Content uploaded by Antonio J. Castro
All content in this area was uploaded by Antonio J. Castro on Oct 13, 2022
Content may be subject to copyright.
Mapping socio-ecological systems in Idaho:
Spatial patterns and analytical considerations
| Susan Parsons
| Antonio J. Castro
| Kathleen A. Lohse
Biological Sciences, Idaho State
University, Pocatello, Idaho, USA
Biology and Geology Department,
Andalusian Center for the Assessment
and Monitoring of Global Change
(CAESCG), University of Almeria,
Division of Earth Sciences, Grant/Award
Number: EAR-1331872; Office of
Experimental Program to Stimulate
Competitive Research, Grant/Award
Handling Editor: Laura L
Policy interest in socio-ecological systems has driven attempts to define and
map socio-ecological zones (SEZs), that is, spatial regions, distinguishable by
their conjoined social and bio-geo-physical characteristics. The state of Idaho,
USA, has a strong need for SEZ designations because of potential conflicts between
rapidly increasing and impactful human populations, and proximal natural ecosys-
by: (1) considering potential biases of clustering methods, (2) cross-validating SEZ
classifications, (3) measuring the relative importance of bio-geo-physical and social
system predictors, and (4) considering spatial autocorrelation. We obtained authori-
tative bio-geo-physical and social system datasets for Idaho, aggregated into 5-km
grids =25 km
, and decomposed these using principal components analyses
(PCAs). PCA scores were classified using two clustering techniques commonly used
in SEZ mapping: hierarchical clustering with Ward’s linkage, and k-means analysis.
Classification evaluators indicated that eight- and five-cluster solutions were optimal
for the bio-geo-physical and social datasets for Ward’slinkage,resultingin31SEZ
composite types, and six- and five-cluster solutions were optimal for k-means analy-
sis, resulting in 24 SEZ composite types. Ward’sandk-means solutions were similar
for bio-geo-physical and social classifications with similar numbers of clusters.
Further, both classifiers identified the same dominant SEZ composites. For rarer
SEZs, however, classification methods strongly affected SEZ classifications, poten-
tially altering land management perspectives. Our SEZs identify several critical
regions of social–ecological overlap. These include suburban interface types and a
high desert transition zone. Based on multinomial generalized linear models,
bio-geo-physical information explained more variation in SEZs than social system
data, after controlling for spatial autocorrelation, under both Ward’sandk-means
approaches. Agreement (cross-validation) levels were high for multinomial models
with bio-geo-physical and social predictors for both Ward’sandk-means SEZs. A
consideration of historical drivers, including indigenous social systems, and current
trajectories of land and resource management in Idaho, indicates a strong need for
rigorous SEZ designations to guide development and conservation in the region.
Received: 28 April 2022 Accepted: 13 May 2022
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided
the original work is properly cited.
© 2022 The Authors. Ecosphere published by Wiley Periodicals LLC on behalf of The Ecological Society of America.
Ecosphere. 2022;13:e4242. https://onlinelibrary.wiley.com/r/ecs2 1of21
Our analytical framework can be broadly applied in SES research and applied in
other regions, when categorical responses—including cluster designations—have a
American West, multivariate ecological data, cluster analysis, Moran eigenvector map,
multinomial GLM, land management, spatial analysis
Humans are a part of ecosystems and shape them from
local to global scales (Berkes & Folke, 1998;
Herrero-Jauregui et al., 2019; Liu et al., 2007;
Ostrom, 2009). This reality has driven the development of
socio-ecological system (SES) theory, which considers
topics at the intersection of human societies and natural
(non-anthropogenic) ecosystem components (Bailey, 2009;
Cherkasskii, 1988;Redmanetal.,2004). A critical topic of
interest in SES research is the delineation and mapping of
SES boundaries that delimit “socio-ecological zones”
(SEZs) to aid in the identification and management of
potentially vulnerable ecosystems (Martín-L
et al., 2017;Winter&Lucas,2017) and their ecosystem
services (Bourne et al., 2016; Maes et al., 2012;
Quintas-Soriano et al., 2018). Here, we define a SEZ to be
a spatial region, distinguishable by its conjoined social
(e.g., socioeconomic, management, land use) and
bio-geo-physical characteristics (e.g., natural communities,
net primary productivity, topography, hydrology).
