasts to reduce the systematic error of the hydrological model. A wavelet based correction method (Bogner & Pappenberger 2011) will be tested for its applicability in the current framework.

Figure 11 shows the workflow of the 6-week flow forecast for the waterways Rhine, Elbe and Danube.

Figure 11: Workflow 6-week flow forecast for the waterways Rhine, Elbe and Danube

Page 29: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

29

Deliverable n° 9.4

For data pre-processing, model runs, post-processing and report generation the Delft-FEWS forecasting system (Werner et al. 2013) is used (see Figure 12).

Figure 12: Time series display of forecasting system FEWS showing the 6 week observed climatological (up-per panel) ESP (middle panel) and the ECMWF-ENS (lower panel) extended forecast initialised at 18. Feb-ruary 2019 for gauge Kaub / Rhine

Forecast Product 4.3

Since 23rd November 2018 6-week flow forecasts are provided to the WP9 stakeholders of IMPREX. Every Tuesday new forecasts are sent via e-mail to the users as PDF-report. The language of the report is German (English version will be used for the IMPREX risk outlook).

As an example Figure 13 (pages 1-2) and Figure 14 (pages 3-4) show the forecast product of the forecast initialised at 18th February 2019 00:00 UTC for gauge Kaub located near the most important bottleneck along the waterway Rhine.

To get a robust estimate of the flow tendency within the next weeks, no absolute values are provided any longer due to the large uncertainties inherent in the hydro-meteorological forecasting chain. Instead of absolute values anomalies referring to the long-term reference climatology are derived. To further reduce the uncertainties and to increase the robustness the forecast values are temporarily aggregated to weekly or monthly means. Hence, long-term forecasts don’t provide information about the flow on a specific date, but they allow estimating flow trends, i.e. if the expected flows tend to be higher or lower than usually observed at this time of year.

The reference climatology of BfG’s hydrological 6-week forecast is the period 1964-2014 and the forecast values are provided as weekly means. From the distribution of the historical reference dataset

Page 30: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

30

of the same time of year five flow classes (quintiles) of equal probability (20 %) are derived, i.e. 20% of the historical values of the respective weeks are falling in each of the following classes: low (0%-20%), slightly lower (20%-40%), normal (40%-60%), slightly higher (60%-80%) and high (80%-100%). For the two flow forecast ensembles forced by the aforementioned meteorological inputs the percentage of ensemble members falling into the respective class are determined. This percentage is interpreted as probability of occurrence.

Additionally the distribution of the mean weekly flows of forecast week 1 to 6 is displayed as box-whisker-plots: the box represents the 25%–75% inter-quantile range, the median is the band inside the box and the whiskers represent the 5% and the 95% quantiles. Outliers are not shown. To classify the tendency of the forecast the distributions of the observed weekly means of the flow of the period 1964-2014 are shown as Box-Whisker plots, too. Although box-whisker-plots are typically used in a more scientific context, several stakeholders provided a positive feedback on this kind of visualization of forecast uncertainty.

Figure 13: PDF-Report of the 6-week forecast for gauge Kaub / Rhine initialised 18. February 2019 00:00 UTC (page 1-2)

Page 31: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

31

Deliverable n° 9.4

Figure 14: PDF-Report of the 6-week forecast for gauge Kaub / Rhine initialised 18. February 2019 00:00 UTC (page 3-4)

Verification 4.4

To assess the skill of the monthly forecasts re-forecasts for the hindcast dates 10.03.2016 – 09.03.2017 generating re-forecasts of the last 20 years have been provided by IMPREX WP 3. It has to be noted that the skill of the real-time forecasts is higher due to the larger number of ensemble members (51 against 11) and higher quality of atmospheric initial conditions. Before the year 2000 the station density of real-time observations is very poor. To minimize the error of model initialization solely forecast dates of the re-forecast data set starting in the year 2000 have been used. Therefore 1699 forecast dates (period 1st Jan. 2000 – 9th March 2016) have been verified in total. Figure 15 shows exemplarily the verification of the Continuous Ranked Probability Skill Score CRPSS (see Deliverable 9.2: “Framework for the assessment of forecast quality and value in the navigation sector” (Klein & Meissner 2017)) at three important gauges located at the waterways Rhine, Danube and Elbe. The CRPSS is a combined quality measure of accuracy, reliability, sharpness and resolution of a probabilistic forecast. As reference forecast the observed climatology was used. A skill score of 1 indicates a perfect forecast, a skill score below zero indicates that climatology is better than the forecast. For all gauges and all lead times the CRPSS is above zero. For gauges Kaub / Rhine and Hofkirchen / Danube the CRPSS of forecast week 5 and 6 is near zero, indicating that the skill of the forecast is similar to the skill of the observed climatology. It is expected that the skill of the forecast increases when the post-processing

