Combining Vis–NIR hyperspectral imagery and legacy measured soil profiles to map subsurface soil properties in a Mediterranean area (Cap-Bon, Tunisia)

doi:10.1016/j.geoderma.2013.06.005

Geoderma

Volumes 209–210, November 2013, Pages 168-176

https://doi.org/10.1016/j.geoderma.2013.06.005 Get rights and content

Highlights

•
Vis-NIR hyperspectral imagery (HI) can only predict surface soil properties.
•
SSF are functions for predicting subsurface soil properties from surface properties.
•
SSF were calibrated from a legacy database of soil profiles in Cap Bon (Tunisia).
•
Combined HI and SSF mapped from 1/3 to 2/3 of the subsurface soil property variances.
•
Vis-NIR hyperspectral imagery is useful for predicting sub-surface soil properties.

Abstract

Previous studies have demonstrated that Visible Near InfraRed (Vis–NIR) hyperspectral imagery is a cost-efficient way to map soil properties at fine resolutions (~ 5 m) over large areas. However, such mapping is only feasible for the soil surface because the effective penetration depths of optical sensors do not exceed several millimeters. This study aims to determine how Vis–NIR hyperspectral imagery can serve to map the subsurface properties at four depth intervals (15–30 cm, 30–60 cm, 60–100 cm and 30–100 cm) when used with legacy soil profiles and images of parameters derived from digital elevation model (DEM). Two types of surface–subsurface functions, namely linear models and random forests, that estimate subsurface property values from surface values and landscape covariates were first calibrated over the set of legacy measured profiles. These functions were then applied to map the soil properties using the hyperspectral-derived digital surface soil property maps and the images of landscape covariates as input. Error propagation was addressed using a Monte Carlo approach to estimate the mapping uncertainties.

The study was conducted in a pedologically contrasted 300 km²-cultivated area located in the Cap Bon region (Northern Tunisia) and tested on three soil surface properties (clay and sand contents and cation exchange capacity). The main results were as follows: i) fairly satisfactory (cross-validation R² between 0.55 and 0.81) surface–subsurface functions were obtained for predicting the soil properties at 15–30 cm and 30–60 cm, whereas predictions at 60–100 cm were less accurate (R² between 0.38 and 0.43); ii) linear models outperformed random-forest models in developing surface–subsurface functions; iii) due to the error propagations, the final predicted maps of the subsurface soil properties captured from 1/3 to 2/3 of the total variance with a significantly decreasing performance with depth; and iv) these maps brought significant improvements over the existing soil maps of the region and showed soil patterns that largely agreed with the local pedological knowledge. This paper demonstrates the added value of combining modern remote sensing techniques with old legacy soil databases.

Introduction

Implementing sustainable agricultural, hydrological and environmental management requires an improved understanding of the soil at increasingly finer scales. The current soil databases that exist are neither exhaustive nor precise enough to be efficiently used for this purpose. An alternative is digital soil mapping (DSM), which can be defined as “the creation and population of spatial soil information systems by numerical models inferring the spatial and temporal variations of soil types and soil properties from soil observation and knowledge and from related environmental variables” (Lagacherie and McBratney, 2006).

Remote sensing images are major sources of input data for digital soil mapping. Until now, these images have mainly been used as spatial inputs for representing the landscape variables that are related to the soils, such as vegetation and parent material (the soil covariates) (McBratney et al., 2003). Boettinger et al. (2008) reviewed the main indicators that could be retrieved from a multispectral image for estimating these soil covariates. Airborne gamma-radiometry was also demonstrated as a suitable source of data for mapping soil parent materials and their alteration rates (Wilford, 2012). After a spatial overlay with the sparse sets of observed and measured sites collected in a given area, the indicators derived from remote sensing have been used as independent variables in regression-like models or as external drift in geostatistic models (McBratney et al., 2003).