Despite recent efforts, there is no consensus on how to
analytically delineate and define SES boundaries, or zones
with consistent social and ecological characteristics, orga-
nized by a common set of system variables, and defined by
diverse sets of socio-ecological dynamics (Jones
et al., 2019; Quintas-Soriano et al., 2022). Current SEZ
mapping efforts (e.g., Cruz-Cardenas et al., 2017;Dressel
et al., 2018;Hamannetal.,2015; Hanspach et al., 2016;
Kok et al., 2016;Martín-L
opez et al., 2017;Vallejos
et al., 2019; Zhang et al., 2011) have used a similar
three-step analytical approach (Appendix S1: Figure S1).
Briefly, these are as follows: (1) collection of spatially
referenced social and bio-geo-physical data, followed gen-
erally by application of dimension-reduction methods
(e.g., principal components analysis [PCA]); (2) cluster
analyses to separately identify social and ecological system
types; and (3) synthesis of social and ecological classifica-
tion types into conjoined SEZs, based on spatial overlap.
While providing potentially useful information, this
approach has at least four shortcomings. First, little thought
has been given to the possible predispositions of certain ana-
lytical methods (e.g., particular clustering algorithms, etc.) in
the determination of the final SEZ types (cf. Aho
et al., 2008). Second, the effectiveness of SEZs in explaining
variation in bio-geo-physical and social data, or serving as
a predictive structure for new data is generally unconsidered.
Third, the relative contributions of bio-geo-physical and
social information to SEZ designations has been
unmeasured. Fourth, explicit spatial effects (e.g., spatial auto-
correlation structures) have been ignored. In this article, we
consider these and other analytical issues, and explore poten-
tial solutions, while developing an SEZ map for the state of
Idaho, in the northwestern region of the United States.
Idaho constitutes a large (216,630 km
inland area in the northwestern United States (Figure 1).
The state is an excellent candidate for SEZ mapping
because of its strong bio-geo-physical gradients, and
diverse land uses and management designations. Over
61% of Idaho’s land area is federally owned or adminis-
tered, resulting in varying levels of protection from
private development and potential ecosystem degradation
(Quintas-Soriano et al., 2021). Idaho, however, is economi-
cally reliant on natural resource acquirement and
utilization (>5% 2019 Idaho gross domestic product [GDP];
United States Bureau of Economic Analysis, 2021), and is
one of the most productive agricultural states, contributing
nearly one-third of the potatoes grown in the United States
(Idaho State Department of Agriculture, 2021). Idaho is
also among the fastest growing US states (United States
Census Bureau, 2021), and its capitol, Boise, is the fastest
growing major American metropolitan area, with a 120%
population increase between 1990 and 2015 (Forbes, 2018).
Boise is projected to continue growing due to its relatively
low cost of living, high job growth rates, and quality of life
(Narducci et al., 2019; Quintas-Soriano et al., 2022).
Idaho’s cultivated cropland and rangeland occur pri-
marily at lower elevations in the southern portion of the
state, including regions along the Snake River plain
(Quintas-Soriano et al., 2018). The northern Idaho
2of21 AHO ET AL.
“panhandle”has a maritime climate and large regions of
dense evergreen forests. Central Idaho is dominated by
rugged mountains and contains the Idaho batholith, a
granitic and granodioritic formation encompassing
approximately 25,000 km
. Northern and central Idaho
contain vast tracts of public land, and including, the
largest designated wilderness area in the lower 48 states,
the Frank Church-River of No Return Wilderness
; United States Forest Service, 2021). Idaho
agricultural economies are likely to be negatively affected
by a number of regional trends including increasing
annual temperatures (Tang et al., 2012), water
scarcity (Castro et al., 2018), invasive plant species
(Shwiff et al., 2018), and reduced steppe fire return inter-
vals (Balch et al., 2013). Further, mountain ecosystems
are expected to experience strong negative effects from
climate change, including loss of biodiversity (Klos
et al., 2014; Kullman, 2004).
