Demattê et al. (2025). A global soil spectral grid based on space sensing

Francos and Ben-Dor, 2022

Viscarra Rossel

et al. (2016)

demonstrated relationships between various soil properties

and soil spectra via EMR interaction with soil in field and laboratory

settings, laying the foundation for soil spectral modeling at continental

and global scales. The next step was to integrate soil spectral libraries

(SSLs) in laboratory and field domains (

Francos and Ben-Dor (2022)

This made it possible to fuse the spectral data from the ground and

satellites, leveraging the cloud computing to predict soil attributes

(

Novais et al., 2025

). Modeling studies enabled the creation of a global-

scale soil reflectance grid at finer spatial resolutions (30 m) thereby

enhancing applications such as land-use planning and precision

agriculture.

Despite such progress in the soil spectral sensing domain, several

issues that constrain upscaling from laboratory to field and global scale

needed to be addressed. For instance,

proposed to identify a transfer function to scale sensing from laboratory

to field and then to RS. Having overcome some challenges, a temporal

imaging system was initiated, where a first attempt was performed by

e et al. (2016)

, on a municipal scale. Afterward,

Gallo et al.

(2018)

improved the system at a regional scale.

e et al. (2018)

implemented an automated system in the Google Earth Engine web

platform (

Gorelick et al., 2017

). The modeling algorithm is named the

Geospatial Soil Sensing System (GEOS3), which generates a product (the

image), named Synthetic Soil Image (SySi). Simultaneously, a great

automatized work was performed by

Rogge et al. (2018)

named Soil

Composite Mapping Processor (SCMaP) at a country scale. Finally,

Richer-de-Forges et al., 2023

e et al. (2020)

applied the GEOS3 to the global domain. This was

tested for the Middle East (

Poppiel et al., 2020

), Europe (

Safanelli et al.,

2020

) and Brazil (

Safanelli et al., 2021

).

Rosin et al. (2023)

went further

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

3

and upgraded GEOS3 to fill the gaps in areas without bare soil exposure,

reaching a continuous predicted bare soil image.

Wesemael et al. (2024)

performed with a similar approach, rather

with different techniques to a continental scale using cloud and

distributed computing, and following a two-pronged approach: one for

detecting bare soils and predicting soil organic carbon (SOC) from the

soil spectra and one for the permanent vegetated areas (grasslands and

woodlands). One key difference between the techniques is that the GEOS

system consistently combines laboratory spectra with satellite data,

while Wesemael

’

s approach relied on Sentinel-2 imagery, unlike GEOS,

which used Landsat. Moreover, Wesemael

’

s approach used satellite data

from March to October which includes months where rainfall takes place

in many parts of Europe. Additionally, Wesemael

’

s method employed

independent validation in specific regions across Europe, whereas GEOS

performed field validations, often correlating image patterns with

extensive soil mapping campaigns. Another distinction lies in the

thresholding techniques for bare soil detection. GEOS applies static

thresholds for the spectral indices (i.e., NDVI and NBR2), while the

SCMaP employed by Wesemael uses a histogram-based method to create

custom thresholds for each Sentinel-2 tile independently, enhancing

regional adaptability. These differences highlight the unique contribu-

tions and scopes of each system.

Numerous studies demonstrated RS-informed characterization of soil

attributes such as organic carbon (

), texture (

), color (

), micro aggregation (

Ben-Dor et al., 2023

), and mineralogy

(

Rosin et al., 2023

). Indeed, all this progress in soil spectral modeling

has been made possible due to the integration of RS and extensive soil

databases. The mosaics formed by RS images permit the fine-scale and

contiguous mapping of soils, thereby addressing the limitations of non-

contiguous approaches that rely on indirect proxies such as relief (

Levin

et al., 2004

). Furthermore, mosaics of RS serve as proxies for soil-

forming factors (

McBratney et al., 2018

). It is particularly useful in

characterizing the spatiotemporal variations of dynamic soil-forming

factors such as vegetation, biota, climate, and soil hydrology

(

Grunwald, 2021

). Moreover, a long-term 40-year time series reveals

trends and changes in the landscape, such as erosion, vegetation dy-

namics, and soil moisture variations, enabling the assessment of soil

degradation, carbon dynamics, and restoration impacts (

Mello et al.,

2023

). The potential digital soil mapping (DSM) usage remarks this as a

significant step forward in soil science. Despite these advances, chal-

lenges remain in algorithm development and region-specific soil mea-

surement needs.

Despite soil maps being available worldwide, most of them are rep-

resented by a coarse spatial resolution, non-contiguous, and restricted to

a specific area, forming a vast puzzle with missing pieces. As soils pre-

sent a huge arrangement with gradual changes, the current map tech-

niques overgeneralize soil nuances. Problems such as gaps, obsolescence

of data, and different production and validation techniques can hamper

comprehensive soil monitoring (

Safanelli et al., 2021

). Furthermore,

there are different approaches to soil mapping from classical surveys to

DSM by remote sensing techniques as well as more recent DSM incor-

porating bare soil reflectance (

). In addi-

tion, the global soil cartographic information available is limited to a

coarse-scale resolution, typically

>

1:1,000,000 (

Arrouays et al., 2020

;

Poggio et al., 2021

). Such scales also tend to disregard essential details,

such as fragile soils, salt pans, or rocky outcrops. However, accurate soil

information is essential for management zoning or precision agriculture

(

Greschuk et al., 2023

). A comprehensive dataset enables effective

management from a farm to a global scale.

The most detailed global-scale soil attributes mapping dataset

currently available is the SoilGrid, with a spatial resolution of 250 m

made with DSM techniques (

Poggio et al., 2021

). It is important to note

that the original data used to create the maps are point-based, while

DSM, which emerged in the early 2000s (

McBratney et al., 2003

;

Lagacherie et al., 2006

), mobilizes techniques to extrapolate the

observations for unsampled areas. In the DSM approach, it is assumed

that spatially continuous covariates represent soil-forming factors. As

McBratney et al. (2018)

demonstrated, this approach offers a relatively

efficient, cost-effective, and environmentally conscious alternative to

traditional soil mapping methods even over vegetated areas. Models

such as machine learning (ML) algorithms can be used in DSM, which

encompasses the spatial modeling of soil properties (

Ma et al., 2019

;

Padarian et al., 2019

Khaledian and Miller, 2020

;

Wadoux et al., 2020

;

Grunwald, 2021

). Optical RS time series have several challenges (i.e.,

clouds and field features such as straw, soil moisture, and rocks), which

interfere with the soil spectral signature as observed in the laboratory

(

Ben-Dor et al., 2009

).

The indicated temporal image techniques have made efforts to avoid

most of these issues, and more will come with new hyperspectral sat-

ellite systems that are starting to be applicable from Earth orbit such as

the current Earth Surface Mineral Dust Source Investigation (EMIT),

PRecursore IperSpettrale della Missione Applicativa (PRISMA), Envi-

ronmental Mapping and Analysis Program (EnMAP), Copernicus

Hyperspectral Imaging Mission for the Environment (CHIME), and

Surface Biology and Geology (SBG). These sensors have great spectral

resolution, yet limited temporal repetition, making it difficult to perform

mapping that relies on change detection. On the other hand, multi-

spectral sensors, such as Landsat with 40 years of temporal information

can depict sequential imagery of soils with and without tillage coverage.

Accordingly, despite Landsat

’

s low spectral and spatial resolutions

(30

–

60 m) outperforms other RS in terms of its temporal resolution.

Most literature shows that Landsat spectra are useful due to their strong

correlation with soil properties, especially clay content (

Fongaro et al.,

2018

) and carbon content (

Padarian et al., 2022

).

This work aimed to construct a global soil spectral grid (30 m spatial

resolution) to identify the spatial variability of the world

’

s soil patterns

in dependence on several spatially and temporally varying environ-

mental factors, using Landsat data. We consider the possible spatio-

temporal relationships to monitor kinetic processes on the Earth

’

s bare

surface with implications for agriculture and environmental science, soil

management, and climate change mitigation strategies (

Ortiz-Bobea

et al., 2021

). To achieve this goal, we combine data from extensive SSL,

wet laboratory soil analyses, and optical Earth observations.

2. Methodological procedures

Fig. 1

delineates the steps involved in constructing the global soil

spectral grid. The datasets ensured broad spatial coverage representing

publicly available soil data. A rigorous soil sample preprocessing

screening, including drying, crushing, sieving, and VNIR-SWIR spectral

measurements, was conducted to ensure laboratory consistency. The

GEOS3 is a modeling concept that models satellite data in a temporal

manner (

e et al., 2018

). Each pixel created is named Temporal

Soil Spectra (TESS), and the entire image is the product finally named

Synthetic Soil Image (SySi). The present case depicts a Landsat time

series providing a robust global representation of soil reflectance.

Clustering methods, including a two-stage self-organizing map process

(SOMP) and Partition Around Medoids (PAM) clustering, reduced the

dimensionality of the SySi data and classified it into spectral groups for

spatial interpretation. The Boruta algorithm determined the significance

of environmental covariates, enabling data-driven analyses of soil-

environment relationships. This approach generates a precise, scalable

soil spectral grid suitable for diverse environmental and agricultural

applications.

