1. Sabine Grunwald; James O Sickman; Nicholas B Comerford
Assessment of Dynamic Soil Carbon Pools at the Watershed Scale Using Regression Kriging
Gustavo M Vasques;
Sabine Grunwald; James O Sickman; Nicholas B Comerford
Funded by: Cooperative Ecosystem Studies Unit – Natural Resources Conservation Service – USDA

2. Rationale
Florida has the highest amount of soil organic carbon per area compared to other States in the U.S. (Guo et al., 2006)
Florida experienced the second highest increase in CO2 emissions from compared to other States (Cassady, 2007)
Guo et al Quantity and Spatial Variability of Soil Carbon in the Conterminous United States. Soil Sci. Soc. Am. J. 70:590–600
Cassady The Carbon Boom: State and National Trends in Carbon Dioxide Emissions Since Environment Florida Research & Policy Center, Tallahassee, FL
Florida has highest amount of soil organic carbon per area in the U.S. and the second highest total amount when considering the state area
Soil is a potential reservoir to sequester atmospheric CO2 and mitigate global warming
There is need for rapid and accurate techniques to synthesize soil information to improve land management
Sustainable land management requires understanding the dynamics of soil carbon and how it relates to landscape factors 2

3. Objectives
To model the distribution of total soil carbon (TC) across the Santa Fe River Watershed (SFRW) at four depths down to 180 cm
To model the distribution of surface soil carbon (SC) fractions in the SFRW
1) Soil is a potential reservoir to sequester atmospheric CO2 and mitigate global warming
2) Study of soil carbon fractions helps to elucidate carbon dynamics on the landscape; Also, they indicate to what scale SC can be sequestered in the soil (RC) and to what scale it can be lost (HC, DOC, and MC)
3) There is need for rapid and accurate techniques to synthesize soil information to improve land management
4) To understand how changes in environmental properties (esp. land use shifts) could affect changes in SC content
To compare different methods to upscale TC and SC fractions
To identify environmental variables that exert major control on SC

4. Sand-rich marine sediments - 90% Karst terrain - 10%
Subtropical climate:
Precipitation – 1,334 mm/yr
Temperature – 20.4 oC
Diverse land uses:
Pinelands - 30%
Wetlands - 14%
Upland forests - 13%
Pasture - 13%
Rangelands - 13%
Urban areas - 11%
Crops - 5%
Influenced by wildfires and hurricanes
High precipitation rates foster vertical leaching of OM in the soil profile
Elevation: m AMSL
Geology:
Sand-rich marine sediments - 90%
Karst terrain - 10% 4

5. Stratified random design along soil-land use trajectories
141 sites and 4 depths
0-30 cm
TC - Total carbon (combustion)
HC - Hydrolysable carbon (6N HCl)
RC - Recalcitrant carbon (TC - HC)
DOC - Dissolved organic carbon
(using hot water extraction)
MC - Mineralizable carbon
(from 15th until 29th day of incubation)
30-60 cm TC
Stratified random sampling along soil-land use trajectories to capture the variability of SC across soil type and land use
>99% of TC is organic carbon cm TC cm TC 5

6. (Environmental Correlation)
Upscaling Methods
All SC properties were log-transformed
Samples were split at each depth into training (2/3) and validation (1/3)
Methods
Trend Model
(Environmental Correlation)
Residuals
(1) LK
Lognormal Kriging
(2)
RK/SMLR
Stepwise Multiple Linear Regression
(3)
RK/RT
Regression Tree
Explain more thoroughly that RK was done in two different modes: RT and SMLR
59 environmental landscape variables were use in the global trend models
RK = Regression Kriging

7. Results – Descriptive Statistics
Property (Depth in cm)
Number of Obs.
Mean (kg m-2)
Median (kg m-2)
Std. Dev. (kg m-2)
Range (kg m-2)
HC (0-30)
141
1.59
1.28
1.33
RC (0-30)
4.67
3.30
6.91
DOC (0-30)
0.34
0.29
0.27
MC (0-30)
0.05
0.04
TC (0-30)
6.26
4.57
8.04
TC (30-60)
3.73
1.67
12.02
TC (60-120)
139
3.59
1.71
11.54
TC ( )
133
1.61
1.08
2.05
0.16 – 18.5
TC (0-100)
11.79
7.49
25.08
RC represented ~75% of the soil TC content, while HC represented ~25%
DOC represented ~5% of TC
MC represented <1% of TC, and ~15% of DOC 7