The conflation of climate change and rapid popula-
tion growth will likely result in increased vulnerability of
many regions in Idaho to ecosystem degradation, includ-
ing water overexploitation (Castro et al., 2018; Cornell
et al., 2019; Quintas-Soriano et al., 2022). We hope that
SEZ mapping will allow tracking of the Idaho ecological
and socioeconomic landscapes as these changes occur.
This information could then be used to develop effective
strategies for Idaho resource management, including
policies for responsible growth and development
(Alessa et al., 2008; Castellarini et al., 2014; Díaz
et al., 2015; Hamann et al., 2015; Hanspach et al., 2016;
opez et al., 2017; Quintas-Soriano et al., 2018;
Sinare et al., 2016).
Data preparation and dimension reduction
Publicly available authoritative bio-geo-physical and
social system datasets were acquired to describe the state
of Idaho at a scale of 5-km
grids. Bio-geo-physical vari-
ables included those describing climate (i.e., average tem-
perature, annual precipitation, average dew temperature,
and average vapor pressure deficit), topography
(i.e., slope, elevation, and lithology), resident flora and
fauna (i.e., biodiversity and land cover), vegetation pro-
ductivity (normalized difference vegetation index and
potential evapotranspiration), and surficial geology
(i.e., lithology and soil order) (Table 1). In parallel, social
variables described basic demographic information
(i.e., population, race, age, income, education, marital
status, and home ownership), human landscape modifi-
cation (i.e., housing density, traffic density, and human
modification index; Theobald, 2013), and economy
(i.e., land use, industry occupation, and surface manage-
ment agency) (Table 2). Each individual grid was
FIGURE 1 Orientation map of Idaho. Shading in inset based on 10-m digital elevation map. Darker colors indicate lower elevations.
assigned a unique ID, resulting in a data matrix, with
rows representing 8796 grid units and columns
representing bio-geo-physical and social variables.
We ran two PCAs in a nested fashion, for both
bio-geo-physical and social system datasets, to obtain use-
ful variables for subsequent cluster analyses. Initial
covariance matrix-based PCAs were run on sets of vari-
ables that were proportional summaries of related cate-
gorical levels. For instance, the “Industry/Occupation”
dataset in Table 2provides 5-km grid population summa-
ries for 17 industry/occupation types. Retained dimen-
sions in initial PCAs were required to explain >70% of
variation in raw data (Table 3).
We created final PCAs—one each for social and
bio-geo-physical datasets—that considered both initial
PCA distillations of categorical variables and raw quanti-
tative data. In total, 18 variables were analyzed in the
final bio-geo-physical PCA, and 22 variables were consid-
ered in the final socioeconomic PCA.
We controlled for potential latitudinal and longitu-
locations as variables in the PCA analyses, but when
calculating the matrix of PCA scores, removing latitu-
dinal and longitudinal loadings from both the scaled
npdata matrix, X,andtheppmatrix of variable
loadings, E, prior to obtaining scores using the
inner product, XE (Legendre & Legendre, 2012). Thus,
we held spatial locations constant while using
location-adjusted loadings from other variables to gen-
erate final PCA principal components. Correlation
matrices were used to generate final PCAs to account
for the highly variable scales of measurement. To gen-
erate final datasets for subsequent cluster analysis,
principal components were retained in the final PCAs
until at least 90% of the variation in the analyzed
datasets was described. This resulted in 11 dimensions
being retained in bio-geo-physical PCA (94.2% varia-
tion explained) and 14 dimensions being retained in
the social system PCA (92.3% variation explained)
(Appendix S1: Figures S1–S9). Eigenvector loadings for
all predictors in initial PCAs and the two final PCAs
are given in Appendix S1: Figures S3–S8.