2.1. Soil samples and dataset

Data on soil attributes and laboratory spectra were obtained from

two global sources, the WoSIS (

Batjes et al., 2020

) and the GSSL dataset

(

) (

). The former provides data from about

96,000 georeferenced profiles (

Batjes et al., 2020

), while the latter GSSL

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

4

dataset provides 8004 topsoil samples (

e et al., 2022

). Only ob-

servations with geographic coordinate accuracy (

<

0.0001 degrees, or

about 10 m) were included in this case. We focused on soil attributes and

topsoil samples up to 20 cm depth, including clay, sand, silt, and SOC.

The number of samples for particle size fractions and SOC was 125,636

and 30,980, respectively.

For standard quality conditions of soil spectral data, the soil samples

from GSSL were air- or oven-dried, crushed, and sieved to a size

<

2 mm.

Sample holders were at least 6 cm in diameter and 1 cm deep, except

when measurements were made with a fiber optic. The spectrometer has

a spectral resolution of at least 10 nm in the VNIR-SWIR region (between

400 and 2500 nm). The illumination source was a halogen lamp, and a

spectralon

©

white reference optimized and calibrated the sensor.

Spectra were recorded as diffuse reflectance varying from 0 to 1 (more

details refer to

e et al., 2022

).

Fig. 1.

Procedural steps for global bare soil grid construction. WoSIS: World Soil Information Service; GSSL: Global Soil Sampling and Site Inventory; GEE: Google

Earth Engine; SySi: Synthetic Soil Image; GEOS3: Geospatial Soil Sensing System; SOMP: Self-Organizing Map Process; PAM: Partition Around Medoids.

Fig. 2.

Location of soil samples with data of clay, sand, silt, SOC, and VNIR-SWIR spectra measured in the laboratory belonging to the GSSL (red dots) and WoSIS

(black dots) datasets.

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

5

2.2. Bare surface composite image production

We analyzed Tier 1 Collection 1 surface reflectance data processed

by the Landsat Ecosystem Disturbance Adaptive Processing System

(LEDAPS) and Surface Reflectance Code (LaSRC) algorithms (

USGS,

2020a, 2020b

) from the Landsat 4 Thematic Mapper (TM), Landsat 5

TM, Landsat 7 Enhanced Thematic Mapper Plus (ETM

+

), and Landsat 8

Operational Land Imager (OLI) sensors. These datasets are accessible

through the Google Earth Engine (GEE) cloud platform (

Gorelick et al.,

2017

). Given the different spectral configurations of the OLI sensors,

such as spectral resolution, Landsat 8 imagery was transformed to ETM

+

following a harmonization process (

Roy et al., 2016

The GEOS3 was used to identify bare surfaces in each available

Landsat satellite image from 1984 to 2022. These images have a tem-

poral resolution of 16 days and a spatial resolution of 30 m. The bare

surfaces were labeled based on several indices: the greenness index

(GREEN), the normalized difference vegetation index (NDVI), the

normalized burn ratio index (NBR), and the normalized burn ratio index

2nd version (NBR2). The term ‘bare surface

’

was assigned when the

following conditions were met: GREEN

<

0.65,

−

0.05

<

NDVI

<

0.25,

NBR

>

−

0.23, and

−

0.05

<

NBR2

<

0.15. Clouds, shadows, and other

non-bare ground pixels were removed using the quality assessment band

(QA band). Each pixel in the QA band has a decimal value indicating

surface, atmospheric, and sensor conditions that affect the overall use-

fulness of the data (

USGS, 2020a, 2020b

). The band is in unsigned 16-bit

format and provides information on various pixel conditions, including

water, cloud, cloud shadow, and snow.

We created a mask to eliminate pixels not marked as

“

clear

”

to ensure

that only land surface reflectance was retained. The mask was then

created and applied to the entire satellite image collection, preserving

only bare land surface reflectance data, thereby eliminating the influ-

ence of clouds or snow. It is noteworthy that the GEOS3 enables the

assessment of spatiotemporal variability in two forms: Firstly, by

analyzing all the time series (i.e., from 1984 to 2024), and secondly, by

periods, which may be annually, monthly, or even daily, depending on

the sensor revisit. The resulting product was a space-time cube of

masked images containing bare surface pixels, characterized by a 30 m

resolution and 16 days of revisit.

The images within the space-time cube were then aggregated, using

the sparse spatiotemporal occurrences, into a single composite product

representing the median surface reflectance value named SySi. SySi

captures natural abiotic surfaces, such as bare soil, sand, and rock out-

crops, where vegetation negligibly affects the reflected signal (

). Masked pixels had no data on the SySi and did not impact

processing; these areas are typically associated with natural vegetation

fragments or soil surfaces under frequent succession or crop residues.

2.3. Comparison between SySi and GSSL

The initial assessment involved comparing data from SySi with the

GSSL spectral library. Given their different spectral resolutions, the

spectra from the GSSL were resampled to align with the six reflectance

bands from Landsat sensors, namely, blue (450

–

520 nm), green

(520

–

600 nm), red (630 nm - 690 nm), NIR (770

–

90 nm), SWIR1

(1550

–

1750 nm), and SWIR2 (2080

–

2350 nm) (

Kaufman and Rousseeuw, 1990

e et al., 2018

).

Afterward, Spearman

’

s correlation coefficients served as comparison

parameters for the two spectra types, where |r|

≥

0.7: Strong relation-

ship; 0.4

≤

|r|

<

0.7: Moderate relationship; and |r|

<

0.4: Weak rela-

tionship, generally considered poor for meaningful predictions.

2.3.1. Definition of SySi spectral groups

A quantitative analysis of spectra patterns was conducted at various

locations worldwide to assess the consistency of SySi. This process

involved a clustering process of SySi

’

s pixels, followed by detailed

interpretation. The spectral groups (SGs) were defined using six image

bands through a SOMP, an unsupervised neural network capable of

reducing the dimensionality of large datasets (

Kohonen, 2001

). This

algorithm attenuates the impact of data complexity or potential outliers.

Initially, SOMP was applied solely to SySi reflectance data at the WoSIS

sampling locations.

SOMP used the spectra data (

“

input space

”

) to generate a lower-

dimensional representation (

“

map space

”

). The map space is divided

into a specific number of components (also known as nodes), resulting in

a hexagonal or rectangular grid. In this study, we defined a hexagonal

grid with 36 nodes, which allows for representing the variability of the

data and drastically reducing the dimensionality of the dataset. Every

node in the map space is associated with a weight vector, representing

the node

’

s position in the input space. While the nodes in the map space

remain fixed, the algorithm moves the weight vectors toward the input

data, aiming to minimize the Euclidean distance without compromising

the topology from the map space.

During the process, SOMP performs several successive iterations

until the weight vectors represent, as closely as possible, the input pat-

terns closer to the nodes in the two-dimensional map. In the second stage

of the process, the spatial analysis occurred using a classical clustering

algorithm called the PAM (

). The SOMP

vectors are considered as local averages of the input data. The final

number of clusters was defined according to the methodology proposed

by

Zappia and Oshlack (2018)

, resulting in 10 clusters. The two-stage

method reduces sensitivity to outliers and noise while lowering

computational costs.

Finally, SGs were extrapolated to the entire data SySi by classifying

each pixel according to the similarity to the closest medoids. The simi-

larity between medoids and pixels was calculated using the Euclidean

distance metric. Euclidean distance is particularly suitable for this task

due to its simplicity, intuitive interpretation, and computational

efficiency.

2.3.2. Relationship between SySi spectral groups and environmental factors

We compared various environmental covariates previously used in

global DSM (

Hengl et al., 2017

) with the six synthetic reflectance bands

from SySi. This comparison enabled us to assess associations between 13

relief attributes (i.e., Aspect, Eastness, Elevation, Gaussian Curvature,

Hillshade, Horizontal Curvature, Maximal Curvature, Mean Curvature,

Minimal Curvature, Northness, Shape Index, Slope and Vertical Curva-

ture) and the SySi bands (blue, green, red, NIR, SWIR1 and SWIR2). The

importance of each covariate was calculated using the Boruta algorithm

(

Kursa and Rudnicki, 2010

). During this process, covariates were ranked

by considering five specific soil attributes: clay, sand, silt, carbonates,

and soil organic carbon. We selected these attributes because they are

spectrally active soil constituents frequently used as physicochemical

and biological indicators for soil quality assessments (

Liu et al., 2020

).

Additionally, once the SGs were defined, they were compared to

environmental factors directly related to soils and their formation. The

factors considered in our analysis included global temperature-moisture

domains, relief, land use for the year 2015 (

Hartmann and Moosdorf, 2012

), world

lithology (

), soil classes (

FAO-UNESCO,

1974

), and topsoil textural classes derived from WoSIS samples. Asso-

ciations between factors and SySi were investigated using correspon-

dence analysis (CA) (

Lele et al., 2007

). This analysis enables the

reduction of data dimensions into orthogonal components, which

explain most of the variation. Continuous and categorical data can be

managed with CA in a contingency table (

Viscarra Rossel et al., 2016

). In

our case, a contingency table with units (spectral group) and variables

(environmental factor) was computed and later converted into a smaller

set of dimensions (principal coordinates). Those dimensions were visu-

alized by plotting the scores for SGs alongside each environmental factor

in ordination graphs.

2.3.3. Boruta algorithm to define SySi importance in soil mapping

The Boruta algorithm provides a means to a robust evaluation of

SySi

’

s importance in DSM processes. The method relies on the random

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

6

forests (RF) classifier to rank the most important covariates for pre-

dicting a specific soil attribute (

Kursa and Rudnicki, 2010

). Initially, the

algorithm extends the covariate space by adding randomly permuted

existing covariates (pCs), which eliminates their correlation with the

response variable (soil attribute). Later, RF predicts a specific soil

attribute and calculates the Z-scores (a covariate importance indicator).

