Building capacities for digital soil organic carbon mapping
Building capacities for digital
soil organic carbon mapping
M. Guevara, G. F. Olmedo E. Stell, J. G. De Lira, C.
Cruz, R. Vargas
The University of Delaware, Plant and Soil Sciences
La Comision Nacional para el Conocimiento y Uso
de la Biodiversidad, CONABIO
Instituto Nacional de Tecnología Agropecuaria, Ar.
Instituto Nacional de Estadistica y Geografia Mx.
Many thanks to:
The SCORPAN model
Digital terrain analysis
Harmonization of databases
Validating and predicting
Uncertainty of SOC estimates
Planetary boundaries that
must not be transgressed
are closely related to soil
organic carbon (SOC)
How can we quantify the
spatial variability of SOC
and predict in places where
no information is available?
We use digital soil mapping to ananlyze the spatial variability of SOC
and its associated uncertainty aiming to provide a baseline estimate
useful for accounting and monitoring soil carbon.
Digital Soil Mapping (DSM) is the creation and the population of a
geographically referenced soil database, generated at a given
resolution by using field and laboratory observation methods coupled
with environmental data through quantitative relationships (see
We assume that different DSM models will capture different portions of
SOC variability. We combine linear models with data driven statistics
for variable selection, prediction to new data and uncertainty
Land modeling efforts to predict climate change rely on SOC
estimates but no uncertainty assessment is available.
High uncertainties and significant differences of SOC estimates
associated with the number of covariates, the pixel size and the
statistical performance of the modeling approaches.
We coupled remote sensing, geomorphometry and climate
surfaces to generate a multiscale and multisource set of prediction
factors. We model, cross validate and estimate uncertainty of SOC
To provide cyber-infrastructure tools and develop country-specific, technical,
analytical and institutional capacities for digital SOC mapping across Central
America and the Caribbean .
To better inform and empower producers and protect the functioning of
the productive landscapes of our territories
Uncertainty in soil data can outweigh climate impact signals in global crop
yield simulations Folberth et al. (2017)
Country-specific information is required by regional-to-global academic and policy
relevant research efforts (GlobalSoilMap.net, FAO-GSM cookbook 2017).
Working principles and specific objectives
Transparency in methods
Open source platforms
To teach the participants how to generate a high quality digital SOC maps based
on state of the art techniques including uncertainty of SOC estimates
The SOC resulting maps could be used as baseline estimates for reporting,
monitoring and verification of SOC stocks
To generate a regional, interpretable and predictable model of SOC variability
Country specific SOC map representative of the first 30cm of depth at the spatial
resolution of 1x1km, derived using digital soil mapping techniques and including
the uncertainty of SOC estimates.
Baseline datasets (SOC ground thruth)
SOC Point data (e.g. Mexico ):
Baseline datasets (SOC covariates 1x1km)
Digital terrain parameters
Legacy and polygon maps (i.e., soil, geology, land use/cover)
Pelletier et al. 2016, Shangguan et al., 2017, Hengl et al. 2017, Fick and Hijmans,
f, a data driven decision
‘I will keep my eye on the quantile culture to see what develops’. Breiman 2001
Quantile regression forests, Meinshausen 2006 The full conditional distribution, a surrogate of uncertainty
Kriging, Support Vector Machines, Cubist, Kernel weighted nearest neighbors
Examples of large scale SOC mapping
Viscarra Rossel et al 2014, Hengl et al. 2017 ...
Digital terrain analysis (DTA) in SAGA GIS
Topography is a major driver of soil variability
Topography can be represented by digital elevation models
We have digital elevation models globally available
Using DTA we can derive topographic attributes (e.g. Slope, aspect, wetness
Derive terrain parameters in SAGA GIS using a digital elevation model of 1x1 km.
Global-to-local and local-to-global SOC mapping
All models are wrong but some of they are useful
Models should not compete but inform each other
Different models (and modeling cultures) will capture different portions of SOC
There are no best method on digital soil mapping (no silver bullets)
Ensemble learning should be part of the pedometrics agenda
The eternal question is, how inform nationwide policy decision based on the best
Box 1976, Breiman 2001, Ho and Pepyne 2002,, Finke 2012, Qiao et al. 2015
Harmonization of datasets
To generate a geographical information system of all available data sets.
For points and grids, same geospatial reference
For grids, same pixel size, 1km. (note: Rasterize first polygon maps)
Exercise 2.1 Harmonize SOC predictors in a regular grid of 1x1km and all
Exercise 2.2 Generate a regression matrix of available SOC data and the values
of the predictors at observation points.
Information sources (some)
1. List of soil covariates available to 1km from ISRIC’s worldgrids.org
2. List of additional covariates
3. Limit of each country http://www.gadm.org/country
4. References (McBratney et al. Batjes et al, Hengl et al.)
5. Code (considering FAO DSM cookbook, http://www.fao.org/3/a-bs901e.pdf)
6. Regression matrix
7. Covariates matrix
8. Digital elevation model processing steps (derive terrain parameters in SAGA