Soil amp Water Science Department University of Florida GIS Research Lab Sabine Grunwald Project Goals Modeling of soil carbon along pedo climatic trajectories across diverse ecosystems ID: 745722
Download Presentation The PPT/PDF document "Rapid Assessment and Trajectory Modeling..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Rapid Assessment and Trajectory Modeling of Soil Carbon Across a Southeastern Landscape
Soil & Water Science Department, University of Florida
GIS Research Lab
Sabine GrunwaldSlide2
Project Goals
: Modeling of soil carbon along
pedo
-climatic trajectories across diverse ecosystems
in Florida
Funding source: National Research Initiative Competitive Grant no. 2007-35107-18368 USDA NIFA - AFRI
Core Project of the
North American
Carbon Program
PD:
S. Grunwald
Co-PIs:
W.G. Harris, N.B.
Comerford
and G.L.
Bruland
Post-Docs:
D.B. Myers and D.
Sarkhot
Graduate students:
G.M.
Vasques
, X.
Xiong
and W.C. Ross
Field and lab staff:
A.
Stoppe
, L. Stanley, A.
Comerford
and S.
MoustafaSlide3
Rationale and Significance
Crutzen, 2002. Nature;
Steffen et al., 2005. Global Change and the Earth System;
Rockström et al., 2009. Nature;
Grunwald et al., 2011. Soil Sci. Soc. Am. J.
Global issues & priorities
Global estimates of terrestrial carbon stocks
UNEP-WCMC. http://www.carbon-biodiversity.net/GlobalScale/Map
Scharlemann et al. (2009): Harmonized World Soil Database (2009)-SOC values up to 1 m depth (1 km spatial resolution) & Ruesch and Gibbs (2008): Biomass carbon map using IPCC Tier 1 methodology and GLC 2000 land cover data.
Lack in understanding of soil
carbon (C) variability
Assessments rely on historic/
legacy soil C data
Soil C – a sink or source ?
Soil C – linkages to processes ?
Total soil C – C pools ?Slide4
Historic and current within ≤ 30m
Historic and current within ≤ 300m
Current (2008/2009)
Resampling of
453
historic sites (out of 1,288 historic pedons – FL Soil Database); 1965-1996
(Soil and Water Science Dept., UF & NRCS)
In 2008/2009 soil sampling at
1014
sites (0-20 cm) based on stratified-random sampling design (land use – soil suborder strata):
TC
SOC
IC
HC RC
BD
TN and TP
SOC Observations (FL)Slide5
N: 1,099
Data source: Florida Soil Characterization Database (FSCD)
Modeling of
Historic SOC
(1 m) – FL
Block Kriging
Block size: 250 x 250 m
Target: Ln-SOC kg m
-2
Nugget: 0.61
Sill: 0.86
Range: 101,088 m
ME: -0.0040 ln[kg m
-2
]
(~ 0.10 kg m
-2
)
Class Pedo-transfer function (PTF)
SOC =
f {LU, order}
SSURGO-Soil Data Mart (NRCS) 1:24,000
STATSGO2-Soil Data Mart (NRCS) 1:250,000
< 5
5 – 10
10 – 15
15 – 20
20 – 50
> 50
Not mapped
Vasques G.M. and S. Grunwald. 201_. Global Env. Change J. (in prep.)
Presented at the World Congress of Soil Sciences (2010)Slide6
SOC statistic
(depth to 1 m)
SSURGO
STATSGO2
FSCD obser-vations
FSCD block kriging
FSCD PTF
Map unit
655,155 map units
2,823
map units
1,099 points
2,282,843
250-m cells
7 soil orders
Minimum (kg m
-2
)
0.67
4.01
0.13
2.82
7.70
Maximum (kg m
-2
)
291.77
264.32
207.98
116.19
144.17
Median (kg m
-2
)
7.90
27.05
6.32
9.00
14.75
Mean (kg m
-2
)
24.17
58.44
12.85
13.95
32.84
Std. dev. (kg m
-2
)
39.3162.6723.6912.2845.63Total mapped area (km2)128,788142,681N/A142,678142,626Total stock (Pg)3.5186.820N/A1.9904.112Mean stock (kg m-2)27.3247.80N/A13.9528.83
Map unit
Florida
Estimates of SOC stocks to 1 m in Florida based on different data/methods was 4.110 ± 1.01 Pg (mean ± std. error)
Vasques G.M. and S. Grunwald. 201_. Global Env. Change J. (in prep.)Slide7
Grunwald S., J.A. Thompson & J.L.Boettinger. 2011. SSSAJ. In press
.
