TM-Based SOC Models Augmented by
Auxiliary Data for Carbon Crediting Programs in
Semi-Arid Environments
Salahuddin M. Jaber and Mohammed I. Al-Qinna
Abstract
This study aimed at testing the hypothesis that augmenting
Landsat TM-based models for predicting soil organic carbon
(
SOC
) with auxiliary data about variables that might affect the
spatial distribution of
SOC
might improve the predictability of
these models in the Zarqa Basin in Jordan (a typical semi-arid
watershed) and enable them to be used for implementing carbon
crediting programs in semi-arid environments. Six modeling
procedures, namely stepwise regression, partial least squares,
recursive partitioning analysis, screening regression analysis,
artificial neural networks, and combined models, were calibrat-
ed and validated for the basin and for the land cover types that
exist in the basin. Although none of the developed models was
powerful for predicting SOC, artificial neural networks models
were more applicable specifically in agricultural lands. How-
ever, the margins of error associated with the best models were
high, and hence hindered the applicability of these models in
carbon crediting programs in semi-arid environments.
Introduction
Global atmospheric concentration of carbon dioxide (the main
greenhouse gas) has increased dramatically since the indus-
trial revolution as a result of human activities including fossil
fuel combustion, land use change, and agricultural practices.
This increase, according to the Fourth Assessment Report
(
FAR
) of the Intergovernmental Panel on Climate Change
(
IPCC
), has lead with very high confidence to global warming
trend of the atmosphere (
IPCC
, 2007a). Given the observed
and projected future tremendous and variable environmental,
societal, and economic impacts of this threat (
IPCC
, 2007b), a
wide variety of policy instruments have been developed to
create incentives for governments to adopt global warming
mitigation actions (
IPCC
, 2007c). Among these instruments
are adopting land management practices that encourage soil
organic carbon (
SOC
) sequestration, which implies transferring
carbon (in the form of carbon dioxide) from the atmosphere
into the soil through the natural processes of humification
(Lal, 2008). Soil organic carbon is considered the lifeblood
of the soil. Its sequestration, in addition to ameliorating the
effects of global warming, has a multitude of benefits. It drives
most of the soil processes, which in turn affect many of the
ecosystem functions and structure and ultimately services to
humans (Moseley and Richards, 2007).
However, in order to gain international acceptance that
SOC
sequestration can be applied as a global warming abatement
measure, there is a need to develop models that can be used
indirectly to estimate
SOC
stocks at different spatial scales and
under natural surface conditions with an acceptable error.
Among these models are the ones that are built upon satellite-
based multispectral remote sensing systems. They provide
cost-effective, non-destructive, and spatially continuous
information that cover large areas rapidly on a repetitive basis,
which is suitable for implementing any governmental programs
granting carbon credits. Nevertheless, a few studies tried to
build models that can be used to predict surface
SOC
based on
information obtained from multispectral remote sensing sen-
sors onboard satellites such as
SPOT1
(e.g., Agbu
et al
., 1990),
Landsat-5 (e.g., Wilcox
et al
., 1994; Jaber and Al-Qinna, 2011;
Jaber and Al-Qinna, 2015), IRS LISS-III (e.g., Ladoni
et al
.,
2010), and MODIS (e.g., Winowiecki
et al
., 2016). These studies
resulted in developing models with varying degrees of success.
Other studies tried to enhance the accuracy of satellite-based
multispectral remote sensing systems
SOC
models by augment-
ing them with auxiliary data such as climate, topography, land
cover, and soil characteristics (e.g., Poggio and Gimona, 2014;
Peng
et al
., 2015; Were
et al
., 2015; Yang
et al
., 2015). However,
none of these studies, except the one which was conducted by
Jaber and Al-Qinna (2011), evaluated the applicability of the
developed models to be used for implementing and monitoring
carbon crediting programs. Jaber and Al-Qinna (2011) evalu-
ated the applicability of using Landsat-5
TM
data for estimat-
ing
SOC
in a typical semi-arid watershed in Jordan (the Zarqa
Basin) by testing a variety of data-mining techniques for the
purpose of implementing and monitoring carbon crediting
programs in semi-arid environments. However, the results were
not promising and none of the developed models was power-
ful for predicting
SOC
. The lowest validation root mean square
error (
RMSE
) obtained was about 10.4 metric tons per hectare
(ton/ha). This margin of error, if compared to the mean decadal
scale carbon sequestration rates due to changes in farming
practices as reported by West and Post (2002) is too high for
direct application in carbon crediting programs. The present
study builds upon the previous one, which was conducted by
Jaber and Al-Qinna (2011). It aims at testing the hypothesis
that augmenting
TM
multispectral data with relevant auxiliary
data about variables that might affect the spatial distribution of
SOC
and breaking down the analysis by land cover type might
improve the predictability of these models in the Zarqa Basin.
This might enable them to be used directly for implement-
ing and monitoring carbon crediting programs in semi-arid
environments and under natural surface conditions. Landsat-5
TM
archives comprise free-of-charge, continuous, worldwide,
and historical data unmatched in quality, detail, coverage, and
length, which span for about 30 years (
NASA
, 2016). Further-
more, the analyses were carried out in a geographic informa-
tion systems (
GIS
) environment, which provide an outstanding
Department of Water Management and Environment, Faculty
of Natural Resources and Environment, Hashemite University,
Zarqa, Jordan (
.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 6, June 2017, pp. 447–457.
0099-1112/17/447–457
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.6.447
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2017
447
