1 Empirical Models Based on the Universal Soil Loss Equation Fail to Predict Discharges from Chesapeake Bay Catchments Boomer, Kathleen B. Weller, Donald E. Jordan, Thomas E. of the Smithsonian Environmental Research Center Journal of Environmental Quality, 2008
75
Embed
1 Empirical Models Based on the Universal Soil Loss Equation Fail to Predict Discharges from Chesapeake Bay Catchments Boomer, Kathleen B. Weller, Donald.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Empirical Models Based on the Universal Soil Loss Equation Fail to Predict Discharges from Chesapeake Bay Catchments
Boomer, Kathleen B.Weller, Donald E.Jordan, Thomas E.of the Smithsonian Environmental Research Center
Journal of Environmental Quality, 2008
2
Presentation Overview
3
1.Abstract2.Background Information3.Methods
1. Location2. Water Quality Data3. Spatial Data4. Data Analysis
Goal: Accurately predict sediment loads/yields in un-gauged basins.
Methods: Test the most widely used equation, USLE with accurate water quality data significant number of catchments from 2 different agencies. Also attempt a multiple linear regression approach.
Results: The USLE and all its derivatives perform very poorly, even using SDR’s. So does the multiple linear regression.
Conclusion: USLE & multiple linear regression are not advised.
7
2. Background Information
8
2. Background Information
A = R K LS CPThe Universal Soil Loss Equation
A = estimated long-term annual soil loss (Mg soil loss ha−1 yr−1)R = rainfall and runoff factor representing the summed erosive potential of all rainfall events in a year(MJ mm ha−1 h−1 yr−1)L = slope length (dimensionless)S = slope steepness (dimensionless)K = soil erodibility factor representing units of soil loss per unit of rainfallerosivity (Mg ha h ha−1 MJ−1 mm−1)CP = characterizes land cover and conservation management practices (dimensionless).
9
2. Background Information
The Revised Universal Soil Loss Equation 2
incorporate a broader set of land coverclasses and attempt to capture deposition in complex terrains
More sub-factors
Daily time step
10
Universal
Soil
Loss
Equation
= Edge of Field
2. Background Information
11
Universal
Soil
Loss
Equation
= Catchment Scale
2. Background Information
12
Sediment
Delivery
Ratios
2. Background Information
13
Sediment
Delivery
Ratios
2. Background Information
1. Estimate from calibration dataor2. Use complex spatial algorithms
Yagow 1998SEDMOD
Exported fromField
Observed atWQ site
Transport
14
2. Background Information
Models that rely on USLE for calibration
GWLF (Generalized Watershed Loading Function; Haith and Shoemaker, 1987)
AGNPS (AGricultural Non-Point Source; Young et al., 1989)
SWAT (Soil & Water Assessment Tool; Arnold and Allen, 1992)
HSPF (Hydrological Simulation Program-Fortran; Bicknell et al., 1993)
SEDD (Sediment Delivery Distributed model; Ferro and Porto, 2000).
15
2. Background Information
DANGER!!!