Predicts the spatially-explicit evolution and behavior of Soil Pixels / Voxels
Explicitly incorporates anthropogenic forcings
Incorporates bio-, topo-, litho-, pedo- and hydrosphere
Provides temporal context to account for ecosystem processes and forcings
Fuses empirical and process-based knowledge
Conceptual Modeling Framework: STEP-AWBH (
“
STEP-UP
”
)
Soil pixel (SA):Slide8
STEP variables:
Soil
Topographic
Ecological / geographic
Parent material
AWBH variables:
Atmosphere / climate Water Biota: LU/LC H(uman)
+
Spatially & temporally
explicit environmental matrix (FL): ~2 TB of data
N: 200+ variables
…..
Soil observations
+
PLSR
CART
Ensemble
regression
trees
… and others
Model
development:
Predict soil-
environmental
properties:
TC
SOC
SOC seq.
Carbon pools
TN, TP
… and more
Model validation:
Uncertainty assessmentSlide9
Data source: NRCS-USDA,
Soil Geographic Database / Soil Data Mart.
Soil Taxonomic Classes – FL
Histosol
Time period: 2000
–
2005;
data source: MODIS satellite data
Net Primary Productivity – FL
SpodosolSlide10
January
February
March
Data source: PRISM
35 – 55
33 – 75
75 – 55
55 – 75
75 – 95
95 – 115
115 – 135
135 – 155
155 – 175
175 – 195
195 – 215
215 – 235
Avg. Monthly
Precipitation
(mm) [1971-2000]
April
May
June
July
August
September
October
November
December
Climatic Data – FL Slide11
Time frame: 1971 – 2000
Data source: PRISM
Climatic Data – FL Slide12
1990
1995
2003
Data sources: Land use / land cover
1970: USGS; 1990 and 1995: Water Management Districts & FL Department of Transportation
2003: Florida Fish and Wildlife Conservation Commission
1970
1970 to 2003
:
↑ Urbanization
(5.4% - 12.1% - 11.0%)
↓ Agriculture
(21.9% - 7.4% - 8.6%)
↓ ↑ Rangeland
(8.8% - 4.7% - 8.2%)
↓
↑
Forest
(29.9% - 23.2% - 26.2%)
↓ Wetland
(21.7% - 4.4% - 5.8%)
Land Use Change (1970 – 2003)
Based on Satellite Data
?Slide13
Inputs (predictor variables): STEP-AWBH environmental variables
Predict SOC stocks
Modeling of Current SOC (0-20 cm) – FL
Methods: Ensemble regression trees (RT) and other data mining methodsSlide14
Total N: 1,014; Randomized 70/30 calibration/validation split of dataset
R
2
RMSE
RPD
Regression trees (RT)
0.49
3.2
1.34
Bootstrapped RT
0.63
2.6
1.64
Boosted RT
0.61
2.7
1.59
Random Forest
0.64
2.6
1.66
Support Vector Machine
0.60
2.8
1.55
Modeling of Current (2009) SOC Stocks (0-20 cm) – FL
Validation results – STEP-AWBH Modeling (kg C m
-2
)
Myers D.B., S. Grunwald et al. 201_. Global Change Biology J. (in prep.)Slide15
Modeling of Current (2009) SOC Stocks
(kg m
-2
)
(0-20 cm) – FL
Predictor variables of importance
: Available water capacity 50 cm 1.0
Soil Great Group 0.85 Land cover / land use (NLCD) 0.83 Land cover / land use (FFWC, 2003) 0.74
Ecologic region 0.50
Soil Order 0.25
Soil Suborder 0.22
… and more
Method
: Random Forest
Independent validation (N: 304)
Myers D.B., S. Grunwald et al. 201_. Global Change Biology J. (in prep.)Slide16
Modeling of
Current (2009) SOC Stocks (20 cm) – FL
Myers D.B., S. Grunwald et al. 201_. Global Change Biology J. (in prep.)
