Impact of Point Rainfall Data Uncertainties on SWAT Simulations Michael Rode & Gerald Wenk Department of Hydrological Modelling Magdeburg, Germany 4th International SWAT Conference 2007 4.- 6. July UNESCO-IHP University, Delft The Netherlands
Motivation Input data uncertainties are increasingly recognised Rainfall data are most important input data for rainfall runoff models Point rainfall data are associated with systematic and random errors Seite 2
Systematic point rainfall measurement errors Mean correction (%) of the average annual precipitation total (1961/90) Moderate wind-sheltered sites in Germany Source: (WMO, 1998) Seite 3
Mean correction (%) of precipitation in the Weiße Elster River Basin Shelter class Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year a 31.6 33.5 26.9 18.3 12.5 10.4 10.8 10.5 12.6 15.5 21.8 26.5 18.2 b 23.3 24.5 20.3 15.1 11.2 9.8 10.0 9.5 11.5 12.7 16.8 19.8 14.6 c 17.3 17.9 15.5 12.7 10.1 8.8 9.1 8.5 10.2 11.0 13.3 15.0 12.0 d 11.5 11.8 10.7 10.0 8.6 7.7 8.0 7.5 8.7 8.8 9.5 10.3 9.3 Time series 1961/90, according to Richter (1995) Seite 4
Objectives Investigate the impact of systematic and random rainfall point measurement errors Assess simulated discharge and nitrogen with SWAT Analyse scaling effects of these errors Seite 5
The Weisse Elster Basin: Study area Overview Catchment area: 5360 km² River length: 253 km Mean discharge: 25.2 m 3 /s Rainfall gauge stations: 49 Discharge gauge stations: 18 Land use Agricultural land: 62% Forest: 23% Urban areas: 13% Seite 6
Methodological approach Add randomly generated correctionvalues to uncorrected rainfall measurement values Assume Gumbel error distribution of PDFs Standard deviations of PDFs are defined by the correction factor Generate 200 time series for each rainfall gauge station (DUE) Compare SWAT simulations with respect to variables and scales Seite 7
Data Uncertainty Engine (DUE) Characterisation and assessment of uncertainty in data Generates time series including systematic and random errors Monte Carlo based approach using pdf s Seite 8
Calibration discharge gauge stations N Restgebiet Oberthau Thekla CatchmentsofDischargeGauges inthe WeisseElsterRiverBasin usedforswat-calibration Zeitz Kleindalzig Böhlen Catchment of Discharge Gauge Adorf Böhlen Gera-Langenberg Greiz Gera-Langenberg Gößnitz Gößnitz Kleindalzig Läwitz Oberthau Läwitz Greiz Restgebiet Straßberg Thekla Zeitz Rivers Discharge Gauge Straßberg Water Quality Gauge Adorf 0 10 20 30 40 50 Kilometers Seite 9
SWAT calibration discharge gauge station Läwitz (98 km²) 7 6 Discharge Läwitz (obs.) Discharge Läwitz (sim.) Reasonable calibration results Discharge [m³/s] 5 4 3 2 Problems to represent the discharge dynamics 1 0 01.11.1996 01.03.1997 01.07.1997 01.11.1997 01.03.1998 01.07.1998 Seite 10
SWAT calibration discharge gauge station Zeitz (2504 km²) 140 Discharge Zeitz (obs.) Discharge Zeitz (sim.) 120 100 Discharge [m³/s] 80 60 40 20 0 01.11.1997 01.05.1998 01.11.1998 01.05.1999 01.11.1999 01.05.2000 Seite 11
SWAT calibration of DIN load, gauge station Gera-Langenberg (2186 km²) 80 Gera-Langenberg measured simulated Reasonable agreement between observed and simulated DIN loads DIN - load [t/d] 60 40 20 Slightly overestimated nitrogen loads Oversimplified representation of nitrate denitrification in the aquifer 0 Oct-95 Apr-96 Oct-96 Apr-97 Oct-97 Apr-98 Oct-98 date Seite 12
Mean simulated discharge using four different correction factors runoff [mm/month] 125 Läwitz 100 75 50 Mean (9,3 %) Mean (12,0 %) Mean (14,6 %) Mean (18,2 %) Systematic errors of mounthly values Large differences in the case of low flows 25 0 1990 1992 1994 1996 1998 2000 Seite 13
Mean simulated discharge using four different correction factors 100 75 Gera-Langenberg Mean (9,3 %) Mean (12,0 %) Mean (14,6 %) Mean (18,2 %) Comparable differences with increasing catchment size runoff [mm/month] 50 25 0 1990 1992 1994 1996 1998 2000 Seite 14
Mean simulated nitrogen using four different correction factors Total Nitrogen Load [kg/(ha*month)] 15,0 12,5 10,0 7,5 5,0 2,5 Läwitz Mean (9,3 %) Mean (12,0 %) Mean (14,6 %) Mean (18,2 %) Small effects on simulated nitrogen loads 0,0 1990 1992 1994 1996 1998 2000 Seite 15
Mean and maximum monthly error ranges of simulated discharge Discharge [mm/month] 50 40 30 20 10 0 0 10 20 30 40 50 Number of Weather Stations Mean (Range) Max (Range) Correction factor of 18.2% Randomly generated rainfall time series Gumbel distribution Decrease of errors with increasing rainfall gauge stations Seite 16
Mean and maximum monthly error ranges of simulated nitrogen Total Nitrogen Load [kg/(ha*month)] 5 Mean (Range) Max (Range) 4 3 2 1 Considerable mean errors only when using small numbers of stations Maximum errors in single month can still be significant 0 0 10 20 30 40 50 Number of Weather Stations Seite 17
Conclusions Systematic rainfall measurement errors can have considerable impact on simulated discharge and nitrogen These errors can be increased by random rainfall errors Effect of random error rapidly decreases with increase rainfall stations Nitrogen load calculations are much less sensitive to random precipitation errors than simulated discharge Seite 18