Alternatively, remote sensing can be considered a cost-efficient way to acquire dense spatial sets of soil property measurements and a means to overcome the lack of soil data that still severely limits the digital soil mapping performances (Lagacherie, 2008). A recent review (Mulder et al., 2011) reported successful attempts to directly map some key soil properties, including soil texture, soil organic carbon, iron content, carbonate content, and soil salinity, from multispectral, hyperspectral and radar images. Some initial promising results have been obtained in describing the complex spatial patterns of soil properties using airborne hyperspectral imagery (Gomez et al., 2012a, Gomez et al., 2012b, Schwanghart and Jarmer, 2011). However, such mapping is only feasible for soil surface properties because the effective penetration depths of optical and radar sensors do not exceed several millimeters (Liang, 1997) and several centimeters (Owe and Van de Griend, 1998), respectively.

Despite this limitation, the aforementioned remote sensing measurements of soil surface property variations can be valuable sources of information for predicting the variations of sub-surface properties. Eighty years of soil surveying have shown significant relations between surface and subsurface properties because i) most of the soils are formed from a single parent material that impacts all of the soil horizons and ii) the soil forming processes that drive the changes in soil properties with depth are themselves impacted by parent materials. Therefore, it should be possible to obtain valuable predictions of sub-surface soil properties from surface measurements when accounting for other landscape drivers that may also influence the pedogenic processes that result in soil property variations with depth, e.g., relief, climate, land use, and water table regime. The soil legacy data available in a growing number of soil databases around the world (Rossiter, 2004) are repositories of knowledge that could be useful in calibrating such predictions, following the approach of Gray et al. (2011) for depicting global soil–landscape relations.

This paper presents an approach for extending the successful mapping of three soil surface properties (clay content, sand content and CEC) to greater depths using hyperspectral imagery in the Cap-Bon Region, Northern Tunisia (Gomez et al., 2012a). Statistical functions inferring the sub-surface soil properties from surface property measurements and landscape variables were first calibrated from a local database of legacy measured and geo-referenced soil profiles. These functions were then coupled with hyperspectral imagery outputs to derive the maps of the subsurface properties with estimations of their associated uncertainties.

Section snippets

Surface–subsurface predictive functions

The value of a soil property S at the ith interval of depth (S_i) can be expressed as follows: $S_{i} = S_{1} + Δ_{i}$ $with Δ_{i} = S_{i} - S_{1} = f (\{S_{1}, L\}) + ε_{i}$ where S₁ is the estimate of the soil surface property S that is assumed to be obtained independently, e.g., from remote sensing, and {L} is a set of easily available landscape variables. The surface–subsurface predictive function f is a statistical function that estimates the differences in the soil property values between the surface and the ith interval of depth (∆_i) with a

Study area

The study area is located in the Cap Bon region in northern Tunisia (36°24′N to 36°53′N; 10°20′E to 10°58′E), 60 km east of Tunis, Tunisia (Fig. 1). This 300-km² area is mainly rural (> 90%) and devoted to cereals in addition to legumes, olive trees, vineyards, and natural vegetation for grazing. It is a hilly area, with elevations ranging from 0 to 226 m. The main soil types are Regosols, Eutric Regosols preferentially associated with sandstone outcrops, Calcic Cambisols, and Vertisols

Soil property variations with depths

Fig. 3 shows the distributions of the three properties for the depth intervals of 0–15 cm, 15–30 cm, 30–60 cm and 60–100 cm for the whole dataset of 152 soil profiles. For all of the considered depths, the soil properties were characterized by a high spatial variation. We observed opposite distribution trends of clay and sand content as a consequence of the highly significant negative correlation between these two variables (Gomez et al., 2012a). The general trend was an increase of clay and a

Added value of hyperspectral images

The hyperspectral-based DSM approach presented in this paper can be first compared with the classical DSM approaches that use currently available landscape covariates derived from digital elevation models and multispectral remote sensing images (see Grunwald (2009) for a review). However, the published papers specifically considering subsurface soil properties are in the minority in the DSM literature, and the papers considering both the soil properties targeted in this study and Mediterranean

