HYPE model simulations for non-stationary conditions in European medium sized catchments

HYPE model simulations for non-stationary conditions in European medium sized catchments

Göran Lindström & Chantal Donnelly, SMHI, SwedenIAHS, 2013-07-23, Göteborg, Sweden.Hw15 - Testing simulation and forecasting models in non-stationary conditions

After the Gudrun stormJanuary 2005

Photo: H.Alexandesson

OutlineObjectives Simulate non-stationary conditions, for this workshop. Evaluate effects of the Gudrun storm in 2005.

Modeled basins Garonne (France, increase in temperature, decrease in discharge) Durance (France, increase in temperature, decrease of glacier) Lissbro (Sweden, forest loss due to Gudrun storm)

HYPE modelHydrological Predictions for the Environment

Simulates daily fluxes and turn-over of water, Nitrogen & Phosphorus Integrated soil- and groundwater, substances follow water flow paths Developed for large-scale applications Routing in rivers & lakes (incl. regulation) Parameters are linked to soil type or land-use, and calibrated Each combination of soil type and land-use is modeled separately First version was developed in 2005-2007, and continuously developed Potential evaporation by air temperature with seasonally varying factor

Soil types Land use SLC+

Soil/Land Use classes (SLC)Most parameters coupled to soil or land-use

HYPE in the world

Durance and Garonne –

- 35000 subbasins

- Median size 215 km2

- Used for hindcasting, operational forecasting and future climate predictions, Q, Nitrogen and Phosphorous

Taken from the E-HYPE pan-European application of the HYPE model

For Durance and Garonne:• Local model taken from E-HYPE (subbasin delineation, landuse, soil-type,

lakes, glaciers, irrigation etc)• Used the local forcing data (but with height adjustment to height of each

subbasin in catchment for temperature)• Calibrated to given Q data by adjusting ’super-parameters’ (also Precip

correction where required)

Trends over data period:Observed Trends:

Simulated vs Observed Trends:

Modeled decrease slightly too weak

Durance Catchment area: 2170 km²

Pmean Tmean Qmean NSE RE (%)1971-1980 1608 2.8 53 0.67 0

1998 -2007 1150 4.1 45 0.55 +2

Garonne Catchment area: 9980 km²

HYPE underestimated the decrease in discharge

Temperature increase not the only cause of decreasing discharge? Temperature increased by ~1.2 ºC (whole period) Precipitation decreased by ~8% (whole period) Data uncertainties? Regulation, irrigation?

Pmean Tmean Qmean NSE RE (%)1971-1980 1232 8.5 222.6 0.78 -1

1998-2007 1086 10.0 155.5 0.50 +12

Does glacial melt in the Durance catchment explain non-stationarity?

Glacier = 8 % of area Glacier = 1 % of area

S-HYPE model for Sweden For support to

implementation of EU Water Framework Directive, forecasting etc.

~ 35000 subbasins, ~15km2 subbasin resolution

Cal/Eval at 400 stations Interpolation

Runoff and discharge

Gudrun storm, January 2005 About 70 M m3 of trees were blown down. 18 people died (in Sweden) Three worst storms in Sweden: 1902, 1969 and 2005 In a region affected by a summer flood in 2004 Worries about increased flood risk after loss of forest Also known as Erwin storm

Gudrun storm January 2005 In the worst hit areas ~8 % of trees were blown down.

Max wind speed (m/s) Loss of forest (m3/ha)

Lissbro

Lissbro97 km2, 81% forested, 1 % lakes

1995 1996 1997 1998 1999 2000 2001 2002 2003 20040

4

8

122004

summer flood

The 10 years before Gudrun

Previous HBV study of clearfellingBrandt et al. (1988), small-scale experiments, central Sweden

Discharge: +165-200 mm/year

Lissbro, Reference (no change in model)4 key parameters adjusted to Lissbro data

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 20100

2

4

6

8

10

12 ComRec

Q (m3/s)

P (mm)

020406080

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

-400-2000200400

AD (mm)

CalibrationBefore storm

"Validation"After storm

NSE=0.86VE=+1%

NSE=0.89VE=-5%

Lissbro, Simulated clearfelling8% of the forest converted to clearfelling

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 20100

2

4

6

8

10

12 ComRec

Q (m3/s)

P (mm)

020406080

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

-400-2000200400

AD (mm)

CalibrationBefore storm

"Validation"After storm

NSE=0.86VE=+1%

NSE=0.89VE=-3%

Change in SLC classes

Lissbro, Decreased PotEvapForest PET ~15% higher than open areas (8% forest loss)

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 20100

2

4

6

8

10

12 ComRec

Q (m3/s)

P (mm)

020406080

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

-400-2000200400

AD (mm)

CalibrationBefore storm

"Validation"After storm

NSE=0.86VE=+1%

NSE=0.89VE=-4%

Change in PET parameter

Conclusions Trends in discharge were fairly well captured by the HYPE model for the two

French basins (but modeled discharge decrease was too weak in Garonne). Glacier development had negligible effect in Durance. The effects of the Gudrun storm on discharge in Lissbro were very small (within

the uncertainty in the model calibration period).