The maximum Z-score (MaxZ) among the pCs is defined and actual

covariates that scored better than MaxZ are defined as

“

important

”

and

retained. Attributes with undetermined importance are submitted to a

two-sided test of equality with the MaxZ. Attributes with an importance

significantly lower than MaxZ were defined as ‘unimportant

’

and

permanently removed from the system. Finally, all pCs were removed

and the procedure was repeated. The algorithm was performed recur-

sively until importance was assigned for all the attributes, or the algo-

rithm had reached the previously set limit of the RF runs.

3. Results and discussion

3.1. Data consistency

Addressing the need for more detailed soil information, this research

“

uncovered

”

90 % of Earth

’

s bare soils under agricultural or natural land

use and land cover at a resolution of 30 m. This product not only offers

finer resolution than those previously available (

≥

250 m) but also

demonstrates high consistency considering early attempts (i.e.,

). This approach can be considered a method of sus-

tainable survey, as it minimizes environmental impacts, costs, and the

time required for execution, reducing analyses needed for large-scale

surveys (

McBratney et al., 2018

). None of the other studies that pro-

duced images of bare surfaces at global scales had reached this scope and

detail level (

Poggio et al., 2021

). The EMR is regionally dependent, and

the algorithm evaluates each case throughout the entire globe.

Furthermore, only a limited number of these studies evaluated the

reliability of RS data (cf.

Safanelli et al., 2020

Tziolas et al., 2020a,

2020b

Huete and Escadafal, 1991

On data consistency, a rigorous control process for soil sampling and

laboratory measurements showed correlation values between observed

and predicted data. Accuracy concerning field observation data varies

from moderate to strong relationships (

). The

values ranged from 0.5 to 0.9. As an example of this consistency, spectra

over the Syrian territory exhibited higher correlation values than those

reported in studies with a local focus. This endeavor is noteworthy,

considering the constraints imposed by the scale of our analysis and the

utilization of a global dataset derived from multiple sources (

Poggio

et al., 2021

).

Table 1

presents the spectra obtained by SySi in comparison with

those from the Global Soil Spectral Library (GSSL), which yielded sig-

nificant results (p

<

0.0001). Spearman

’

s correlations (r) ranged from

0.67 to 0.84, with higher values observed for the blue, green, and SWIR

spectral bands, key ranges for soil properties modeling (

). This observation indicates that considering the whole studied

area of bare soil, the filtering of the GEOS3 system found relative con-

sistency between the ground and space sensing data.

3.2. Relation between SySi

’

s spectral behavior and environmental factors

The SySi successfully captured the Earth

’

s surface reflectance pat-

terns intrinsic to environmental variability (

a), enabling the

observation of spectral characteristics at various locations worldwide

(

b). Despite focusing on only six bands from Landsat sensors, it

represents the wavelengths in the VNIR-SWIR spectral ranges and a

practical evaluation was achieved by analyzing spectral shapes and

reflectance patterns using SySi. Ten spectral groups (SG) represented the

diversity of the world

’

s soils. Subsequently, these groups were compared

with environmental factors directly influencing soil formation and

characteristics.

SGs 1 and 2 were the most common groups (

b), primarily

associated with hilly croplands and grasslands in tropical and temperate

regions. The parent material consists of unconsolidated sediments,

metamorphic rocks, or, in some cases, evaporites. These factors are

predominantly found in medium-textured soils (in SG1). Furthermore,

SG 1 and 2 spectra showed similar characteristics, with medium

reflectance intensity and convex shape. The higher reflectance intensity

of SG1 is attributed to the increased presence of coarse particles and the

soils

’

lower soil organic carbon content (SOC), distinguishing it from

SG2 (

e et al., 2020

). This finding can be used to distinguish soil, i.

e.: by texture (

), SOC (

), color (

), or floodplains (

Joint observations of

a and b show that SG3 prevailed in plain

pastures and croplands, highlighted in western China and western Asia.

Similarly, SGs 4 and 5 predominated in natural landscapes with complex

relief. SGs 3, 4, and 5 exhibited the lowest reflectance intensities, with

shapes varying from convex (SG3) to flat or concave (SG5) caused by the

soil physicochemical conditions. These three SGs also exhibited the

highest SOC among all SGs, which reduced reflectance in the VNIR re-

gion, particularly in the visible region, as

Stoner and Baumgardner

(1981)

observed. Interestingly, SG3 was associated with soils containing

argic horizons (i.e., Phaeozems, and Planosols), resulting in lower soil

surface clay content due to remotion processes (

Buol et al., 2011

As shown in

a, SG4 corresponded to soil classes with dark

surface horizons because the curve presents low albedo in the visible

region. As an example of these soil types, we identified the Chernozems,

Vertisols, or Rendzinas, with blackish colors due to organic matter

accumulation or moisture (

Liu et al., 2020

). The map of soil groups

(

b) showed a higher occurrence of SG4 in some regions of Russia,

Ukraine, and Kazakhstan, which is consistent with the occurrence of the

soil as mentioned earlier in classes of the

FAO-UNESCO (1974)

world

soil map (see Supplemental Figs. S1 and S2). Nevertheless, the low

reflectance of SG 5 (

Stoner and Baumgardner, 1981

a) is influenced by parent material, specifically

volcanic rocks rich in mafic minerals. These minerals absorb more

electromagnetic energy, producing a distinctive pattern in the soil

spectra (

). Moreover, the fine and very

fine soil texture of the locations with SG5 (

b) also contributed to

the predominance of these reflectance patterns.

In contrast, SGs 6 to 10 have been observed mainly in tropical,

subtropical, or temperate deserts, where soils are mainly unconsolidated

with higher quartz or salt concentrations reflecting most light that af-

fects them. These characteristics and the low SOC content are the main

reasons for the highest reflectance (

Howari et al., 2002

;

Stenberg and

Viscarra-Rossel, 2010

). SG 6 was observed sparsely throughout Asia,

while SG 7 was predominantly found in the Namibia Desert (south-

western Africa) and the Thar Desert (India). These two groups were

primarily associated with Yermosols, soils common in semiarid to arid

hot climates.

SG 8 corresponded to Arenosols and Solonchaks, sandy and salic

soils, observed in a semi-desert area south of the Sahara, the Simpson

Desert (Australia), or temperate deserts such as the Taklamakan in

China. Desert areas with salt-affected soils (e.g., Solonchaks) were also

observed in SG 9, highlighting the potential of remote sensing products

Table 1

Spearman correlation values between satellite data and laboratory spectra (n

=

104,004).

Simulated bands (SySi)

Landsat bands
Blue

Green

Red

NIR

SWIR l

SWIR 2

Blue

0.84

0.83

0.69

0.71

0.64

0.51

Green

0.82

0.71

0. 72

0.65

0.51

Red

0.71

0.74

0.69

0.70

0.63

0.53

NIR

0.75

0.78

0.73

0.75

0.69

0.59

SWIR 1

0.77

0.80

0.76

0.79

0.77

0.66

SWIR 2

0.74

0.77

0.74

0.77

0.75

0.67

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

7

such as SySi in identifying and mapping such soils (

Naimi et al., 2021

;

Wang et al., 2021

). Finally, SG 10 was the dominant group in the Sahara

and is strongly associated with Yermosols (

FAO-UNESCO, 1974

).

Considering the characteristics of such soils and the regional climatic

conditions (tropical desert), we expected a higher intensity of this SG

due to its very convex shape. Middle-Western and boundary of the

Sahara, with sandy soil prevalence, are the most representative regions

with the SG 10 (

b).

3.3. SySi relevance for Earth topsoil modeling

The findings indicate a strong correlation between SySi and topsoil

attributes or soil classes suggesting that it is a valuable tool for soil

characterization, mapping, and classification (

e et al., 2020

)

although only the soil surface was evaluated. It has been demonstrated

that bare surface images can enhance DSM, particularly at local and

regional scales (

Ng et al., 2023

Urbina-Salazar et al., 2023

;

Rizzo et al.,

2023

). Consequently, we compared the environmental covariates pre-

viously utilized in the global DSM (

) and the six reflectance bands from the SySi. Similarities in soil

patterns can be observed in both grids (

Fig. 4

a and b). The potential of

these variables to support global mapping, such as the mapping of

topsoil color, was discussed following

Rizzo et al. (2023)

.

Spectral patterns are present at specific positions due to combina-

tions of specific environmental factors such as climate, land use, and

geology. The long-term mean of precipitation and temperature, repre-

senting climatic covariates, were weakly to moderately relevant to infer

soil properties (

a). We expected a more robust correlation, espe-

cially considering that different soil moisture regimes affect the leaching

of base cations, enriching the surface layers of soil with pigments such as

oxide. This climatic factor is the leading cause of horizontal and vertical

variations in soil color (

Liu et al., 2020

). Many factors influence the

comparison with climate zones, but an important consideration is the

low accuracy of the climate maps. Soil spectral data can change drasti-

cally within a small area, thus, higher spatial resolution covariates could

enhance inference on spatially explicit soil properties and soil forming

processes.

Among the LULC covariates (

b), Forest, Cropland, Shrubland

and Grassland achieved the highest importance in the model, as these

LULC types exert a strong influence on reflectance (

). We found that only the plain areas were moderately relevant

among relief data (

c). Topographic landscape covariates are

Fig. 3.