GROSS EROSION VS. SEDIMENT TRANSPORT
FIELD OBSERVATIONS VS. REGIONAL SPATIAL DATA
Van Rompaey et al., 2003 – 98 catchments in europe, poor results
Wischmeier and Smith 1978; Risse et al., 1993; Kinnell, 2004a) – Not for Catchment
16
3. Methods
17
3. Methods
Water Quality Data
18
3. Methods
SERC DATAcontinuous monitoring stations in 78 basins within 166,000 km2
USGSS DATAcontinuous monitoring stations in 23 additional basins within 166,000 km2 Chesapeake Bay watershed
101-90,530 ha
no reservoirs
Water Quality Data
21
0 20 40 60 80 100
forest
agriculture
residential/commercial 0-30%
5-100%
0-40%
USGS DATA
3. Methods Water Quality Data
22
SERC WATER QUALITY DATA
3. Methods
•Automated samplers, continuous stage,
•flow-weighted water samples composited weekly
•<= 1 year, 1974-2004
•Annual mean flow rates * flow-weighted mean conc = annual avg loads
•Yield=load/area
USGS WATER QUALITY DATA
•Samples collected daily or determined by ESTIMATOR model
Water Quality Data
23
3. Methods Regional Spatial Data
Source Year ResolutionConverted Resolution
USGS National Elevation Dataset 1999 27.78 m 30 m
USGS National Landcover Database 1992 30 m
USDA-NRCS STATSGO Soils Database 1995 1:250,000 30 m
RESAC Dataset (% impervious) 2003 30 m
Spatial Climate Analysis Service 2002 1:250,000
24
3. Methods Regional Spatial Data
USLE Analysis
25
3. Methods USLE AnalysisGRID-BASED USLE ANALYSIS
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
R (rainfall erosivity) =
Derivded from linear interpolationof national iso-erodent map
26
3. MethodsGRID-BASED USLE ANALYSIS
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
K (surface soil erodibility)=
STATSGO resampled to 30m
USLE Analysis
27
3. MethodsGRID-BASED USLE ANALYSIS
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
L (slope length) =
NED DEM resampled to 30 m
USLE Analysis
28
3. MethodsGRID-BASED USLE ANALYSIS
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
S (slope steepness)=
NED DEM resampled to 30 m
USLE Analysis
29
3. MethodsGRID-BASED USLE ANALYSIS
C (cover management)=
Consolidated NLCD 30m
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
***no differentiation of erosion control practices
USLE Analysis
30
3. MethodsGRID-BASED USLE ANALYSIS
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
RKLSCP RKLSCP RKLSCP
P (support practice factor) =
1
***no differentiation of erosion control practices
USLE Analysis
31
3. Methods
Revised-USLE2 Analysis
32
3. Methods RUSLE-2
•Automated
•identifies potential sediment transport routes using raster grid cumulation and max downhill slope methods
•Identifies depositional zones
•L= surface overland flow distance from origin to deposition or stream
•CP (cover and practice) calculated from the RUSLE database (wider range of land cover characteristics)
33
3. Methods
SDR’s
34
3. Methods SDR’s
3 LUMPED-PARAMETER
2 SPATIALLY EXPLICIT
“life is a box, but spatial relationships matter”
“where life is a box and space is only considered in terms of area”
35
3. Methods
Runoff
36
3. Methods Runoff
CN (curve number) method used to estimate runoff potential and annual runoff
STATSGO hydrosoilgrp + LC = CN
Monthly time step annual value
(for multiple linear regression)
37
3. Methods
Multiple Linear Regression
38
3. Methods Multiple Linear Regression
Considered additional parameters
•Physiographic province
•Watershed size
•Variation in terrain complexity
•Topographic relief ratio
•Land cover proportions
•Percent impervious area
•Runoff potential
•Annual average runoff (CN method)
39
4. Results/Discussion
40
4. Results
USLE vs RUSLE2
41
USLE vs RUSLE 24. Results
42
USGS Data
Pearson r = 0.95, p<0.001
USLE vs RUSLE 24. Results
43
4. Results
USLE & RUSLE2 (SDR) vs SERC & USGS
44
USLE & RUSLE2 (SDR) vs SERC & USGS4. Results
45
Negative spearmen r =
USLE & RUSLE2 (SDR) vs SERC & USGS4. Results
46
All p values = not statistically significant
Negative spearmen r =
USLE & RUSLE2 (SDR) vs SERC & USGS4. Results
47
4. Results
USLE Parameters vs USLE
48
USLE parameters vs USLE4. Results
49
4. Results
Univariate Regressions
50
4. Results Univariate Regressions
51
4. Results Univariate Regressions
52
4. Results
Best Subsets Multiple Regression Analysis
53
4. Results Best Subsets Multiple Regression Analysis
54
4. Results Best Subsets Multiple Regression Analysis
55
4. Results Best Subsets Multiple Regression Analysis
(dead sheep)
56
4. Results
Other Multiple Linear Regressions
57
4. Results Other Multiple Linear Regressions
58
LC
LC
4. Results Other Multiple Linear Regressions
PC
PC
59
5. Conclusion
60
5. Conclusion
Widespread misuse of USLE and derivatives
61
5. Conclusion
Widespread misuse of USLE and derivatives
Multiple linear regression fails
fierceromance.blogspot.com
62
5. Conclusion
Widespread misuse of USLE and derivatives
Multiple linear regression fails
elevated sediment loads short term events
Static models do not represent dynamic interactions among parameters, which change on a small time step
questionable spatial data
fierceromance.blogspot.com
63
5. Conclusion
Widespread misuse of USLE and derivatives
Multiple linear regression fails
elevated sediment loads short term events
Static models do not represent dynamic interactions among parameters, which change on a small time step
questionable spatial data
“…trends collectively suggest scientists […] have not captured the linkages between the catchment landscape setting and the physical mechanisms that regulate erosion and sediment transport processes.”