Page 32: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

32

methods in development are implemented in the current forecasting system. Overall the skill at waterway Elbe is higher than the skill at the other waterways for most of the lead times.

Figure 15:CRPSS of the forecasted weekly mean flows at Kaub / Rhine (black), Hofkirchen / Danube (blue) and Neu-Darchau / Elbe (red) of the period 02. January 2000 – 09. March 2016

Page 33: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the
Page 34: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

34

Seasonal forecast German waterways Rhine, Elbe, Danube 5

In order to provide forecast information for several months ahead, which is needed for long-term planning of transport, logistic and industrial companies (see chapter 1), seasonal forecast products are considered in IMPREX. As in the case of monthly forecasts, seasonal forecasts are developed for gauges relevant for navigation along the waterways Rhine, Danube and Elbe River (see Figure 2).

The forecast methods applied are the same as in the case of monthly forecasting (see chapter 4). To calculate the forecasts two different meteorological data sets are used as input for the hydrological model:

• Observed meteorological observations („Climatological“ meteorological forecast) • Numerical weather prediction: meteorological ensemble forecasts from the European Centre

for Medium-Range Weather Forecasts (ECMWF)

Meteorological Forecast Data 5.1

As meteorological forcing data to calculate seasonal flow forecasts the seasonal forecasting system SEAS from ECMWF is used. Seasonal forecasts with a lead time of 7 months are provided every month operationally. The fifth generation of ECMWF seasonal forecasting system SEAS5 (ECMWF 2017, Johnson et al. 2018, Owens & Hewson 2018) replaced System 4 in November 2017. The horizontal resolution of the model is 0.4°x0.4° (approx. 36 km). The ensemble consists of 51-members created using a combination of Sea Surface Temperature SST and atmospheric initial condition perturbations and the activation of stochastic physics (ECMWF 2017). Re-Forecasts with the ensemble size of 25 members starting on the 1st of each month for the years 1981-2016 are generated with the same model as used for the operational forecast. Re-forecasts from the old System 4 and the new system SEAS5 provided by IMPREX WP3 have been used to bias-/drift-correct the meteorological seasonal forecast data and to assess the potential skill of seasonal forecasts.

As observed meteorology used to create the climatological meteorological forecast (ESP approach), observed precipitation, air temperature and global radiation from the HYRAS data set (Rauthe et al. 2013) available on a 5 km x 5 km grid for the period 1951-2015 are directly used as model input. This dataset was also used to calibrate the hydrological model LARSIM used to calculate the long-term forecast for waterways Rhine, Danube and Elbe.

Workflow 5.2

The hydrological model LARSIM-ME used for the long-term flow forecasting at BfG is based on the model software LARSIM (details see chapter 4.2).

The long-term flow forecasts are initialized 1st of each month at 00:00 UTC. Real-time meteorological station data (precipitation, temperature and global radiation) is interpolated to the 5 km x 5km model grid and used as meteorological forcing to initialize LARSIM-ME at the forecast date. For the forecast period the two meteorological products are processed: 51 ensemble members of resampled observations from the past already available on the 5 km x 5 km model grid, and 51 ensemble member of SEAS5. The output from SEAS5 (precipitation, air temperature and global radiation), interpolated to a 25 km x 25 km grid (multiple of the 5 km x 5 km model grid), was drift-corrected with the meteorological observation dataset HYRAS used for the calibration of LARSIM-ME. Currently the simple bias correction method linear scaling (Lenderink et al. 2007), already successful applied for drift correction of seasonal forecasts (Crochemore et al. 2016), was used to drift correct the meteorological