Soil spectral patterns worldwide. a) Mean spectra for the ten soil spectral groups with minimum and maximum values and b) map of the ten soil spectral

groups (SG).

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

8

related to water dynamics and indicate erosion intensity, redistribution

of soil material, particle sorting processes, and thermal regimes. Some of

the terrain attributes are critical covariates (e.g., Topographic Wetness

Index, Multiresolution Index of Valley Bottom Flatness, and LS factor) in

DSM (

McBratney et al., 2018

). However, these secondary topographic

data were not used in our study due to their computational demand

which makes the processing difficult, even on cloud computing plat-

forms (

Safanelli et al., 2020

c shows the relation of SG. SG1

predominates arid (i.e., Middle East) and semiarid areas (i.e., north-

eastern Brazil) while SG4 prevails in mountainous landscapes (i.e., north

and southwestern Asia).

Lithology showed a strong relationship with soil spectral behavior.

Basaltic rock areas were close to SG5 (

d) as are Ferralsols (

f).

Indeed, these patterns are consistent with field observation, such as the

typical occurrence of Ferralsols on Basalt in southwestern Brazil (

Gallo

et al., 2018

). In summary, all this information is essential for decision-

making regarding managing or protecting specific areas.

3.4. Remote sensing informed global analysis of soil variation and

diversity

Using a surface soil proxy is essential in quantifying various soil at-

tributes as

Clingensmith and Grunwald, 2022

e et al. (2020)

have already observed. This approach is

well supported by

>

50 years of laboratory research focused on estab-

lishing relationships between soil attributes and reflectance, mainly

conducted in laboratory settings (

). Indeed, the statistics of each variable of the matrix (

Fig. 6

) from

our study show the significant importance of soil properties and spectral

data on a global scale. Elevation and SWIR2 band show the highest

importance with 78 and 80 %, respectively, followed by spectral data,

while the lowest values were found for SOC and calcium carbonate. The

combination of elevation and spectral data achieved a higher R

2

than

using elevation alone, which is in agreement with the findings of

Fon-

garo et al. (2018)

The evidence presented here suggests that SySi bands provide critical

information to describe the spatial variability of topsoil color, which is

consistent with findings by

Rizzo et al. (2023)

. This statement is justified

because bare surface images capture real-time data about soils through

reflectance. While SySi bands effectively capture topsoil color vari-

ability, they have several limitations: weak correlations with specific

soil properties, inadequate representation of certain soil groups,

spatially varying accuracy due to environmental factors, uncertainties

arising from the sensor or atmospheric conditions, and the generaliza-

tion of temporal aspects by harmonizing the pixels. These issues should

be considered to improve the reliability of soil map products.

3.5. Insights for societal issues

The majority of DSM studies do not use SySi information and

therefore may face difficulties in capturing adequate soil variability,

which is a challenge for accurate pedological delineation and mapping.

Fig. 4.

Distribution of spectral patterns and statistics over Earth

’

s bare surfaces. (a) True-color image of bare surfaces (Landsat ETM

+

band 1

=

blue, band 2

=

green,

and band 3

=

red). (b) Red, green, and blue (RGB) image depicting pixel-based spatialization of the three principal components (PC1, PC2, PC3) on the bare surfaces.

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

9

For example, users cannot observe soil variation beneath the soil surface

(

a). In

b, a comparison is shown between pedological

delineation without SySi (cyan line) and with SySi (brown line). SySi

provides critical important background information that makes the

result different and affects several subsequent decisions, such as soil

management, fertilization, and conservation.

c highlights a clas-

sical soil survey delineation (yellow line) based on the site (borehole)

plot and relief. Applying the SySi (

d), the areas with different

spectra (related to other soil properties) are highlighted and affect new

locations to insert the sites, causing an utterly new delineation, as

Safanelli et al. (2021)

observed in a similar situation.

Both conventional (

Hartemink et al., 2013

Negasa et al., 2017

) and

digital (

Chen et al., 2022

;

Ma et al., 2019

) soil sampling methods may

fail to capture the full range of soil variation if they do not consider soil

Fig. 5.

Comparison of the position of spectral groups (red circles) to a) climate zones of the world (

), b) land use and land cover (

),

c) landforms distribution (

Hartmann and Moosdorf, 2012

), d) Earth

’

s geological groups (

), e) soil textural classes (

Batjes et al., 2017

), and f) soil

classes (

FAO-UNESCO, 1974

) (a, b, c, d, e, in black circles).

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

10

as a continuous feature across the landscape. Conventional surveying

typically relies on manual sampling and observation, with surveyors

collecting soil samples based on predefined locations, often using relief-

focused approaches. In contrast, digital surveying incorporates tech-

nology, spatial data, and modeling techniques, such as RS and GIS, to

map soil properties more continuously across a landscape. Traditional

surveyors typically use relief to delineate soil classes on the landscape;

however, when bare soil information is added, multiple underlying

variations are revealed, leading to significant changes in delineation. In

summary, when positioning soil samples, SySi data could be considered,

when available, to ensure a more accurate representation. For instance,

incorporating SySi data in stratified random sampling designs is a

Fig. 6.

Contribution of SySi bands and terrain covariates with soil attributes modeling (sand, silt, clay, carbonate, and SOC). Values in percentage (%).

Fig. 7.

Scale is not presented due to high variability ranging from farm to state scale. The examples are from Brazil. a) A Google image in true color composition

showing sugarcane crop in Goi

´

as state, Brazil, b) SySi 5R4G3B composition with traditional soil survey (using only relief) in blue lines, and new soil type surveyed in

brown lines (using relief plus SySI), c) conventional soil survey delineation (yellow line) at a site; d) the same site with new delineation using SySi and the locations to

soil sampling, e) precision agriculture grid sampling, f) same site with new soil site allocation using SySi, g) a Google Earth image with natural color covered by

sugarcane cultivation, h) the same site with SySi, highlighting soil variations, i) the same area with elevation model, j) SySi image of S

˜

ao Paulo, and P

´

arana States,

Brazil, in 5R4G3B color composition, k) soil potential productivity map(green indicates a high potential and red indicates a low potential) (

Greschuk et al., 2023

), l)

Google Earth image of a cropland area, and m) same area with SySi 5R4G3B composition (

Mello et al., 2023

), the red arrows go from the laboratory sample pictures

to the sampling places.

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

11

valuable asset to improve the placement of soil sampling locations in

future studies.

Precision agriculture and soil management zoning are strategies for

collecting soil samples at specific locations to guide nutrient manage-

ment with optimized fertilizer applications. Optimizing soil manage-

ment relies on accurately mapping soil variability controlled by various

soil-environmental factors (

e, f). In the first case, if a farmer makes

a regular grid, it is not possible to observe the differences. However,

with SySi the sampling allocation can be targeted to areas of soil vari-

ation, which subsequently improves the accuracy of the precision

farming system.

g, h, and i compare a standard image, an image

with SySi, and an image with only relief (DEM). Soil variation does not

necessarily correspond to relief variation; thus, a pedologist can only

interpret the delineation using this traditional information. Indeed, pe-

dologists frequently delineate the soil transitions based on landforms,

with or without the support of aerial photographs or altitude maps.

However, SySi provides new information to be integrated into this

process. It captures direct measurements of the topsoil, representing the

diversity of multiple soil properties. Thus, while the relief may appear

uniform, spectra reveal changes that would otherwise go unnoticed.

Pedologists likely lose accuracy in their delineations of soils without

incorporating soil spectral information.

Spectral reflectance can be linked to particle size distribution, carbon

content, carbon stock, pH, chemistry elements, clay content and

mineralogy (

Nocita et al., 2015

Novais et al., 2025

). Therefore, it is

reasonable to expect that spectral data could not only help improve DSM

and agricultural productivity but also be applied to assess soil health.

j shows areas of bare soil in four Brazilian states (Paran

´

a, S

˜

ao

Paulo, Minas Gerais, and Mato Grosso do Sul), where lighter colors

represent soils formed from sandstone, and dark purple indicates basalt.

k shows a map of soil potential productivity (

Greschuk et al.,

2023

), where green features represent higher potential productivity,

while red areas represent lower potential productivity. These latter areas

are characterized by high clay, iron, and carbon content. Notably, areas

of high productivity correlate with regions shown in dark blue in the

SySi images.

l and m shows that SySi images reveal sandy areas (patches)

with poor drainage within a generally clayey site, which agrees with

Mello et al. (2023)

. In addition, significant relationships between the

bare color composition images and soil types could be observed

worldwide. Supplemental notes 1 and 2 detail the performance of

spectral clustering for Asia and America in Figs. S1 and S2. We observe

in many cases that areas indicated as sandy soil were clayey. This

behavior occurs because straw plays a critical goal and sometimes

cannot be depicted by the algorithm (

4. Challenges and limitations of the approach

Despite the promising results, there is still room for improvement in

the bare soil spectra. We observed inconsistencies in areas with high

straw covering the soil surface. We also observed the importance of

laboratory spectral libraries to train the model and validate the data

from a satellite. The more robust and representative SSL the better the

model at the satellite level. Thus, obtaining these inconsistencies in the

final product is plausible, while human experts must be involved in the

interpretation of the spectra. We strongly encourage efforts to improve

algorithms on the detection of bare soil, for instance, identifying soils

free of vegetation but remaining wet. On the one hand, upcoming

hyperspectral images (i.e., EMIT, PRISMA, EnMAP, CHIME, and SBG)

may partially address this issue as they are relatively recent in terms of

spatial coverage and are not designed to capture the entire globe (except

the Terrestrial Ecosystem and Airborne Observatory with Global

Ecosystem Resolution - TANAGER-1). On the other hand, globally

available Landsat and Sentinel data, for instance, are valuable because

they can image larger areas with bare soil, particularly as no-till agri-

cultural practices become more common. The new hyperspectral data

will gain in quality but lose coverage over larger spatial areas. There-

fore, we should focus on both spatial resolution and spectral range in the

future.