TABLE 1 Raw bio-geo-physical variables used in the analyses.
Variable Description Source Resolution
Elevation (m) Digital elevation model (DEM) NED 10 m
Slope (0–90) Obtained from DEM NED 10 m
Dew temperature (C) 30-year average annual mean PRISM 1981–2010 1 km
Temperature (C) 30-year average annual mean PRISM 1981–2010 1km
Vapor pressure deficit (hectopascals) 30-year average annual median PRISM 1981–2010 1km
Precipitation (average mm/month) 30-year average annual mean PRISM 1981–2010 1km
10-year average annual mean MODIS 2000–2010 1km
NDVI NDVI =(actual NDVI 200) +50 MODIS 2001–2006 250 m
Land cover types NLCD 2011 30 m
Basic rock types MRData Variable vector
Soil order SSURGO (2022),
Organismal species counts BiodiversityMapping.org,
Jenkins et al. (2013),
Pimm et al. (2014)
Note: Footnoted entries indicate datasets that were proportional summaries of related categorical levels, which were analyzed in initial (nested) principal
components analyses. See Appendix S2 for lithology and soil order summaries.
Abbreviations: MODIS, Moderate Resolution Imaging Spectroradiometer (MODIS, 2022); MRdata, Mineral Resources Online Spatial Data (MRData, 2022);
NDVI, normalized difference vegetation index; NED, National Elevational Dataset (NED, 2018); NLCD, National Land Cover Database (NLCD, 2022); PRISM,
Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM, 2015); SSURGO, Soil Survey Geographic Database (SSURGO, 2022); STATSGO2,
Digital General Soil Map of the United States (STATSGO2, 2022).
Herbaceous wetlands, woody wetland, deciduous forest, evergreen forest, mixed forest, shrub and scrub, grassland, open water, and ice and snow.
Sedimentary, igneous, metamorphic, and unconsolidated.
Alfisol, andisol, aridisol, entisol, inceptisol, mollisol, and vertisol.
Amphibians, fish, birds, mammals, reptiles, and trees.
4of21 AHO ET AL.
TABLE 2 Raw social system datasets used in the analyses.
Variable Description Source Resolution
Population size ACS 2008–2012, average US Census Bureau Census Block Groups
Median age ACS 2008–2012, average US Census Bureau Census Block Groups
High school degrees ACS 2008–2012, percentage US Census Bureau Census Block Groups
Bachelor degrees ACS 2008–2012, percentage US Census Bureau Census Block Groups
Aggregate income/household ACS 2008–2012, average US Census Bureau Census Block Groups
Housing units ACS 2008–2012, average US Census Bureau Census Block Groups
Owner occupied homes/ha ACS 2008–2012, average US Census Bureau Census Block Groups
Married with children/ha ACS 2008–2012, percentage US Census Bureau Census Block Groups
Divorced/ha ACS 2008–2012, percentage US Census Bureau Census Block Groups
Traffic density 2011 Daily average ITD Polyline, then 1 km
Human modification index Index of 0–1 for North America Kennedy et al. (2019) 230 m
ACS 2008–2012, percentage US Census Bureau Census Block Groups
Stewardship agency or ownership ISMA 2002 variable vector
Land use types NLCD 30 m
ACS 2008–2012, Average US Census Bureau Census Block Groups
Note: Footnoted entries indicate datasets that were proportional summaries of related categorical levels and which were analyzed in initial (nested) principal
components analyses (Table 3).
Abbreviations: ACS, American Community Survey (ACS, 2022); ISMA, Idaho Surface Management Agency (ISMA, 2022); ITD, Idaho Transportation
Department (ITD, 2022); NLCD, National Land Cover Database (NLCD, 2022).