Page 35: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

35

Deliverable n° 9.4

forcings. We corrected daily values of the different variables on a monthly basis, which means each daily value of the same month is corrected by the same scaling. As meteorological seasonal forecasts tend to drift towards the model climate with increasing lead-time, giving rise to model bias, separate bias correction factors have been estimated for each forecast initialization date (calendar month) and monthly lead time (month 1 to month 7). In the final step the corrected precipitation, temperature and global radiation are downscaled to the 5 km by 5 km model grid. The drift correction parameters are estimated based on the re-forecast data set of SEAS5 for the period 1981-2016. The flow re-forecasts driven by the meteorological re-forecasts are used to estimate the parameters of the statistical post-processing method EMOS (Gneiting et al. 2005) to derive the total uncertainty of the operational flow forecast.

An additional module currently under development and not operationally yet is a statistical error correction of the flow forecasts to reduce the systematic error of the hydrological model. A wavelet based correction method (Bogner & Pappenberger 2011) is tested for its applicability in the current framework. Figure 16 shows the workflow of the 6-week flow forecast for the waterways Rhine, Elbe and Danube.

Figure 16: Workflow Seasonal flow forecast for the waterways Rhine, Elbe and Danube

As in case of the 10-day and the monthly forecast the forecasting framework Delft-FEWS (Werner et al. 2013) is used for data pre-processing, model runs, post-processing and report generation for seasonal forecasts, too (see Figure 17).

Page 36: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

36

Figure 17: Time series display of forecasting system FEWS showing the observed climatological (upper pan-el) ESP (middle panel) and the ECMWF-SEAS forecast initialised at 01. February 2019 for gauge Kaub / Rhine

Forecast Product 5.3

The seasonal forecast products are quite similar to the ones of the monthly forecast (see chapter 4.3). Due to the large uncertainties monthly mean flows are analysed instead of weekly mean flows. As in case of the monthly forecast product, the probabilities of five flow quintiles (lower, slightly lower, normal, slightly higher, higher) are derived from the ensemble forecasts for the next 6 months to get an impression of the flow tendency within the coming months.

Additionally the distribution of the mean monthly flows of the next months is displayed as box-whisker-plots. To classify the tendency of the forecast the distributions of the observed monthly means of flow of the period 1964-2014 are shown as Box-Whisker plots, too. Figure 18 shows exemplarily the products for the seasonal forecast for gauge Kaub initialised at 01st February 2019.

Due to the limited skill of the seasonal forecasts for Central Europe (see chapter 5.4) with a predictability of 1-2 months it was decided in the course of IMPREX not to pre-operationally publish seasonal forecasts, but to focus on the monthly / 6-week forecasts, which has more potential to be used in daily business of the stakeholders. Nevertheless the aspect of seasonal flow forecasts for the Central European waterways is important and will be a focal point of future work.

Page 37: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

37

Deliverable n° 9.4

Figure 18: Seasonal Forecast for gauge Kaub / Rhine initialised 01. February 2019 00:00UTC. Upper fig-ure: categorical forecast of five flow classes, lower figure: probability distribution of mean monthly flows

Verification 5.4

To assess the skill of the seasonal forecasts re-forecasts of System 4 and SEAS5 for the hindcast period 01.01.1981 – 31.12.2016 have been provided by IMPREX WP 3. Using these re-forecasts as meteorological forcings a seasonal flow re-forecast dataset was created. Figure 19, Figure 20 and Figure 21 show exemplarily the verification using the correlation coefficient between the ESP-based as well as the System4 and SEAS5-based forecasts and the observation for the gauges Kaub / Rhine, Hofkirchen / Danube and Neu Darchau / Elbe as a function of forecast horizon (month 1 to month 6) and initialization month (January to December). Dark coloured pixels indicate high forecast skill. It is obvious that for both approaches and all stations, the skill significantly diminishes with increasing lead-time, but that the use of the System4 forecasts and SEAS5 forecasts leads to additional skill for the majority of lead-times and initialization months at all gauges. The spring forecasts have the highest

Page 38: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

38

skill also for longer lead times due to the snow memory. Overall, the forecast skill for the Elbe (gauge Neu Darchau) is higher than for Rhine (gauge Kaub) and Danube (gauge Hofkirchen). No significant improvement between the old System4 and the current system SEAS5 can be observed. Due to the limited skill of the seasonal forecasts with a predictability of 1-2 months it was decided to not publish the seasonal forecasts and to concentrate on the 6-week forecast described in chapter 4.