Working with RS and open (public) data poses several issues and

limitations. Traditionally, soil carbon analysis has been performed by

Walkley and Black

’

s method, despite the combustion method being

considered the most accurate (

Drover and Manner, 1975

). Although,

currently an extensive dataset is unavailable. Thus, we have to work

with what we have until soil carbon monitoring adopts state-of-the-art

soil analytics and new-generation global databases (

Novais et al.,

2025

). Another point is that wet soil laboratories can have differences in

their analysis, which will also impact the sensor analysis (

Paiva et al.,

2022

). Importantly, the spectral reflectance is dependent on the soil

analysis. Open data, even when fused with private data, comes from

different soil laboratories which add uncertainties. The timing of the soil

analysis also matters, mostly for carbon. A carbon analysis from 1980 in

a given site, can be different in the same site 20 years later. Thus, carbon

analysis made at different times also interferes with the relationship of

spectral modeling.

The RS technique also has issues. In the laboratory, there are fewer

uncertainties in measuring the reflectance and harmonization of SSL can

be done today using the IEEE P4005 WG strategy (

Ben-Dor et al., 2024

https://esalqgeocis.wixsite.com/english

Orbital sensor harmonization and calibration is still an issue that needs

more attention when compared to laboratory settings (

Ben-Dor et al.,

2024

). The sensors inside the Landsat system, and any other platform,

also degrade over time which may impact the spectral result if routine

radiometric calibration is not well performed. The spectral resolution is

another issue for multispectral sensors since it is not able to detect the

spectral features of straw or any other residue in the field. Other issues

could be cloud interference, soil moisture variations, and aggregation

(

Ben-Dor et al., 2009

). Many of these issues were overcome by auto-

mated systems that can choose the best images and low cloud coverage,

evaluate the best pure pixel for soil, and create the median of the spectra

to diminish the error.

An overlap between the dates of RS data acquisition and the soil

sampling can be achieved in projects that use legacy and continuous

sampling when new data is being collected. The use of historical soil

data, such as many of the soil sampling points in the WoSIS dataset,

typically leaves a significant time gap between the date of sampling and

the date of RS data acquisition. This would be expected to introduce

additional uncertainty, especially if the work is using bare soil com-

posite images to quantify soil properties such as SOC. This element is

known to change under agricultural land management practices. The

same is true for other work using remotely sensed data stacks for pre-

diction. For example,

Sothe et al. (2022)

produced a SOC map of Canada

using soil sample data from 1977 to 1995 and RS data from 2000 to

2019. Does this work reflect SOC nowadays? It is more likely a map of

SOC from the median age of the data, as we have also stated in our

manuscript. Significant losses of SOC in agricultural landscapes over the

last 40 years make this temporal gap a notable challenge for future

research. Despite this problem, the technique still provides useful in-

sights into SOC, with less uncertainty in some regions and more uncer-

tainty in others. Therefore, SOC content assessments must be

approached with caution.

In summary, spatial resolution matters to capture a mixture of soil

types in a given field. The spectral resolution limitation may not mimic

the absorption bands of many chromophores, and thus, underestimate

the real soil condition. The mismatch between soil sample support (i.e.,

a small soil sample collected over an area of about 20 cm

2

, considering

the augered borehole) and spatial resolution of the sensor (e.g., Landsat

with 30 m spatial resolution) may add ambiguities in the soil-spectral

modeling. It is necessary to consider that electromagnetic energy

comes from the sun for 149,600,000 km at lightspeed until it reaches the

Earth

’

s surface. The EMR crosses the atmosphere, reaches the soil and

reflects and goes back to the sensor for

>

800 km from the target. In this

way, it suffers interference from aerosols such as clouds and smog. After

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

12

addressing all these challenges and issues, we achieved a Spearman

’

s

correlation between soil laboratory and satellite spectra between 0.51

and 0.84, with 80 % of the correlations over 0.7 (

Table 1

). This outcome

demonstrates coherence between bare soil information and field data.

5. Conclusions

We successfully developed Earth

’

s topsoil reflectance spectra at a 30

m spatial resolution. The GEOS3 processing is the system and the SySi

the product that effectively covers agricultural and natural areas,

providing societal insights. The Landsat time series spanned over 38

years of imaging, and its overlapping data has demonstrated a signifi-

cant similarity to field truth spectra. Therefore, the system has

“

un-

covered

”

soil in over 90 % of the world

’

s agricultural areas. Considering

that the spectral signature was captured by a sensor 800 km above the

ground with a spatial resolution of 30 m, while the soil sample was

collected in the field at 0.1

×

0.1 m, this outcome is an impressive

advancement in soil science. We also obtained ten different spectral

groups regarding the shape and intensity of specific soil properties.

The spectral data, even being limited, showed remarkable variation,

reflecting each specific soil type

’

s unique and complex nature. As a rich

information source, the spectra can be integrated into pedo-transfer

functions to estimate several soil properties, such as sand, silt, clay,

cation exchange capacity, mineral composition, and nutrients, which

showed a significant correlation with SySi. Elevation and SWIR 1 and 2

were the variables that explained the higher importance in correlation

with these attributes. GEOS3 also presented a contiguous map that im-

proves soil synoptic visualization, decreasing errors in mapping. This

may pave the way for improved soil monitoring and facilitate an un-

derstanding of soil variation and the operationalization of smart

agriculture.

The global SySi provides a valuable tool for monitoring and

analyzing soil properties across diverse landscapes. These findings can

impact disciplines such as DSM, soil surveying, precision agriculture,

soil productivity, land use planning, tillage, conservation management,

and irrigation. This product can guide the future allocation of soil

sampling sites, thereby reducing the costs of human labor, soil analysis,

and field campaigns. They can also be helpful for scientists, consultants,

farmers, and policymakers and complement climate monitoring efforts.

In the future, as higher spectral, spatial and temporal resolution data

become available, this paper can serve as a proof of concept to inform

future global studies mapping topsoil properties in a novel way. Despite

this, we advocate that there is still space to explore create new algoritms

for bare soil detection.

CRediT authorship contribution statement

Jos

´

e A.M. Dematt

ˆ

e:

Writing

–

review

&

editing, Writing

–

original

draft, Supervision, Resources, Funding acquisition, Formal analysis,

Data curation, Conceptualization.

Rodnei Rizzo:

Visualization, Meth-

odology, Investigation, Formal analysis, Conceptualization.

Nícolas

Augusto Rosin:

Software, Methodology.

Raul Roberto Poppiel:

Soft-

ware, Methodology, Formal analysis, Conceptualization.

Jean Jesus

Macedo Novais:

Writing

–

review

&

editing, Visualization, Validation,

Investigation, Formal analysis.

Merilyn Taynara Accorsi Amorim:

Software, Methodology.

Heidy Soledad Rodriguez-Albarracín:

Soft-

ware, Methodology.

Jorge Tadeu Fim Rosas:

Software, Methodology.

Bruno dos Anjos Bartsch:

Software, Methodology.

Letícia Guadagnin

Vogel:

Software, Methodology.

Budiman Minasny:

Writing

–

review

&

editing, Supervision, Formal analysis.

Sabine Grunwald:

Writing

–

re-

view

&

editing, Data curation.

Yufeng Ge:

Writing

–

review

&

editing,

Data curation.

Eyal Ben-Dor:

Writing

–

review

&

editing, Supervision,

Formal analysis.

Asa Gholizadeh:

Writing

–

review

&

editing, Data

curation.

Cecile Gomez:

Writing

–

review

&

editing, Data curation.

Sabine Chabrillat:

Writing

–

review

&

editing, Data curation.

Nicolas

Francos:

Writing

–

review

&

editing, Data curation.

Dian Fiantis:

Writing

–

review

&

editing, Data curation.

Abdelaziz Belal:

Writing

–

review

&

editing, Data curation.

Nikolaos Tsakiridis:

Writing

–

review

&

editing, Data curation.

Eleni Kalopesa:

Writing

–

review

&

editing,

Data curation.

Salman Naimi:

Writing

–

review

&

editing, Data cura-

tion.

Shamsollah Ayoubi:

Writing

–

review

&

editing, Data curation.

Nikolaos Tziolas:

Writing

–

review

&

editing, Data curation.

Bhabani

Sankar Das:

Writing

–

review

&

editing, Data curation.

George Zalidis:

Writing

–

review

&

editing, Data curation.

Marcio Rocha Francelino:

Writing

–

review

&

editing, Data curation.

Danilo Cesar de Mello:

Writing

–

review

&

editing.

Najmeh Asgari Hafshejani:

Writing

–

re-

view

&

editing, Data curation.

Yi Peng:

Writing

–

review

&

editing, Data

curation.

Yuxin Ma:

Writing

–

review

&

editing, Conceptualization.