ManagementBusinessFinancial, ComputerEngineeringScience, EducationLegalArtsMedia, HealthcareTechnical, HealtcareSupport, ProtectiveService,
FoodPrepServing, CleaningMaintenance, PersonalCareService, Sales, AdministrativeSupport, FarmingFishingForestry, ConstructionExtraction,
InstallMaintainRepair, Production, Transportation, and MaterialMoving.
BLM, USFS, PRIVATE, STATE, NPS, BOR, DOE, IR =Indigenous Reservation, and OTHER.
Cultivated, pasture and hay, high intensity development, low intensity development, medium intensity development, designated open space.
Asian, Black, Hawaiian/Pacific, Latin, Mixed race, Indigenous, White, and Other.
TABLE 3 Principal components retained in initial principal components analyses (PCAs) (based on proportional summaries of
categorical data) and final bio-geo-physical and social system PCAs (that used both scores from initial PCAs and quantitative data).
PCA model No. original variables Dimensions retained
Percentage of variance explained in
Biodiversity 6 3 80.6
Land cover 9 2 89.0
Soil order 7 3 87.1
Lithology 4 2 80.8
Final PCA 18 11 94.2
Industry occupation 17 7 70.5
Management agency 9 2 87.3
Land use 6 1 93.7
Race/ethnicity 8 2 98.1
Final PCA 22 14 92.3
We classified scores of the final PCAs of
bio-geo-physical and social system datasets using
both agglomerative hierarchical classification with
Ward’s linkage (Ward, 1963)andk-means analysis
(MacQueen, 1967). We chose these classifiers because
they are the most frequently used clustering methods
in SES mapping efforts (Cervantes-Jiménez et al., 2017;
Hamann et al., 2015;Koketal.,2016; Jones et al.,
opez et al., 2017; Zhang et al., 2011),
and because they are frequently applied in the cluster-
ing of PCA scores (Gardner, 1991;Mimuraetal.,2013).
Preceding classification, scores from final PCAs were
weighted by multiplying principal component’sscores
by their corresponding eigenvalues. The algorithm of
Hartigan and Wong (1979)wasusedink-means analy-
sis (Hartigan & Wong, 1979).
To determine an optimum number of clusters gener-
ated by classifiers, we applied three classification eval-
uators to the 24 simplest classification solutions (2–25
clusters) of the bio-geo-physical and social system
synthetic (PCA score) datasets. The classification eval-
uators were as follows: (1) the within-to-between ratio
(McClain & Rao, 1975); (2) Dunn’s index (Dunn, 1974);
and (3) the Calinski–Harabasz index (Calinski &
Harabasz, 1974). In order to examine the correspon-
dence/incongruence of classification solutions from
Ward’sandk-means analysis, we calculated the per-
cent agreement of grid assignments to clusters using
contingency tables, for both the bio-geo-physical and
socioeconomic clustering solutions. Percentage agree-
ment and permutational significance tests were
obtained using the function best agreement from the
R-package asbio (Aho, 2020). To address computa-
tional constraints imposed by the large grid matrix,
only 100 iterations were used in permutation testing.
Socio-ecological zone designation
Because a common 5 km grid system was used for clus-
ter analyses of both the bio-geo-physical and socioeco-
nomic datasets, we were able to efficiently overlay
corresponding maps of the bio-geo-physical and social
system cluster regions. This resulted in SEZ compos-
ites, with certain zones potentially implying manage-
ment concerns or higher levels of sustainability than
others (Winter & Lucas, 2017).
Quantifying spatial autocorrelation—
Moran eigenvector maps
To define potential spatial autocorrelation patterns
and effects at multiple spatial scales, we generated
distance-based Moran eigenvectorsmaps(dbMEMs)
(Griffith, 1996) from the spatial coordinates of grid centroids.
For the required truncation threshold value, we used the
length of the largest edge from a minimum spanning span
of the grid-centroid-based Euclidean distance matrix (see
Legendre & Legendre, 2012, p. 863). Other methods for the
generation of dbMEMs are described by Dray et al. (2018).
By definition, we obtained 8796 distance-based MEM
eigenvectors, reflecting the number of unique grids.