Figure 19: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Kaub / Rhine: climatological meteoro-logical forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast pe-riod analysed 1981-2016

Figure 20: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Hofkirchen / Danube: climatological meteorological forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast period analysed 1981-2016

Figure 21: Correlation coefficient between the forecasted mean monthly flows and the observed mean monthly flows for different meteorological forecast products at gauge Neu-Darchau / Elbe: climatological meteorological forecast (ESP) left panel, ECMWF System 4 middle panel, ECMWF SEAS5 right panel. Hindcast period analysed 1981-2016

Page 39: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the
Page 40: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

40

Summary 6

Within IMPREX several pre-operational forecast products have been developed to support improved transport cost planning in the inland waterway transport sector. Besides the forecasts products, which are the visible output of IMPREX for the stakeholders, deliverable D9.4 describes the underlying forecasting systems, which have been set-up in the course of IMPREX at BfG. These forecasting systems form the backbone of future forecasting services providing users with specific forecast products. All forecast products presented in D9.4 have been co-designed with the WP9 stakeholders of IMPREX.

The short-to-medium term water-level forecasts are useful to optimize transport management (ship loads, timing of transports), short-term stock management of companies relying on waterway transport, and waterway maintenance (e.g. dredging). Monthly to seasonal forecast products are required for medium- to long-term planning and optimization of the water bound logistic chain (stock management, fleet composition, adjustment of the industrial production chain, modal split planning), as well as for the sediment management of harbours and tidal influenced waterways.

In order to provide forecasts on different time scales supporting decision making in IWT a 10-day probabilistic waterlevel forecast was developed for the River Rhine and the related forecast products are published work-daily since April 2017 to WP9 stakeholders of IMPREX. It is the first time that probabilistic forecasts with a lead time of 10 days are available for navigational users of the River Rhine.

For the long-term planning monthly to seasonal forecast products have been developed and analysed. Due to the limited skill of seasonal forecasts in Central Europe with a predictability of 1-2 months at maximum it was decided to focus on the 6-week forecast using ECMWF-ENS as meteorological forcings. This 6-week forecast showed a higher potential to be used in daily business of the stakeholders, which is essential to get a sound feedback on the usefulness of such products. It became obvious that ECMWF-ENS has several advantages compared to the seasonal meteorological forecast ECMWF-SEAS5. In the first 14 days it has a significantly higher spatial resolution (18 km compared to 36 km), it uses the most recent model version of the IFS forecasting model and forecasts are issued twice a week compared to once a month.

Since November 2018 – in the critical phase of the extreme low flow situation all over Europe – the 6-week flow forecasts are provided to the WP9 stakeholders of IMPREX pre-operationally on every Tuesday. This is the first time that long-term forecasts are provided to navigational users at the river Rhine, Elbe and Danube in order to optimize management procedures on a longer term. But it has to be mentioned that in order to improve skill and to provide additional forecast products to the end users the 6-week forecasts have to be enhanced within the coming months (even beyond the project life span of IMPREX) including technical improvements bias/drift correction of the meteorological input based on re-forecasts, error correction of the forecasted flows and statistical post-processing to estimate the predictive uncertainty. At the second IMPREX WP9 transport user workshop in Koblenz on 21st March 2019, all the existing forecast products and possible future forecast products have been discussed with the users to ensure the usefulness of the forecast products developed in the context of IMPREX. The pre-operational probabilistic 10-day waterlevel forecast and the 6-week flow forecast for the waterway described in section 3 and 4 are heavily used for medium-term and long-term planning by the users, especially in the extreme low flow period 2018. The users are especially satisfied with the quality of the 10-day forecast provided since April 2017. For the 6-week forecasts, provided since November 2018 to

Page 41: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

41

Deliverable n° 9.4

the users, the period was too short to gain in depth experience with the product. Nevertheless the product is already used in decision making by several users.