Jo

˜

ao Augusto Coblinski:

Writing

–

review

&

editing, Data curation.

Alexandre M.J.-C. Wadoux:

Writing

–

review

&

editing, Conceptuali-

zation.

Igor Savin:

Writing

–

review

&

editing, Data curation.

Brendan

P. Malone:

Writing

–

review

&

editing, Data curation.

Konstantinos

Karyotis:

Writing

–

review

&

editing, Data curation.

Robert Milewski:

Writing

–

review

&

editing, Data curation.

Emmanuelle Vaudour:

Writing

–

review

&

editing, Data curation.

Changkun Wang:

Writing

–

review

&

editing, Data curation.

Elsayed Said Mohamed Salama:

Writing

–

review

&

editing, Data curation.

Keith D. Shepherd:

Writing

–

review

&

editing, Investigation, Formal analysis, Data curation.

Declaration of competing interest

The authors declare that they have no known competing financial

interests or personal relationships that could have appeared to influence

the work reported in this paper.

Acknowledgments

First Author ackowledge the CNPq for a researcher scholarship. The

authors would like to express their gratitude to the S

˜

ao Paulo Research

Foundation (FAPESP) for the financial support provided through grant

number 2021/05129-8. The authors also wish to thank researchers and

institutions in other countries for providing data. We thank the GeoCiS

research group

for technical

support.

Appendix A. Supplementary data

Supplementary data to this article can be found online at

org/10.1016/j.scitotenv.2025.178791

https://doi.org/10.1016/j.geodrs.2020.e0025z

Data availability

The authors do not have permission to share data.

References

Arrouays, D., McBratney, A., Bouma, J., Libohova, Z., Richer-de-Forges, A.C., Morgan, C.

L.S., Roudier, P., Poggio, L., Mulder, V.L., 2020. Impressions of digital soil maps: the

good, the not so-good, and making them ever better. Geoderma Reg. 20, e00255.

https://doi.org/10.5194/essd-9-1-2017

Batjes, N.H., Ribeiro, E., van Oostrum, A., Leenaars, J., Hengl, T., Mendes de Jesus, J.,

2017. WoSIS: providing standardized soil profile data for the world. Earth Syst Sci

Data 9, 1

–

14.

https://doi.org/10.5194/essd-12-299-2020

Batjes, N.H., Ribeiro, E., Van Oostrum, A., 2020. Standardized soil profile data to support

global mapping and modelling (WoSIS snapshot 2019). Earth Syst Sci Data 12,

299

–

320.

.

Ben-Dor, E. and Dematt

ˆ

e, J.A.M. Remote sensing of soil in the optical domains. (2024).

In: Remote Sensing Handbook, Second edition, Ed. Prasad S. Thenkabail, Taylor and

Francis Inc Press, pg. 643

–

703, DOI:

https://doi.org/10.1201/9781003442127-21

https://doi.org/10.1016/j.rse.2008.09.019

(November 2024. Chapter 16).

Ben-Dor, E., Chabrillat, S., Dematt

ˆ

e, J.A.M., Taylor, G.R., Hill, J., Whiting, M.L.,

Sommer, S., 2009. Using imaging spectroscopy to study soil properties. Remote Sens.

Environ. 113, S38

–

S55.

https://doi.org/10.1016/j.geoderma.2023.116636

Ben-Dor, E., Granot, A., Wallach, R., Francos, N., Heller Pearlstein, D., Efrati, B.,

Bor

ů

vka, L., Gholizadeh, A., Schmid, T., 2023. Exploitation of the SoilPRO

®

(SP)

apparatus to measure soil surface reflectance in the field: five case studies. Geoderma

438, 116636.

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

13

Ben-Dor, E., Karyotis, K., Chabrillat, S., Schmid, T., 2024. Standardization and protocol

development for soil spectral reflectance in laboratory and field settings: an

overview of IEEE SA P4005 activity. In: IGARSS 2024-2024 IEEE International

Geoscience and Remote Sensing Symposium. IEEE, pp. 268

–

270.

10.1109/IGARSS53475.2024.10641689

Buol, S.W., Southard, R.J., Graham, R.C., McDaniel, P.A., 2011. Soil genesis and

classification: sixth edition. In: Soil Genesis and Classification: Sixth Edition.

doi.org/10.1002/9780470960622

.

Chen, S., Arrouays, D., Mulder, V.L., Poggio, L., Minasny, B., Roudier, P., Walter, C.,

2022. Digital mapping of GlobalSoilMap soil properties at a broad scale: a review.

Geoderma 409, 115567.

https://doi.org/10.1016/j.geoderma.2021.115567

.

Clingensmith, C.M., Grunwald, S., 2022. Predicting soil properties and interpreting Vis-

NIR models from across continental United States. Sensors Journal 22 (Article 3187),

1

–

17.

https://doi.org/10.3390/s22093187

.

Dematt

ˆ

e, J.A.M., Alves, M.R., Terra, F.S., Bosquilia, R.W.D., Fongaro, C.T., Barros, P.P.S.,

2016. Is it possible to classify topsoil texture using a sensor located 800 km away

from the surface? The Brazilian Journal of Soil Science 40, e0150335.

org/10.1590/18069657rbcs20150335

Dematt

ˆ

e, J.A.M., Fongaro, C.T., Rizzo, R., Safanelli, J.L., 2018. Geospatial Soil Sensing

System (GEOS3): a powerful data mining procedure to retrieve soil spectral

reflectance from satellite images. Remote Sens. Environ. 212, 161

–

175.

org/10.1016/j.rse.2018.04.047

.

Dematt

ˆ

e, J.A.M., Safanelli, J.L., Poppiel, R.R., Rizzo, R., Qui

˜

nonez Silvero, N.E.,

Mendes, W. de S., Bonfatti, B.R., Dotto, A.C., Urbina Salazar, D.F., Mello, F.A. de O.,

Paiva, A.F. da S., Souza, A.B., Santos, N.V. dos, Nascimento, C.M., de Mello, D.C.,

Bellinaso, H., Neto, L.G., Amorim, M.T.A., Resende, M.E.B. de, Vieira, J. da S.,

Queiroz, L.G. de, Gallo, B.C., Say

˜

ao, V.M., Lisboa, C.J. da S., 2020. Bare Earth

’

s

surface spectra as a proxy for soil resource monitoring. Sci. Rep. 10, 4461.

doi.org/10.1038/s41598-020-61258-w

.

Dematt

ˆ

e, J.A.M., Paiva, A.F.d.S., Poppiel, R.R., Rosin, N.A., Ruiz, L.F.C., Mello, F.A.d.O.,

Minasny, B., Grunwald, S., Ge, Y., Ben-Dor, E., Gholizadeh, A., Gomez, C.,

Chabrillat, S., Francos, N., Ayoubi, S., Fiantis, D., Biney, J.K.M., Wang, C., Belal, A.,

Silvero, N.E.Q., 2022. The Brazilian Soil Spectral Service (BraSpecS): a user-friendly

system for global soil spectra communication. Remote Sens. 14 (3), 740.

org/10.3390/rs14030740

https://doi.org/10.1080/00103627509366586

Drover, D.P., Manner, H., 1975. A comparison between the walkley-black and a dry

combustion method for determining organic carbon in a humic brown soil, Papua

New Guinea. Communica- tions in Soil Science and Plant Analysis 6 (5), 495

–

500.

https://doi.org/10.1016/bs.agron.2023.10.00

Evangelista, S.J., Field, D.J., McBratney, A.B., Minasny, B., Ng, W., Padarian, J.,

Dobarco, M.R., Wadoux, A.M.J.-C., 2024. Soil security-zstrategizing a sustainable

future for soil. Adv. Agron. 183.

.

FAO, 1974. Soil Map of the World. Volume I, Legend. Unesco, Paris.

https://data.apps.fa

o.org/map/catalog/srv/eng/catalog.search#/metadata/cc45a270-88fd-11da

-a88f-000d939bc5d8

http://www.fao.org/3/a-i5199e.pdf

FAO and ITPS, 2015. Status of the World

’

s Soil Resources (SWSR)

–

Main Report. Food

and Agriculture Organization of the United Nations and Intergovernmental

Technical Panel on Soils, Rome, Italy.

(650 p.).

Fongaro, C.T., Dematt

ˆ

e, J.A.M., Rizzo, R., Lucas Safanelli, J., Mendes, W.D.S., Dotto, A.

C., Vicente, L.E., Franceschini, M.H.D., Ustin, S.L., 2018. Improvement of clay and

sand quantification based on a novel approach with a focus on multispectral satellite

images. Remote Sens. 10 (10), 1555.

https://doi.org/10.3390/rs10101555

https://doi.org/10.1016/j.

Francos, N., Ben-Dor, E., 2022. A transfer function to predict soil surface reflectance from

laboratory soil spectral libraries. Geoderma 405, 115432.

geoderma.2021.115432

.

Gallo, B.C., Dematt

ˆ

e, J.A.M., Rizzo, R., Safanelli, J.L., Mendes, W.S., Lepsch, I.F., Sato, M.

V., Romero, D.J., Lacerda, M.P.C., 2018. Multi-temporal satellite images on topsoil

attribute quantification and the relationship with soil classes and geology. Remote

Sens. 10, 1571.

https://doi.org/10.3390/rs10101571

.