Conclusions

In this paper, we presented an approach that enabled the estimation of subsurface soil properties from surface properties by merging data from legacy measured profiles, hyperspectral imagery and landscape covariates. We investigated surface–subsurface functions, including linear models, Random Forests variables, and landscape covariates for mapping soil properties. The error propagation was addressed using a Monte Carlo approach to estimate the mapping uncertainties. The main outcomes of this

Acknowledgments

The authors are indebted to UMR LISAH (IRD, France) and to CNCT (Centre National de Cartographie et de Télédétection, Tunisia) for providing the AISA-Dual images for this study. This hyperspectral data acquisition was granted by IRD, INRA and the French National Research Agency (ANR) (ANR-O8-BLAN-C284-01). We are also indebted to Yves Blanca (IRD-UMR LISAH Montpellier) for the field work.

References (38)

T.F.A. Bishop et al.
Modelling soil attribute depth functions with equal-area quadratic smoothing splines
Geoderma
(1999)
R. Ciampalini et al.
Detecting, correcting and interpreting the biases of measured soil profile data: a case study in the Cap Bon Region (Tunisia)
Geoderma
(2013)
J. Gallichand et al.
Mapping clay content for subsurface drainage in the Nile Delta
Geoderma
(1993)
C. Gomez et al.
Regional predictions of eight common soil properties and their spatial structures from hyperspectral Vis–NIR data
Geoderma
(2012)
J.M. Gray et al.
Distribution patterns of World Reference Base soil groups relative to soil forming factors groups relative to soil forming factors
Geoderma
(2011)
R. Grimm et al.
Soil organic carbon concentrations and stocks on Barro Colorado Island — digital soil mapping using Random Forests analysis
Geoderma
(2008)
S. Grunwald
Multi-criteria characterization of recent digital soil mapping and modeling approaches
Geoderma
(2009)
P. Lagacherie et al.
Chapter 1 Spatial soil information systems and spatial soil inference systems: perspectives for digital soil mapping
Developments in Soil Science
(2006)
B.P. Malone et al.
Mapping continuous depth functions of soil carbon storage and available water capacity
Geoderma
(2009)
A.B. McBratney et al.
From pedotransfer functions to soil inference systems
Geoderma
(2002)

A.B. McBratney et al.

On digital soil mapping

Geoderma

(2003)

B. Minasny et al.

A conditioned Latin hypercube method for sampling in the presence of ancillary information

Computers & Geosciences

(2006)

V.L. Mulder et al.

The use of remote sensing in soil and terrain mapping — a review

Geoderma

(2011)

W. Schwanghart et al.

Linking spatial patterns of soil organic carbon to topography — a case study from south-eastern Spain

Geomorphology

(2011)

T. Stuffler et al.

The EnMAP hyperspectral imager — an advanced optical payload for future applications in Earth observation programmes

Acta Astronautica

(2007)

J. Triantafilis et al.

Mapping clay content variation using electromagnetic induction techniques

Computers and Electronics in Agriculture

(2005)

R.A. Viscarra Rossel et al.

Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties

Geoderma

(2006)

J. Wilford

A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis

Geoderma

(2012)

H. Abdu et al.

Geophysical imaging of watershed subsurface patterns and prediction of soil texture and water holding capacity

Water Resources Research

(2008)

Cited by (37)