SOC (kg m
-2
)
Spatial resolution: 30 mSlide17
SOC sequestration
(g C m
-2
yr
-1
)
SOC Sequestration in Florida (1965 – 2009)
Historic & current sites ≤ 30 m (N: 194)
Grunwald et al., 201_. Front Ecol. Env. J. (in prep.)
SOC sequestration (g C m
-2
yr
-1
)
Mean: 11.6; Median: 17.7
STDev: 93.3
Max: 511.3
Time frame of sequestration (yrs)
Mean: 30.3; Median: 29.6
STDev: 5.3
Max: 43.5Slide18
Predictor variables of importance
:Surficial geology 100
Land use 1995 75.4
Long-term max. temp. May 75.4Long-term max. temp. March 62.9
Long-term max. temp. April 35.9
Soil Great Group 27.3Land use 1970 25.9
MODIS EVI (day 137) 22.8MODIS EVI (day 169) 22.7Landsat Bd. 3 20.6
Forest canopy cover 17.5 …. and more
Modeling of SOC Sequestration Rates
(g C m
-2
yr
-1
) (0-20 cm) – FL
Methods
: Ensemble trees (bagging mode)
10% V-fold cross-validation
Grunwald et al., 201_. Front Ecol. Env. J. (in prep.)
STEP-AWBH
model evaluation
(
g C m
-2
yr-1):MSE = 85.93
MAD = 47.61Slide19
Significance of research:
Predict high-resolution soil C pixels across large landscapesReduce the uncertainty of soil C assessmentModel spatial variability of soil C (C pools and nutrients) along climate and land use trajectories
Model soil change in dependence of anthropogenic induced stressorsSlide20
Soil attributes
=
f
(VNIR)
Rapid and cost-effective sensing of Soil C and Pools using visible/near-infrared (VNIR) diffuse reflectance spectroscopy
Soil attributes
=
f
(VNIR; MIR)
Spectral soil C modelingSlide21
Authors
Spectra Type
Area
N
Properties
R
2
Cal.
R
2
Val.
Vasques et al. 2008. Geoderma
VNIR
SFRW
554
TC
0.98
0.86
Vasques et al. 2009. SSSAJ
(Ahn et al., 2009. Ecosystems)
VNIR
SFRW
102
TC
RC
SC
HC
MC
0.93
0.93
0.89
0.92
0.87
0.86
0.82
0.40
0.70
0.65
Vasques et al. 2010. JEQ
VNIR
FL (hist.)
7120
SOC
0.97
0.79
Myers et al. 2011. in prep.
VNIR
FL (2009)
1014
SOC (RC, HC)
0.930.89McDowell et al. 2011. in prep.VNIR & MIRHawaii306SOC0.93 (VNIR)0.97 (MIR)V-fold cross-validationSarkhot et al., 2011. GeodermaVNIRTX514TCHCSOCIC0.940.960.950.930.850.77
0.86
0.81
Research Results VNIR & MIR Slide22
Follow-up Research Project
(NRCS, Grunwald – UF & McBratney – U Sydney)
Rapid soil C assessment across the U.S.
Soil C ↔ Land use/land cover, ecoregion, climate, …
Soil C
↔ VNIR
Apply research methodology tested in FL to U.S.
FL Slide23
http://soils.ifas.ufl.edu/faculty/grunwald
sabgru@ufl.edu