Gomez, C., Vaudour, E., F

´

eret, J.-B., de Boissieu, F., Dharumarajan, S., 2022. Topsoil clay

content mapping in croplands from Sentinel-2 data: influence of atmospheric

correction methods across a season time series. Geoderma 423, 115959.

org/10.1016/j.geoderma.2022.115959

.

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017.

Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sens.

Environ. 202, 18

–

27.

https://doi.org/10.1016/j.rse.2017.06.031

.

Greschuk, L.T., Dematt

ˆ

e, J.A.M., Silvero, N.E.Q., Rosin, N.A., 2023. A soil productivity

system reveals most Brazilian agricultural lands are below their maximum potential.

Sci. Rep. 13, 14103.

https://doi.org/10.1038/s41598-023-39981-y

https://doi.org/10.3389/fsoil.2021.714323

Grunwald, S., 2021. Grand challenges in pedometrics-AI research. Frontiers in Soil

Science-Pedometrics 1, 714323.

(1

–

8).

Hartemink, A.E., Krasilnikov, P., Bockheim, J.G., 2013. Soil maps of the world.

Geoderma 207

–

208, 256

–

267.

https://doi.org/10.1016/j.geoderma.2013.05.003

.

Hartmann, J., Moosdorf, N., 2012. The new global lithological map database GLiM: a

representation of rock properties at the Earth surface. Geochem. Geophys. Geosyst.

13, 12004.

https://doi.org/10.1029/2012GC004370

.

Hengl, T., Mendes de Jesus, J., Heuvelink, G.B.M., Ruiperez Gonzalez, M., Kilibarda, M.,

Blagoti

´

c, A., Shangguan, W., Wright, M.N., Geng, X., Bauer-Marschallinger, B.,

Guevara, M.A., Vargas, R., MacMillan, R.A., Batjes, N.H., Leenaars, J.G.B.,

Ribeiro, E., Wheeler, I., Mantel, S., Kempen, B., 2017. SoilGrids250m: global gridded

soil information based on machine learning. PLoS One 12 (2), e0169748.

doi.org/10.1371/journal.pone.0169748

.

Howari, F.M., Goodell, P.C., Miyamoto, S., 2002. Spectral properties of salt crusts formed

on saline soils. J. Environ. Qual. 31, 1453

–

1461.

https://doi.org/10.2134/

jeq2002.1453

Huete, A.R., Escadafal, R., 1991. Assessment of biophysical soil properties through

spectral decomposition techniques. Remote Sens. Environ. 35, 149

–

159.

org/10.1016/0034-4257(91)90008-T

https://doi.org/10.1016/j.agsy.2016.02.010

J

´

onsson, J.

¨

O.G., Davíðsd

´

ottir, B., 2016. Classification and valuation of soil ecosystem

services. Agric. Syst. 145, 24

–

38.

Kaufman, L., Rousseeuw, P.J., 1990. Finding Groups in Data.

https://doi.org/10.1002/

9780470316801

https://doi.org/10.1016/j.

Khaledian, Y., Miller, B.A., 2020. Selecting appropriate machine learning methods for

digital soil mapping. Appl. Math. Model. 81, 401

–

418.

apm.2019.12.016

https://doi.org/10.1016/j.geoderma.2014.10.019

Knox, N.M., Grunwald, S., McDowell, M.L., Bruland, G.L., Myers, D.B., Harris, W.G.,

2015. Modelling soil carbon fractions with VNIR and MIR spectroscopy. Geoderma

239-240, 229

–

239.

https://doi.org/10.1007/978-3-642-

Kohonen, T., 2001. Self-organizing maps, 30, 502.

56927-2

https://doi.org/10.18637/jss.v036.i11

Kursa, M.B., Rudnicki, W.R., 2010. Feature selection with the Boruta package. J. Stat.

Softw. 36 (11), 1

–

13.

.

Digital soil mapping: an introductory perspective. In: Lagacherie, P., McBratney, A.B.,

Voltz, M. (Eds.), 2006. Developments in Soil Science, 1st ed. Elsevier, Amsterdam;

Boston, ISBN 978-0-444-52958-9

.

Lal, R., 2014. Soil conservation and ecosystem services. International Soil and Water

Conservation Research 2 (3), 36

–

47.

https://doi.org/10.1016/S2095-6339(15)

30021-6

https://doi.org/10.1111/j.1461-

Lele, S.R., Dennis, B., Lutscher, F., 2007. Data cloning: easy maximum likelihood

estimation for complex ecological models using Bayesian Markov chain Monte Carlo

methods. Ecol. Lett. 10, 551

–

563.

0248.2007.01047.x

.

Levin, N., Ben-Dor, E., Karnieli, A., 2004. Topographic information of sand dunes as

extracted from shading effects using Landsat images. Remote Sens. Environ. 90,

190

–

209.

https://doi.org/10.1016/j.rse.2003.12.008

Liu, F., Rossiter, D.G., Zhang, G.L., Li, D.C., 2020. A soil colour map of China. Geoderma

379, 114556

.

Ma, Y., Minasny, B., Malone, B.P., Mcbratney, A.B., 2019. Pedology and digital soil

mapping (DSM). Eur. J. Soil Sci. 70 (2), 216

–

235.

https://doi.org/10.1111/

ejss.12790

.

McBratney, A.B., Mendonça Santos, M.L., Minasny, B., 2003. On digital soil mapping.

Geoderma 117, 3

–

52.

https://doi.org/10.1016/S0016-7061(03)00223-4

.

McBratney, A.B., Minasny, B., Stockmann, U. (Eds.), 2018. Pedometrics, 1st ed. vol. XIII.

Springer, p. 720.

https://doi.org/10.1007/978-3-319-63439-5

https://doi.org/10.1038/s41598-023-36219-9

Mello, F.A.O., Dematt

ˆ

e, J.A.M., Bellinaso, H., et al., 2023. Remote sensing imagery

detects hydromorphic soils hidden under agriculture system. Sci. Rep. 13, 10897.

Naimi, S., Ayoubi, S., Zeraatpisheh, M., Dematte, J.A.M., 2021. Ground observations and

environmental covariates integration for mapping of soil salinity: a machine

learning-based approach. Remote Sens. 13, 4825.

https://doi.org/10.3390/

rs13234825

.

Negasa, T., Ketema, H., Legesse, A., Sisay, M., Temesgen, H., 2017. Variation in soil

properties under different land use types managed by smallholder farmers along the

toposequence in southern Ethiopia. Geoderma 290, 40

–

50.

https://doi.org/10.1016/

j.geoderma.2016.11.021

Ng, W., Minasny, B., McBratney, A., de Caritat, P., Wilford, J., 2023. Digital soil mapping

of lithium in Australia. Earth System Science Data Discussions 2023, 1

–

25.

doi.org/10.5194/essd-15-2465-2023

Nocita, M., Stevens, A., van Wesemael, B., Aitkenhead, M., Bachmann, M., Barth

`

es, B.,

Ben-Dor, E., Brown, D.J., Clairotte, M., Csorba, A., Dardenne, P., Dematt

ˆ

e, J.A.M.,

Genot, V., Guerrero, C., Knadel, M., Montanarella, L., Noon, C., Ramirez-Lopez, L.,

Robertson, J., Sakai, H., Wetterlind, J., 2015. Soil spectroscopy: an alternative to wet

chemistry for soil monitoring. Adv. Agron. 132, 139

–

159.

https://doi.org/10.1016/

bs.agron.2015.02.002

.

Novais, J.J.M., Melo, B.M.D., Junior, A.F.N., Lima, R.H.C., Souza, R.E., Melo, V.F.,

Amaral, E.F., Tziolas, N., Dematt

ˆ

e, J.A.M., 2025. Online analysis of Amazon

’

s soils

through reflectance spectroscopy and cloud computing can support policies and the

sustainable development. J. Environ. Manage. 375, 124155.

10.1016/j.jenvman.2025.124155

.

Ortiz-Bobea, A., Ault, T.R., Carrillo, C.M., Chambers, R.G., Lobell, D.B., 2021.

Anthropogenic climate change has slowed global agricultural productivity growth.

Nat. Clim. Chang. 11 (4), 306

–

312.

https://doi.org/10.1038/s41558-021-01000-1

.

Padarian, J., Minasny, B., McBratney, A.B., 2019. Using deep learning for digital soil

mapping. Soil 5 (1), 79

–

89.

https://doi.org/10.5194/soil-5-79-2019

.

Padarian, J., Stockmann, U., Minasny, B., McBratney, A.B., 2022. Monitoring changes in

global soil organic carbon stocks from space. Remote Sens. Environ. 281, 113260.

https://doi.org/10.1016/j.rse.2022.113260

.

Paiva, A.F.S., Poppiel, R.R., Rosin, N.A., Greschuk, L.T., Rosas, J.T.F., Dematt

ˆ

e, J.A.M.,

2022. The Brazilian Program of soil analysis via spectroscopy (ProBASE): combining

spectroscopy and wet laboratories to understand new technologies. Geoderma 421

(1), 115905.

https://doi.org/10.1016/j.geoderma.2022.115905

https://doi.org/10.5194/soil-7-

Poggio, L., De Sousa, L.M., Batjes, N.H., Heuvelink, G.B., Kempen, B., Ribeiro, E.,

Rossiter, D., 2021. SoilGrids 2.0: producing soil information for the globe with

quantified spatial uncertainty. Soil 7 (1), 217

–

240.

217-2021

.