In the beginning of IMPREX there were a lot of hesitations about the usefulness of extended forecast horizons, but since at least the extreme low flow situation in 2018 the extended forecasts are actively used by an increasing number of stakeholders from different sectors (19 companies and institutions, i.a. Waterway and Shipping Administration WSV, Federal Ministry of Transportation and Digital Infrastructure BMVI, BASF, EnBW, TransnetBW, Imperial, ...). At it seems that the more and more parties become aware that hydrological forecast covering various time-scales is a useful, maybe even crucial information for recent IWT.

All pre-operational forecast products developed within the context of IMPREX have been implemented in the operational forecasting system Delft-FEWS. This facilitates a possible shift of the test and pre-operational forecast products to operational services of BfG after the end of IMPREX.

Page 42: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

42

References

Bergström, S. (1995): The HBV model. In: V. P. Singh (Ed.): Computer models of watershed hydrology. Water Resources Publications, Colorado, USA, 443-476

Bogner, K. & F. Pappenberger (2011): Multiscale error analysis, correction, and predictive uncertainty estimation in a flood forecasting system. Water Resources Research 47

Bougeault, P., Z. Toth, C. Bishop, B. Brown, D. Burridge, D. H. Chen, B. Ebert, M. Fuentes, T. M. Hamill, K. Mylne, J. Nicolau, T. Paccagnella, Y.-Y. Park, D. Parsons, B. Raoult, D. Schuster, P. S. Dias, R. Swinbank, Y. Takeuchi, W. Tennant, L. Wilson & S. Worley (2010): The THORPEX Interactive Grand Global Ensemble. Bulletin of the American Meteorological Society 91(8), 1059-1072

Box, G. E. P. & D. R. Cox (1964): An analysis of transformations. Journal of the Royal Statistical Society Series B 26(2), 211-243

Broersen, P. & A. Weerts (2005): Automatic Error Correction of Rainfall-Runoff models in Flood Forecasting Systems. Conference Proceedings: IMTC 2005 – Instrumentation and Measurement Technology Conference, Ottawa, Canada, 17-19 May 2005.

CCNR (2016): Annual Report 2016 - Inland Navigation in Europe - Market Observation. Central Commission for the Navigation of the Rhine, Strasbourg, France, http://www.ccr-zkr.org/files/documents/om/om16_II_en.pdf

Crochemore, L., M. H. Ramos & F. Pappenberger (2016): Bias correcting precipitation forecasts to improve the skill of seasonal streamflow forecasts. Hydrol. Earth Syst. Sci. 20(9), 3601-3618

Deltares (2012): Technical Reference SOBEK-RE. Deltares, Delft, The Netherlands

ECMWF (2017): SEAS5 user guide - Version 1.1. ECMWF, Reading, UK

Gascon, E., D. Lavers, T. M. Hamill, D. S. Richardson, Z. Ben Bouallegue, M. Leutbecher & F. Pappenberger (2019): Statistical post-processing of duel-resolution ensemble precipitation forecasts across Europe. Quarterly Journal of the Royal Meteorological Society (submitted)

Gneiting, T., A. E. Raftery, A. H. Westveld & T. Goldman (2005): Calibrated probabilistic forecasting using ensemble model output statistics and minimum CRPS estimation. Monthly Weather Review 133(5), 1098-1118

Hemri, S. & B. Klein (2017): Analog-Based Postprocessing of Navigation-Related Hydrological Ensemble Forecasts. Water Resources Research 53(11), 9059-9077

Hemri, S., D. Lisniak & B. Klein (2015): Multivariate post-processing techniques for probabilistic hydrological forecasting. Water Resources Research

Johnson, S. J., T. N. Stockdale, L. Ferranti, M. A. Balmaseda, F. Molteni, L. Magnusson, S. Tietsche, D. Decremer, A. Weisheimer, G. Balsamo, S. Keeley, K. Mogensen, H. Zuo & B. Monge-Sanz (2018): SEAS5: The new ECMWF seasonal forecast system. Geosci. Model Dev. Discuss. 2018, 1-44