Remote sensing of the Earth's soil color in space and time
2023, Remote Sensing of Environment
Abstract
Soil color is a key indicator of soil properties and conditions, exerting influence on both agronomic and environmental variables. Conventional methods for soil color determination have come under scrutiny due to their limited accuracy and reliability. In response to these concerns, we developed an innovative system that leverages 35 years of satellite imagery in conjunction with in-situ soil spectral measurements. This approach enables the creation of a global soil color map with a fine spatial resolution of 30 m x 30 m. The system initially identifies bare earth areas worldwide using reflectance bands acquired from Landsat 4 through Landsat 8 between 1985 and 2020. Soil color was quantified using the CIE-XYZ coordinates, utilizing 8005 soil spectral measurements within the visible range (380–780 nm) as ground truth data. We established transfer functions to convert Landsat reflectance bands to standardized XYZ color coordinates. These transfer functions were subsequently applied to images of bare surfaces, covering approximately 38.5% of the Earth's surface. We validated the resulting global soil color map using statistical indices derived from an independent set of ground-truth spectral data, demonstrating a high degree of agreement. By creating the world's first global soil color map, we have set a baseline for future spatial and temporal monitoring of soil conditions, thus enhancing our understanding and management of our planet's vital soil resources.
Digital mapping of the soil available water capacity: tool for the resilience of agricultural systems to climate change
2023, Science of the Total Environment
Soil available water capacity (AWC) is a key function for human survival and well-being. However, its direct measurement is laborious and spatial interpretation is complex. Digital soil mapping (DSM) techniques emerge as an alternative to spatial modeling of soil properties. DSM techniques commonly apply machine learning (ML) models, with a high level of complexity. In this context, we aimed to perform a digital mapping of soil AWC and interpret the results of the Random Forest (RF) algorithm and, in a case study, to show that digital AWC maps can support agricultural planning in response to the local effects of climate change. To do so, we divided this research into two approaches: In the first approach, we showed a DSM using 1857 sample points in a southeastern region of Brazil with laboratory-determined soil attributes, together with a pedotransfer function (PTF), remote sensing and DSM techniques. In the second approach, the constructed AWC digital soil map and weather station data were used to calculate climatological soil water balances for the periods between 1917–1946 and 1991–2020. The result showed the selection of covariates using Shapley values as a criterion contributed to the parsimony of the model, obtaining goodness-of-fit metrics of R² 0.72, RMSE 16.72 mm m⁻¹, CCC 0.83, and Bias of 0.53 over the validation set. The highest contributing covariates for soil AWC prediction were the Landsat multitemporal images with bare soil pixels, mean diurnal, and annual temperature range. Under the current climate conditions, soil available water content (AW) increased during the dry period (April to August). May had the highest increase in AW (∼17 mm m⁻¹) and decrease in September (∼14 mm m⁻¹). The used methodology provides support for AWC modeling at 30 m resolution, as well as insight into the adaptation of crop growth periods to the effects of climate change.
Field-scale spatial correlation between soil and Vis-NIR spectra in the Cerrado biome of Central Brazil
2022, Geoderma Regional
Citation Excerpt :
In this way, Vis-NIR spectroscopy has the potential to determine soil spatial variability and properties, with a high spatial resolution (Rizzo et al., 2016). Vis-NIR spectroscopy has also been used to map soils (Bazaglia Filho et al., 2013) and their properties, such as organic matter (Conforti et al., 2013), clay content (Piikki et al., 2013; Ramirez-Lopez et al., 2019), particle size distribution, and cation exchange capacity (Lagacherie et al., 2013; Ramirez-Lopez et al., 2019), as well as to associate several properties with latent variables from the multivariate approach (Webster and Burrough, 1974; Odlare et al., 2005; Rizzo et al., 2016; Chauhan et al., 2021; Sleep et al., 2022). Webster and Burrough (1974), Jang et al. (2021), and Chauhan et al. (2021) illustrated the discriminant analysis as one of the best methods to assist in the survey and discretization of soil classes in the field, including the transition limits between them.