Poppiel, R.R., Lacerda, M.P.C., Rizzo, R., Safanelli, J.L., Bonfatti, B.R., Silvero, N.E.Q.,

Dematt

ˆ

e, J.A.M., 2020. Soil color and mineralogy mapping using proximal and

remote sensing in Midwest Brazil. Remote Sens. 12 (7), 1197.

https://doi.org/10.3390/rs15123070

10.3390/rs12071197

.

Richer-de-Forges, A.C., Chen, Q., Baghdadi, N., Chen, S., Gomez, C., Jacquemoud, S.,

Martelet, G., Mulder, V.L., Urbina-Salazar, D., Vaudour, E., Weiss, M., Wigneron, J.-

J.A.M. Dematt

ˆ

e et al.

Science of the Total Environment 968 (2025) 178791

14

P., Arrouays, D., 2023. Remote sensing data for digital soil mapping in French

research

—

a review. Remote Sens. 15, 3070.

Rizzo, R., Wadoux, A.M.J.-C., Dematt

ˆ

e, J.A.M., Minasny, B., Barr

´

on, V., Ben-Dor, E.,

Francos, N., Savin, I., Poppiel, R., Silvero, N.E.Q., Terra, F. da S., Rosin, N.A.,

Rosas, J.T.F., Greschuk, L.T., Ballester, M.V.R., Rico G

´

omez, A.M., Belllinaso, H.,

Safanelli, J.L., Chabrillat, S., Fiorio, P.R., Salama, E.S.M., 2023. Remote sensing of

the Earth

’

s soil color in space and time. Remote Sens. Environ. 299, 113845.

doi.org/10.1016/j.rse.2023.113845

.

Rogge, D., Bauer, A., Zeidler, J., Mueller, A., Esch, T., Heiden, U., 2018. Building an

exposed soil composite processor (SCMaP) for mapping spatial and temporal

characteristics of soils with Landsat imagery (1984

–

2014). Remote Sens. Environ.

205, 1

–

17.

https://doi.org/10.1016/j.rse.2017.11.004

.

Rosin, N.A., Dematt

ˆ

e, J.A.M., Poppiel, R.R., Silvero, N.E.Q., Rodriguez-Albarracin, H.S.,

Rosas, J.T.F., Greschuk, L.T., Bellinaso, H., Minasny, B., Gomez, C., Marques

Júnior, J., Fernandes, K., 2023. Mapping Brazilian soil mineralogy using proximal

and remote sensing data. Geoderma 432, 116413.

https://doi.org/10.1016/j.

geoderma.2023.116413

.

Roy, D.P., Kovalskyy, V., Zhang, H.K., Vermote, E.F., Yan, L., Kumar, S.S., Egorov, A.,

2016. Characterization of Landsat-7 to Landsat-8 reflective wavelength and

normalized difference vegetation index continuity. Remote Sens. Environ. 185,

57

–

70.

https://doi.org/10.1016/j.rse.2015.12.024

.

Safanelli, J.L., Chabrillat, S., Ben-Dor, E., Dematt

ˆ

e, J.A.M., 2020. Multispectral models

from bare soil composites for mapping topsoil properties over Europe. Remote Sens.

12, 1369.

https://doi.org/10.3390/rs12091369

Safanelli, J.L., et al., 2021. Fine-scale soil mapping with Earth Observation data: a

multiple geographic level comparison. Rev. Bras. Ci

ˆ

enc. Solo 45, e0210080.

doi.org/10.36783/18069657rbcs20210080

Sayre, R., Karagulle, D., Frye, C., Boucher, T., Wolff, N.H., Breyer, S., Wright, D.,

Martin, M., Butler, K., Van Graafeiland, K., Touval, J., Sotomayor, L., McGowan, J.,

Game, E.T., Possingham, H., 2020. An assessment of the representation of

ecosystems in global protected areas using new maps of World Climate Regions and

World Ecosystems. Global Ecology and Conservation 21, e00860.

10.1016/j.gecco.2019.e00860

.

Sothe, C., Gonsamo, A., Arabian, J., Kurz, W.A., Finkelstein, S.A., Snider, J., 2022. Large

soil carbon storage in terrestrial ecosystems of Canada. Glob. Biogeochem. Cycles 36,

e2021GB007213.

https://doi.org/10.1029/2021GB007213

https://doi.org/10.1007/978-

Stenberg, B., Viscarra-Rossel, R.A., 2010. Diffuse reflectance spectroscopy for high-

resolution soil sensing. Proximal Soil Sensing 29

–

47.

90-481-8859-8_3

sssaj1981.03615995004500060031x

Stoner, E.R., Baumgardner, M.F., 1981. Characteristic variations in reflectance of surface

soils. Soil Sci. Soc. Am. J. 45, 1161

–

1165.

https://doi.org/10.2136/

https://doi.org/10.3390/rs12091389

Tziolas, N., Tsakiridis, N., Ben-Dor, E., Theocharis, J., Zalidis, G., 2020a. Employing a

multi-input deep convolutional neural network to derive soil clay content from a

synergy of multi-temporal optical and radar imagery data. Remote Sens. 12, 1389.

https://doi.org/10.1016/j.rse.2020.111793

Tziolas, N., Tsakiridis, N., Ogen, Y., Kalopesa, E., Ben-Dor, E., Theocharis, J., Zalidis, G.,

2020b. An integrated methodology using open soil spectral libraries and Earth

Observation data for soil organic carbon estimations in support of soil-related SDGs.

Remote Sens. Environ. 244, 111793.

https://d9-wret.s3.us-west-2.amazonaws.com/ass

U.S. Geological Survey, 2020a. Landsat 4-7 Collection 1 (C1) Surface Reflectance

(LEDAPS) product guide. URL.

ets/palladium/production/s3fs-public/atoms/files/LSDS-1370_L4-7_C1-Surfac

eReflectance-LEDAPS_ProductGuide-v3.pdf

(WWW Document) (accessed September

1, 2021).

U.S. Geological Survey, 2020b. Landsat 8 Collection 1 (C1) Land Surface Reflectance

Code (LaSRC) product guide URL

https://d9-wret.s3.us-west-2.amazonaws.co

m/assets/palladium/production/s3fspublic/atoms/files/LSDS-1368_L8_C1-LandSur

faceReflectanceCode-LASRC_ProductGuide-v3.pdf

(WWW Document) (accessed

September 1, 2021).

Urbina-Salazar, D., Vaudour, E., Richer-de-Forges, A.C., Chen, S., Martelet, G.,

Baghdadi, N., Arrouays, D., 2023. Sentinel-2 and Sentinel-1 bare soil temporal

mosaics of 6-year periods for soil organic carbon content mapping in Central France.

Remote Sens. 15, 2410.

https://doi.org/10.3390/rs15092410

Vaudour, E., Gholizadeh, A., Castaldi, F., Saberioon, M., Bor

ů

vka, L., Urbina-Salazar, D.,

Fouad, Y., Arrouays, D., Richer-de-Forges, A.C., Biney, J., Wetterlind, J., Van

Wesemael, B., 2022. Satellite imagery to map topsoil organic carbon content over

cultivated areas: an overview. Remote Sens. 14, 2917.

https://doi.org/10.3390/

rs14122917

.

Viscarra Rossel, R.A., Chen, C., Grundy, M.J., Searle, R., Clifford, D., Campbell, P.H.,

2015. The Australian three-dimensional soil grid: Australia

’

s contribution to the

GlobalSoilMap project. Soil Research 53 (8), 845

–

864.

https://doi.org/10.1071/

SR14366

Viscarra Rossel, R.A., Behrens, T., Ben-Dor, E., Brown, D.J., Dematt

ˆ

e, J.A.M.,

Shepherd, K.D., Shi, Z., Stenberg, B., Stevens, A., Adamchuk, V., Aïchi, H.,

Barth

`

es, B.G., Bartholomeus, H.M., Bayer, A.D., Bernoux, M., B

¨

ottcher, K.,

Brodský, L., Du, C.W., Chappell, A., Fouad, Y., Ji, W., 2016. A global spectral library

to characterize the world

’

s soil. Earth Sci. Rev. 155, 198

–

230.

10.1016/j.earscirev.2016.01.012

.

Wadoux, A.M.C., Minasny, B., McBratney, A.B., 2020. Machine learning for digital soil

mapping: applications, challenges and suggested solutions. Earth Sci. Rev. 210,

103359

.

Wang, F., Yang, S., Wei, Y., Shi, Q., Ding, J., 2021. Characterizing soil salinity at multiple

depth using electromagnetic induction and remote sensing data with random forests:

a case study in Tarim River Basin of southern Xinjiang. China. Science of The Total

Environment 754, 142030.

https://doi.org/10.1016/j.scitotenv.2020.142030

.

Wesemael, B.V., Abdelbaki, A., Ben-Dor, E., Chabrillat, S., d

’

Angelo, P., Dematt

ˆ

e, J.A.M.,

Genova, G., Gholizadeh, A., Heiden, U., Karlshoefer, P., Milewski, R., Poggio, L.,

Sabetizade, M., Sanz, A., Schwind, P., Tsakiridis, N., Tziolas, N., Yagüe, J.,

ˇ

Zí

ˇ

zala, D.,

2024. Geoderma 452, 117113.

https://doi.org/10.1016/j.geoderma.2024.117113

.

Zappia, L., Oshlack, A., 2018. Clustering trees: a visualization for evaluating clusterings

at multiple resolutions. Gigascience 7, 1

–

9.

https://doi.org/10.1093/gigascience/

giy083