Klein, B. & D. Meissner (2016): Vulnerability of Inland Waterway Transport and Waterway Management on Hydro-meteorological Extremes. Deliverable 9.1, IMPREX - Improving Predictions of Hydrological Extremes - Grant Agreement Number 641811, http://www.imprex.eu/system/files/generated/files/resource/d9-1-imprex-v2-0.pdf

Klein, B. & D. Meissner (2017): Framework for the assessment of forecast quality and value in the navigation sector. Deliverable 9.2, IMPREX - Improving Predictions of Hydrological Extremes - Grant

Page 43: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

43

Deliverable n° 9.4

Agreement Number 641811, http://www.imprex.eu/system/files/generated/files/resource/d9-2-imprex-v2-0.pdf

Lenderink, G., A. Buishand & W. van Deursen (2007): Estimates of future discharges of the river Rhine using two scenario methodologies: direct versus delta approach. Hydrology and Earth System Sciences 11(3), 1143-1159

Leutbecher, M. & T. N. Palmer (2008): Ensemble forecasting. Journal of Computational Physics 227(7), 3515-3539

Lindstrom, G., B. Johansson, M. Persson, M. Gardelin & S. Bergstrom (1997): Development and test of the distributed HBV-96 hydrological model. Journal of Hydrology 201(1-4), 272-288

Ludwig, K. & M. Bremicker (2006): The Water Balance Model LARSIM –Design, Content and Applications. 22. C. Leibundgut, S. Demuth and J. Lange (Eds), Freiburger Schriften zur Hydrologie, Institut für Hydrologie, Universität Freiburg im Breisgau, Freiburg, 141 pp.

Marsigli, C., A. Montani & T. Paccagnella (2014): Perturbation of initial and boundary conditions for a limited-area ensemble: multi-model versus single-model approach. Quarterly Journal of the Royal Meteorological Society 140(678), 197-208

Meißner, D., B. Klein & M. Ionita (2017): Development of a monthly to seasonal forecast framework tailored to inland waterway transport in central Europe. Hydrol. Earth Syst. Sci. 21(12), 6401-6423

Meissner, D. & S. Rademacher (2010): Die verkehrsbezogene Wasserstandsvorhersage für die Bundeswasserstraße Rhein. KW Korrespondenz Wasserwirtschaft 3(9), 485-491

Montani, A., D. Cesari, C. Marsigli & T. Paccagnella (2011): Seven years of activity in the field of mesoscale ensemble forecasting by the COSMO-LEPS system: main achievements and open challenges. Tellus Series a-Dynamic Meteorology and Oceanography 63(3), 605-624

Owens, R. & T. R. E. Hewson (2018): ECMWF Forecast User Guide. ECMWF, Reading, doi: 10.21957/m1cs7h

Park, Y. Y., R. Buizza & M. Leutbecher (2008): TIGGE: Preliminary results on comparing and combining ensembles. Quarterly Journal of the Royal Meteorological Society 134(637), 2029-2050

Rauthe, M., H. Steiner, U. Riediger, A. Mazurkiewicz & A. Gratzki (2013): A Central European precipitation climatology - Part I: Generation and validation of a high-resolution gridded daily data set (HYRAS). Meteorologische Zeitschrift 22(3), 235-256

Werner, M., J. Schellekens, P. Gijsbers, M. van Dijk, O. van den Akker & K. Heynert (2013): The Delft-FEWS flow forecasting system. Environmental Modelling & Software 40, 65-77

Wood, A. W., E. P. Maurer, A. Kumar & D. P. Lettenmaier (2002): Long-range experimental hydrologic forecasting for the eastern United States. Journal of Geophysical Research-Atmospheres 107(D20)

Page 44: SEMI-OPERATIONAL FORECASTING SYSTEM FOR RHINE, … · inland waterway transport to improve transport cost planning and the competitiveness of the inland navigation sector within the

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant agreement N° 641811

44

IMPREX has received funding from the European Union Horizon 2020 Research and Innovation

Programme under Grant Agreement N° 641811