8. Results - Upscaling
Property (Depth in cm)
RMSE of Training
(kg m-2)
RMSE of Validation
Total Stock
(Mt) LK
RK/
SMLR RT
HC (0-30)
1.14
1.94
1.59
0.73
2.49
1.73
6.21
9.88
9.81
RC (0-30)
4.74
8.32
6.30
2.85
6.36
2.57
17.3
20.21
21.77
DOC (0-30)
0.10
1.43
1.16
0.14
10.88
1.19
1.30
45.72
5.34
MC (0-30)
0.02
1.70
1.31
11.43
1.22
0.18
47.98
4.49
TC (0-30)
5.16
9.65
11.23
3.34
6.96
9.99
23.0
25.36
27.74
TC (30-60)
2.41
11.72
10.96
12.6
20.27
6.20
10.5
65.30
11.65
TC (60-120)
8.28
27.30
11.98
2.02
19.32
3.42
10.6
37.29
20.57
TC ( )
0.44
2.94
2.75
3.46
5.28
8.98
12.38
TC (0-100)
12.7
29.39
26.35
7.21
16.67
15.82
39.3
82.59
90.13
OK performed best in training mode for all properties because it is an exact interpolator, so estimated and observed values should match at the sampling sites
Reduced clay content (average = 15%) can explain why correlations are low between SC and environmental properties
When OK performed better, this is an indicative that the spatial dependency structure is more important a determinant than the environmental factors
Best methods are in bold 8

9. SC Fractions - Variograms
logRC (0-30)
logDOC (0-30)
Semivariance
Distance (m)
logHC (0-30)
logMC (0-30)
Variograms of the log-transformed properties, not the residuals
Semivariograms show a similar range for the different SC fraction 9

10. SC Fractions - Best Methods
SC fractions were a function of/highly correlated with TC, so the distribution of SC fractions follows the distribution of TC
Correlations with environmental variables were not very strong
Selected variables:
RC(0-30) – Depth to water table
DOC (0-30) – Soil hydrologic group 10

11. Total Carbon - Variograms
logTC (30-60)
logTC ( )
Distance (m)
Semivariance
logTC (0-30)
logTC (60-120)
Semivariograms show a similar range for TC at different depths 11

12. Total Carbon - Best Methods
Florida soils are atypical:
They do not follow the typical decreasing pattern of SC along the profile because of spodosols
The widespread presence of wetlands promotes the accumulation of SC due to slow decomposition
Selected variables:
TC(30-60) – Elevation (mouth of the river), Landsat 7 Band 2
TC( ) – Spodosols, Entisols, Slope, Clay percent, Agriculture, Env Geology = Medium fine sand and silt 12

13. Total Carbon - Best Methods
TC (0-30) is controlled mainly by the land use, especially by the presence of wetlands
TC (30-60) is high close to the mouth of the river, and low in upland forests and agricultural fields
TC (60-120) and TC ( ) are influenced by the presence of Spodosols 13

14. Total Carbon – TC (0-100) High values – close to wetlands or Spodosols
Medium values – areas of pine plantation, agriculture, and improved pasture
Low values (SE) – due to the presence of Entisols in areas of high infiltration rate and low productivity; usually areas of upland forest
Low values (NW) – Ultisols 14

15. Conclusions TC and SC fractions behaved similarly across the SFRW
SC in the SFRW was highest in the surface, in saturated horizons, and in deep spodic horizons
This is also true for similar landscapes in the SE US
TC and SC fractions behaved similarly across the SFRW
Overall, major controlling factors of SC storage were soil type, land use and hydrologic properties 15

16. Outlook
Land management to enhance SC in Florida should focus on two strategies:
Protect/improve SC-rich areas (e.g. areas of Spodosols, wetlands)
Identify strategies to conserve SC
Identify strategies to conserve SC
Land restoration
Control of urban expansionA
Adoption of more sustainable management practices in productive areas
Identify strategies to improve SC storage in productive areas (agriculture, forestry, cattle ranching)
No-till
Improved pasture
Land management should aim at promoting soil aggregation
Enhance SC storage in areas with low SC (e.g. areas of excessive drainage)
Identify strategies to improve SC storage 16

17. Thank you
Questions?