dsm in argentina: challenges to overcome - marcos angelini, soil institute, national institute of...
TRANSCRIPT
Mr. Marcos Angelini [email protected]
Soil Institute National Institute of Agriculture Technology
DSM in Argentina: challenges to overcome
“Towards a Global Soil Partnership for Food Security and Climate Change Mitigation and Adaptation”
Day 3: Tools for Soil Mapping
Olmedo, G.; Rodriguez, D.; Moretti, L.; Zurita, J.; Lopez, A; Schulz, G; Pezzola, A.; Cruzate, G.; Morales, C.; Tenti, L.
Content
•Context
•National Project
• Study cases
• Soil map point based
• Polygon maps disaggregation
DIGITAL SOIL INFORMATION Implementation & Validation New Methodologies for Up-dating the Soil Information
Objectives General
• Need of updated digital soil information that enables soil resource inventories and that identifies areas of opportunities & risks.
• To adapt and apply the appropriate methodology for each study area according to its requirements.
• To train researchers in the DSM methodology
• To generate soil inventories identifying areas of opportunities & risks
• To develop a Soil Data Base accessed via web.
Specific
35 researchers staff – 2009 to 2012
Study Areas
• Ing. Juárez: 500 km2. 80% of area is native vegetation. Slope < 0.5 %. Water Erosion. Soil Map 1:500 000.
• San Justo: 1600 km2. Crop lands. Water erosion and flood risk. Soil map scale 1:50 000.
• La Paz: Farm 50 Km2. Slopes> 5%. Partially deforest. Water Erosion. Soil Map 1:50 000 & 1:20 000
• Tupungato: 710 km2. Fruit and wine production area.
• Patagones: 1600 km2. Partially deforested. Severe wind erosion. Soil Map 1:250 000.
Multi-linear Regression
Properties to predict
Environmental Covariate
EC 0-30 cm ASPECT, CN_BL, LS, MRRTF, MRVBF, WTI
EC 30-60 cm SLOPE, ASPECT, CN_BL, MBI
pH 0-30 cm DEM, SLOPE, ASPECT, CN_BL, CI, MRRTF, WTI
pH 30-60 cm ASPECT, CN_BL, LS, MRRTF
Clay 0-30 cm SLOPE, ASPECT, Curv1KK, MRVBF, MBI
Silt 0-30 cm SLOPE, ASPECT, Curv1KK, MRVBF, MBI
Sand 0-30 cm SLOPE, ASPECT, Curv1KK, MRVBF, MBI
Clay 30-60 cm CI, MRVBF, MBI
Silt 30-60 cm CI, MRVBF, MBI
Sand 30-60 cm CI, MRVBF, MBI
Depth Prof. SLOPE, CN_BL, MRRTF, MRVBF
Covariates selected by stepway
Exclusion of outliers
70% of data for make models
30% of data for validation
RStudio
Modeling
Validation
Properties to predict
R2 adjusted SSE MPE RMSPE
EC 0-30 cm 0.27 128.06 0.81 3.04
EC 30-60 cm 0.19 469.82 1.03 2.09
pH 0-30 cm 0.27 14.12 -0.55 0.71
pH 30-60 cm 0.31 8.11 -0.24 0.46
Clay 0-30 cm 0.13 1263.39 3.66 5.21
Silt 0-30 cm 0.13 7948.79 9.24 13.08
Sand 0-30 cm 0.13 15567.76 -12.93 18.32
Clay 30-60 cm 0.13 728.43 0.96 3.64
Silt 30-60 cm 0.13 4582.95 2.39 9.12
Sand 30-60 cm 0.13 8974.72 -3.38 12.77
Depth Prof. 0.24 229074.57 46.09 64.90
(Sun et al., 2011; Hengl et al., 2004)
Results
(a) EC 0-30 cm (dS/m); (b) pH 0-30 cm; (c) Clay 0-30 cm en (%); (d) Silt 0-30 cm en (%); (e) Sand 0-30 cm (%); (f) Depth profile (cm.)
MLR prediction +
Kriging of residuals
Methodology
1. To get environmental covariates 2. Sampling soil map (Taxonomic Classes) 3. To make and apply decision tree model 4. To repeat loops of 2) and 3) 5. Cross-validation with soil profile points
Environmental covariates DEM Potassium_c ChNBL AAChN TWI Landsat Distance V_OFD
OFD Slope Environments NDVI01 NDVI09 Converg_I Stream_P LS
1/5
Sampling of taxonomic soil classes
Soil A: 60% Soil B: 40% A1 =
Soil B Soil A
Cartographic Unit Soil Map
A1
• Random sampling
2/5
Used Covariates: 100% DEM 100% Potassium_c 98% ChNBL 83% AAChN 69% TWI 41% Landsat 40% Distance 39% V_OFD 32% OFD 28% Slope 25% Environment 23% NDVI01 18% NDVI09 17% Converg_I 13% Stream_P 12% LS
Decision tree model
3/5
Conclusions Recently developed tools for soil mapping have been combined to obtain soil maps point
based, what has shown promising results. The maps not only offer information about the edaphic variables, but also an estimation of point value and error. Other authors (Hengl et al, 2004; 2007a) have found some advantages using RK instead of ordinary kriging. Although this research do not analyze differences between the two methods, it can be seen that the maps obtain are of a higher quality.
When working on disaggregation polygons maps, the applied methodology allows the
use of information about the different soil classes that form the cartographic units. In this manner it can be obtained raster maps of soil classification associated to the probability of event occurrence.
It is suggested to continue applying these tools in order to find a solution to real
problems and state the advantages compared to other methodologies.