use changes and live management operations such as ... watershed model (SWAT) using weather forecasts ... the watershed for weather station data should.
H31F-0726 Cornell University Cornell University Soil & Water Research Group
Hydrological Modeling Where No Meteorological Stations Exist SOIL & WATER LAB
Study Overview, cont.
Assuming nearest reporting stations for PPT.
• ECMWF • NCEP
Closest Weather Stations. Variables.
Summer 2007 KALB 2.0
Gridded Forecast Models from TIGGE
1.8
Date range: 4/1/2007 – present
1.2
• 0Z runs, model hours 06 though 30
Predicted
Results CFSR Results Town Brook
• 0.45deg Res. Global. • 50 perturbations. • Median from 50 member pt forecast.
50
a) CFSR NSE=0.627
Predicted Flow (mm/d) 20 30 40 10 0 Predicted Flow (mm/d) 20 30 40
50
b) test8[, 1] * 24 * 60 * 60/37000 Ideal Met. NSE=0.524
0
10
Gumara Watershed • Tana Basin • 1200km2 Town Brook watershed
c) test8[, 1] * 24 * 60 * 60/37000 Easton et al. 2008 NSE=0.564
50
• New York State. • 37km2 catchment of Cannonsiville basin
Cross River catchment • 44km2 catchment of NYC water supply SWAT 2005, initialized with ArcSWAT.
0
10
20 30 40 ObservedFlow Flow(mm/day) (mm/d) Observed
Below, predicted vs. observed stream flow graph with the furthest distance station in the background in darkest/largest symbols and closer stations represented by lighter/smaller symbols. Legend shows the location along with locations respective NSE.
Yemen
Gumara Uganda
Kenya
0.01 - 1.85 1.85 - 4.50 4.50 - 8.28 8.28 - 14.10 14.10 - 25.94
Optimal Predicted Flow (mm/d) 20 30
PPT(dm/yr)
CFSR performs better than all regional weather stations as represented by the better linear relationship with the lightest blocks in the graphic.
10
Ethiopia
CFSR 0.68 Danburry 0.54 Dutchess 0.46 Westchester 0.29 Monticello 0.17
40
Djibouti
Somalia
Goals
50
Left, three scatter plots showing the optimal calibrations of SWAT on the Town Brook watershed. At the top a) the optimal calibration results using reanalysis data from CFSR, giving the best results when compared to an “Ideal meteorological station” calibration shown in the center b) and the previously published ideal calibration from Easton et al. 2008 shown in c). All calibrations were performed using the same period.
CFSR Results Cross River Saudi Arabia
Lake Tana
0
10
20 30 Observed Flow (mm/d)
40
Acknowledgements and Interesting Side Points
Hypothesis Gridded weather models have some skill, thus substituting short range forecasts centered in the watershed for weather station data should better represent conditions in the watershed than stations at some X distance away. Question: What is this distance?
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
1.6
1.8
2
1.60
1.80
Measured
1.8
Study Watersheds and Model
US has: • ~60km spacing between real-time stations Africa • 600km-1000km regularly spaced stations in region
0.0
2.0
• 1.0deg Res. Global. • 20 perturbations. • Median from 20 member pt. forecast.
Sudan
0.2
Summer 2007 ECMWF Ens
NCEP Global Ensemble Forecast System (GEFS).
450 km
0.8
0.4
ECMWF Ensemble.
300
1.0
0.6
Eritrea
Need consistent forcing variables (ppt, temp, etc). • And no missing time steps We would like to produce short term forecasts. with management custom dictionary to Geospatial We want tomapping develop “live” tools. characterize ditches. allocation Parametersplanning included: • Live irrigation Applicable world wide.
y = 1.452x R² = -0.0689
1.6
1.4
0 75 150
Study Overview
44 37 1200
Best Met NSE 0.54 0.52 0.68
CFSR NSE 0.68 0.63 0.70
• PPT, as well as Max and Min Temperature.
.23
Finding a set of stations to run with real-time at 30km is unlikely. This type of agreement would be a dream situation for us!
Cross River Town Brook Gumara
Best Met Distance (km) 13 0/3 0/30
TIGGE Ensemble Results
Ethiopian Watersheds
• Mehta et. al. 2004.
Location
Watershed Area (km2)
0
• 30km apart, PPT
r2
Climate Forecast System Reanalysis (CFSR) • Hourly 1979 – Today TIGGE Ensemble Archive.
Predicted Flow (mm/d) 20 30 40
Spatial Issues
Best Meteorological Station vs CFSR
10
Issues in Hydrologic Modeling
Data Overview
0
Hydrologic modeling at daily or sub daily resolution requires accurate precipitation and temperature data, which is particularly important when planning land use changes and live management operations such as irrigation allocation. Weather gage data pose a variety of challenges, including the fact that they are nonexistent in many watersheds of interest. We show that by integrating global forecast products and reanalysis datasets from the atmospheric modeling community into the hydrologic models commonly used, we can circumvent some of the issues with point-of-measure weather gauges. Traditionally in hydrologic forecasting, a precipitation forecast is used from a nearby precipitation station on which the hydrologic model has been calibrated. Here we present the results of two studies. First, we compare calibrating a watershed model (SWAT) using weather forecasts from ensembles of raw gridded atmospheric models interpolated to the center of subbasins against using the closest weather gage measurements. Second, we perform a similar study using global reanalysis datasets. In addition, we look at what scales and radii using direct gridded model output may introduce equal or less error to watershed modeling projects than using the closest station.
Results, cont.
New York Watersheds
Authors would like to acknowledge ECMWF, who supplied the data for this study via the TIGGE online user interface. http://tigge-portal.ecmwf.int/ All computation for this study was performed in the publicly accessible Amazon EC2 grid. Authors are looking for collaborators who might wish to run similar studys without the pain of extracting the datasets themselves.
y = 1.139x R² = 0.36162
1.6
1.4
Predicted
Introduction
Fuka, D.R., MacAlister, C. Demissie, S.S., Wale, A., Walter, M.T., Easton, Z.M., and Steenhuis, T.S.
1.2
1.0
0.8
0.6
0.4
0.2
0.0 0
0.2
0.4
0.6
0.8
1
1.2
1.4
Measured
Summer 2007 NCEP GEFS 2.0
1.8
y = 0.7452x R² = 0.61175
1.6
1.4
Predicted
Biological & Environmental Engineering
1.2
1.0
0.8
0.6
0.4
0.2
0.0 0.00
0.20
0.40
0.60
0.80
1.00
Measured
1.20
1.40
2.00
Left, three scatter plots showing the optimal calibrations of SWAT on the Town Brook watershed using NCEP and ECMWF forecast ensemble means. At the top the optimal calibration results using the closest real-time operational met station KALB, center, using the ECMWF ensemble, and bottom, using the NCEP global ensemble. Forecasts of precipitation and temperature for hours 6 through 30 were used as forcing variables.
Discussion There is validity to using reanalysis datasets in place of weather station data to force hydrologic modeling systems when nearby weather stations are not available or have inconsistent records There is validity to using short term forecast archives in place of weather station data to force hydrologic modeling systems. The distance to the where the model skill produces the same accuracy as the weather stations is closer than we thought. Variables from the atmospheric models are physically linked, and we should likely keep them that way.
Conclusion This study found that forecast archives and CFSR reanalysis data provide a potentially valuable alternative to “nearest weather station” data for hydrological modelers. Indeed, this study found that the CFSR data may actually be a better representation of the weather than point observations, at the scale of small watersheds (