fierceromance.blogspot.com
64
5. Conclusion
Models that rely on USLE for calibration
GWLF (Generalized Watershed Loading Function; Haith and Shoemaker, 1987)
AGNPS (AGricultural Non-Point Source; Young et al., 1989)
SWAT (Soil & Water Assessment Tool; Arnold and Allen, 1992)
HSPF (Hydrological Simulation Program-Fortran; Bicknell et al., 1993)
SEDD (Sediment Delivery Distributed model; Ferro and Porto, 2000).
USE WITH CAUTION (DON’T USE)
65
5. Conclusion
1. Identify predictor variables that conceptually link landscape and stream characteristics to flow velocity, stream power, and the ability to transport sediment
In order to accurately predict sediment discharges in ungauged drainage basins, scientists need to:
66
5. Conclusion
1. Identify predictor variables that conceptually link landscape and stream characteristics to flow velocity, stream power, and the ability to transport sediment
2. Incorporate metrics to indicate potential sediment sources within streams, including bank erosion and legacy sediments
In order to accurately predict sediment discharges in ungauged drainage basins, scientists need to:
67
5. Conclusion
1. Identify predictor variables that conceptually link landscape and stream characteristics to flow velocity, stream power, and the ability to transport sediment
2. Incorporate metrics to indicate potential sediment sources within streams, including bank erosion and legacy sediments
3. Develop predictions for temporal scales finer than the long-term annual average time frame
In order to accurately predict sediment discharges in ungauged drainage basins, scientists need to:
68
5. Conclusion
1. Identify predictor variables that conceptually link landscape and stream characteristics to flow velocity, stream power, and the ability to transport sediment
2. Incorporate metrics to indicate potential sediment sources within streams, including bank erosion and legacy sediments
3. Develop predictions for temporal scales finer than the long-term annual average time frame
In order to accurately predict sediment discharges in ungauged drainage basins, scientists need to:
WE NEED CONSISTENT AND VERIFIABLE RESULTS!!!
69
6. My Comments
70
I would also emphasize not only a need for stronger scientific theories but finer resolution spatial data. The closer raster based data (and any other spatial data for that matter) becomes to the actual landscape, the greater chance there is of describing the processes that control sediment yields (in additional to other “contaminants”) at the catchment scale. The stronger the GIS database, the greater potential for success (however, it remains to bee seen exactly what level of spatial and temporal detail is needed to optimize results and minimize costs).
6. My Comments
71
6. My Comments
Weakest Points
Violates a fundamental principle of spatial analysis
No spatially explicit terms in multiple linear regression
Long term based model applied to tiny temporal window(observed data have low probability of representing “average conditions”)
72
6. My Comments
Weakest Points
Violates a fundamental principle of spatial analysis
No spatially explicit terms in multiple linear regression
Long term based model applied to tiny temporal window(observed data have low probability of representing “average conditions”)
73
6. My Comments
Weakest Points
Violates a fundamental principle of spatial analysis
No spatially explicit terms in multiple linear regression
Long term based model applied to tiny temporal window(observed data have low probability of representing “average conditions”)
74
6. My Comments
Weakest Points
Violates a fundamental principle of spatial analysis
No spatially explicit terms in multiple linear regression
Long term based model applied to tiny temporal window(observed data have low probability of representing “average conditions”)