Visible and near-infrared (Vis-NIR) spectroscopy is a tool to determine soil spatial variability and has been used to map soils and their properties. Considering that physical, chemical, mineralogical, and morphological soil properties can affect the intensity and the depth of the spectral reflectance band in the Vis-NIR region, the objectives of this work were to: (i) evaluate the potential of the reflectance inflection difference (RID) to discriminate soils; and (ii) verify potential spatial correlations of the RID with soil properties, compared with the full spectra, in order to build thematic maps at a field scale. In a farm of 375 ha, 78 soil samples from the 0.87–0.92-m depth were collected in a regular grid of 200 m, with a focus on the soil diagnostic horizon (B_w horizon). The sampled soils were a Latossolo Vermelho-Amarelo ácrico (Haplic Ferralsol) and a Latossolo Vermelho distrófico (Rhodic Ferralsol). Twenty-two physical, chemical, mineralogical, and morphological soil properties were determined, and the Vis-NIR spectra between 400 and 2500 nm were measured. Considering the presence of an inflection band and its relationship with soil properties, the spectral bands used to calculate the RID were between (base 1/base 2): 400–510, 730–930, 1290–1450, 1800–1950, 2000–2218, and 2218–2290 nm. The RID failed to map the spatial variability of soil properties, with a Kappa index of 39%; therefore, it is not a good parameter for building thematic maps of soil parameters. In addition, the complete spectrum (mainly in 400–510, 730–930, 1290–1450, 1800–1950, 2000–2218, and 2218–2290 nm) was better spatially correlated with soil properties than the decomposition of the spectrum by the RID. Soil classification and level of discretization as affected by spectral variability were also discussed here. Three soil groups were discriminated mainly by the Ki and Kr indexes and clay content. Moreover, the variability of the spectra was conditioned by the spatial variability of the mentioned variables. The clay content for soils with a discrepant particle size (group 1 compared with groups 2 and 3) and the Ki and Kr indexes for soils with a homogeneous particle size (between groups 2 and 3), associated with the full Vis-NIR spectral analysis, allowed building thematic maps with a good precision, without the need of mathematical models; this was possible by the modification of the reflectance intensity and the size of the concavity of the spectral band.
Land use mosaics in Mediterranean rainfed agricultural areas as an indicator of collective crop successions: Insights from a land use time series study conducted in Cap Bon, Tunisia
2021, Agricultural Systems
In cultivated landscapes, land use patterns related to the diversity of crops, their spatial arrangement into patches and their succession over several years influence many biophysical processes and the production of ecosystem services and disservices. Understanding the determinants of these patterns is a prerequisite for the development of acceptable alternative land use patterns. Most studies deem crop distribution patterns at the landscape level to be the result of individual allocations of crop successions to fields designed at the farm level. However, in some parts of the world, there are collective crop successions that apply to groups of adjacent fields on different farms.
The objective of this study was to examine the extent to which the spatiotemporal distribution of crops at the landscape scale relates to collective crop successions.
The study was based on a Mediterranean rainfed agricultural landscape (67.7 km²) located in northeastern Tunisia, in which collective successions respond to constraints related to agricultural land fragmentation. Combined with field mapping, remotely sensed land use time series were used to identify three-year crop sequences, classify them into crop succession types and identify clusters of adjacent fields with the same type of crop succession. We assumed that such a cluster was the result of a collective succession if the determinants of the individual crop succession locations did not explain its size (expressed in the number of fields). We related such determinants to the characteristics of the fields and their land-use environment and defined them statistically. Then, we developed a spatial permutation test to distinguish clusters resulting only from the determinants of the individual crop succession locations from those resulting from collective succession.
The results show that the collective successions mainly comprised biennial successions (wheat sown alternately with legumes, spices or forage crops). These successions were synchronized between adjacent fields based on wheat cultivation; all fields in the same cluster had wheat in the same year. Collective successions were secondarily comprised of fodder-dominant successions. These collective successions involved approximately 40% of the fields and their total area in the study area. These fields belonged to different clusters ranging in size from two to 96 adjacent fields.
The developed approach is a tool for mapping the likely presence of collective successions and considering this factor in the definition of sustainable land use scenarios at the landscape level for better soil and water management.
A case-based method of selecting covariates for digital soil mapping
2020, Journal of Integrative Agriculture