However, only the first 4392 dbMEMs had positive eigen-
values (Figure 2). Following convention (Legendre &
Legendre, 2012), we limited our spatial autocorrelation
considerations to dbMEMs with positive eigenvalues.
Loadings for the first 25 dbMEMs are overlain on
our Idaho grid system in Figure 2. Darker colors indicate
larger positive eigenanalysis scores for grids. Note that
lower dimensional dbMEMs represent broadscale auto-
correlation patterns, whereas higher dimensions repre-
sent autocorrelation structures at increasingly finer
spatial scales (Borcard et al., 2004).
Partitioning effects of predictors on SEZs
To quantify the explanatory contributions of
bio-geo-physical and social system predictors to the gener-
ation of SEZs, while controlling for spatial autocorrelation,
we used multinomial generalized linear models (GLMs)
(Venables & Ripley, 2002). Specifically, we fit SEZ types as
a simultaneous function of final bio-geo-physical PCA
scores for principal components 1–11 (see above), final
social system PCA scores for principal components 1–14
(see above), and dbMEMs (representing multiscale spatial
autocorrelation structures). Only the first 25 dbMEMs
(Figure 2) had appreciable levels of association with
SEZ composite classes under either Ward’s linkage
(Appendix S1: Figure S19) or k-means (Appendix S1:
Figure S20) classifiers. Thus, for computational efficiency,
only these dbMEMs were used to represent spatial auto-
correlation in multinomial models. Under this approach,
the full (saturated) multinomial model had 49 predictors,
representing bio-geo-physical (11 predictors), social (14 pre-
dictors), and spatial processes (25 predictors). Because they
each resulted from eigenanalyses, model predictors (PCA
scores) representing these individual processes were statis-
To quantify multinomial model efficacy, we calcu-
lated Efron’s pseudo-R
(Efron, 1978), adjusted for the
6of21 AHO ET AL.
number of parameters in GLMs, using an extension pro-
posed by Zhang (2017):
is the residual deviance (minus two times
the model log-likelihood) and dev
is the null deviance,
that is, the deviance in an intercept-only model. We mea-
sured the partial explanatory power of bio-geo-physical
and social predictors when controlling for the effect of
the other predictor, and/or controlling for spatial auto-
correlation (dbMEMs), using a surrogate of the partial R
statistic based on deviance:
represents the deviance-based partial
indicates the adjusted pseudo-R
(Equation 1) for a reference model that includes the
predictor(s) of interest, and R
of a nested model that excludes
only the predictor(s) of interest. Equation 2can be justi-
fied on the grounds that residual deviance is a general-
ized form of model error (equivalent to the residual of
squares [SSE] in conventional general linear models)
(Aho, 2013), and substitution of SSE for residual deviance
in Equation 2results in the conventional formulation for
We validated our SEZ composite classification by
comparing multinomial predicted grid SEZs for differ-
ent subsets of predictors to corresponding mapped
Ward’sandk-means SEZs. Agreement was measured
using percent agreement, the kappa statistic ( b
Congalton & Mead, 1983), and user and producer
FIGURE 2 The first 25 distance-based Moran eigenvector maps (dbMEMs) for analysis of Idaho 5-km grids. Darker colors indicate
larger positive loadings in Moran eigenvectors.
accuracy for individual SEZs. For the latter three mea-
sures, the mapped SEZs were used as the reference
For all GIS applications, including the generation of
maps with SEZ composite overlays, we used ESRI
ArcGIS for Desktop 10.8 (https://www.esri.com/en-us/
home). The R statistical environment (R Core
Team, 2020) was used for most statistical analyses. The
function princomp from the stats package was used to
generate PCAs. The functions hclust and k-means
from the stats package were used to generate Ward’s
linkage and k-means classifications, respectively.