Selecting a proper set of covariates is one of the most important factors that influence the accuracy of digital soil mapping (DSM). The statistical or machine learning methods for selecting DSM covariates are not available for those situations with limited samples. To solve the problem, this paper proposed a case-based method which could formalize the covariate selection knowledge contained in practical DSM applications. The proposed method trained Random Forest (RF) classifiers with DSM cases extracted from the practical DSM applications and then used the trained classifiers to determine whether each one potential covariate should be used in a new DSM application. In this study, we took topographic covariates as examples of covariates and extracted 191 DSM cases from 56 peer-reviewed journal articles to evaluate the performance of the proposed case-based method by Leave-One-Out cross validation. Compared with a novices’ commonly-used way of selecting DSM covariates, the proposed case-based method improved more than 30% accuracy according to three quantitative evaluation indices (i.e., recall, precision, and F1-score). The proposed method could be also applied to selecting the proper set of covariates for other similar geographical modeling domains, such as landslide susceptibility mapping, and species distribution modeling.
Multi-temporal bare surface image associated with transfer functions to support soil classification and mapping in southeastern Brazil
2020, Geoderma
Citation Excerpt :
Another possible factor impacting surface-subsurface spectra correlations correspond to agricultural practices, where soil management could be increasing the system’s complexity (de Mendes et al., 2019). Furthermore, Lagacherie et al. (2013) describes differences in the performances of depth-transfer models, depending on soil formation and the processes involved. Predictions of soil attributes had satisfactory performance for surface and subsurface clay content (R2 = 0.62 and 0.63; RMSE = 82.87 and 88.24 g kg−1), iron concentration (R2 = 0.72; RMSE = 5.88 g kg−1) and soil color (R2 for hue, value and chroma were 0.57, 0.73 and 0.63, respectively; RMSE corresponded to 1.28, 0.26, 0.52) (Table 2).
Detailed soil maps are essential for agricultural management, but they are scarce in many regions. Even with the recent development of digital soil mapping (DSM) strategies, providing an adequate spatial representation of soils is still a challenging task. Therefore, this work aims to define a DSM approach, which combines proximal and remote sensing data to describe the spatial variation of soil attributes and types. The study was carried out in a site at southeastern Brazil, where 326 sampling points were defined and collected at two depths. Soil Vis-NIR spectra and physico-chemical attributes were measured in laboratory. A bare surface synthetic image (SYSI) was created from multi-temporal Landsat images and later validated with lab. spectra. Geographically weighted regression was used to calibrate depth transfer functions, which were applied to SYSI, generating a subsurface soil synthetic image (SYSI_sub). Soil attributes at both depths were mapped with SYSI, SYSI_sub and terrain derivatives. A soil classification key was designed following the Brazilian Soil Classification System and boolean logic. Soils were classified based on the soil attributes maps and boolean key. Hence, Monte-Carlo simulation (MCS) was performed to evaluate the error propagation from predicted attribute maps to soil types map. Correlations between satellite and lab. spectra varied from 0.68 to 0.8, indicating good capacity of SYSI in retrieving bare soil reflectance. Depth transfer functions also had good performance, with R² ranging from 0.62 to 0.72. The soil attribute maps with best performance were clay content (R² = 0.63), iron concentration (R² = 0.72) and soil color (hue R² = 0.57; value R² = 0.73; chroma R² = 0.63). Soil organic matter and chemical attributes were poorly predicted, with R² between 0.12 and 0.38. MCS indicated that uncertainties in attributes maps might result in confusion between Ferralsols and Acrisols, Regosols and Luvisols, as well as Luvisols and Acrisols. Comparison between digital and conventional maps of soil classes, presented satisfactory kappa (34.65%) and global accuracy (54.46%). The technique presents an improvement to DSM, as it integrates soil sensing and depth transfer functions into DSM.

View all citing articles on Scopus

View full text

Combining Vis–NIR hyperspectral imagery and legacy measured soil profiles to map subsurface soil properties in a Mediterranean area (Cap-Bon, Tunisia)

Highlights

Abstract

Introduction

Section snippets

Surface–subsurface predictive functions

Study area

Soil property variations with depths

Added value of hyperspectral images

Conclusions

Acknowledgments

Geoderma

Geoderma

Geoderma

Geoderma

Geoderma

Geoderma

Geoderma

Developments in Soil Science

Geoderma

Geoderma

Geoderma

Computers & Geosciences

Geoderma

Geomorphology

Acta Astronautica

Computers and Electronics in Agriculture

Geoderma

Geoderma

Geophysical imaging of watershed subsurface patterns and prediction of soil texture and water holding capacity

Water Resources Research