The package fpc (Hennig, 2020) was used to imple-
ment classification evaluators. The package nnet
(Venables & Ripley, 2002) was used to create multino-
mial models. Functions from the package asbio
(Aho, 2020) were used to measure and test agreement
between Ward’sandk-means classifications and to
Kand other error matrix statistics for SEZ-type
We attempted to identify “optimal”classification solu-
tions with an intermediate (>2) number of clusters,
whose efficacy was supported by several evaluators
(cf. Aho et al., 2008). Classification evaluator results for
Ward’s linkage are shown in Figure 3. We chose eight-
and five-cluster solutions as relatively effective prunings
of the Ward’s linkage classifications of bio-geo-physical
and social datasets, respectively (Figure 3), and chose six-
and five-cluster solutions for k-means analyses of
bio-geo-physical and social system datasets, respectively
(Appendix S1: Figure S13).
We named bio-geo-physical types based on their gen-
eral topography and physiognomy (Appendices S2 and S4).
Ward’s bio-geo-physical types were as follows: (1) shrub
FIGURE 3 Standardized cluster evaluator scores for Ward’s linkage classification of the biophysical and sociological datasets. Mean
evaluator scores are shown as a thickened black line. The eight-cluster solution from the biophysical dataset (deemed effective by the
Calinski–Harabasz [C-H] index, and particularly the Dunn index) and the five-cluster solution from the socioeconomic dataset (deemed
particularly effective by the Calinski–Harabasz index) are indicated with dashed lines. W/B ratio, within-to-between ratio.
8of21 AHO ET AL.
steppe, (2) arid foothills, (3) river plain, (4) valley, (5) arid
mountains, (6) maritime mountains, (7) lowland forest,
and (8) upland forest (Figure 4). The k-means
bio-geo-physical types were as follows: (1) shrub steppe,
(2) arid foothills, (3) river plain, (4) arid mountains,
(5) maritime mountains, and (6) forest (Figure 4).
The k-means bio-geo-physical types were largely equiva-
lent to or spatial conflations of Ward’s types. For
instance, river plain and valley types and lowland
forest and upland forest types from the Ward’s classifica-
tion were combined into single river plain and forest
types in the k-means classification (Figure 4). We
named social system types based largely on their
management and human population densities
(Appendices S2, S3, and S5). We designated Ward’s social
system types as follows: (1) dry agriculture/public land,
(2) rural public lands, (3) exurban and rural/private
land, (4) urban, and (5) high-density urban (Figure 5).
The k-means social system types provided poorer
resolution of heavily urbanized regions (the Ward’s
high-density urban type was not detected by k-means)
but greater resolution of sparsely populated areas with
an additional rural low-income type in central Idaho
Summaries of Ward’s and k-means bio-geo-physical
types, including species richness, physical characteristics,
physiognomy, parent material, and soil order, are given
in Appendix S4. Summaries of Ward’s and k-means social
system types, including census means, ownership/
management, and workforce types, are given in
The largest (in land area) bio-geo-physical types for
the Ward’s optimal classification were shrub steppe
) and arid mountains (41,025 km
same types had the largest areas under k-means analysis
of the bio-geo-physical dataset (49,550 and 44,650 km
respectively). The largest social system types under the
Ward’s classification were dryland agriculture/public
land (128,425 km
) and rural public land (68,200 km
These types also had the largest areas under k-means
FIGURE 4 Overlay of the optimal Ward and k-means classification results for the bio-geo-physical dataset: (a) eight-cluster Ward’s
solution and (b) six-cluster k-means solution.
analysis, although their rank order was reversed
(85,800 and 95,750 km
Agreement in the assignment of particular grids to
clusters was generally high for Ward’sandk-means classi-
fications, particularly for solutions with few clusters, or a
similar number of clusters (Figure 6). Percent agreement
for solutions with the same cluster number ranged from
62% (7 clusters) to 94% (2 clusters) for the bio-geo-physical
dataset, and from 63% (3 and 9 clusters) to 95% (4 clusters)
for the social dataset (Figure 6). Percent agreement in all
tested comparisons of Ward’sandk-means classification
solutions with the same number of clusters was statisti-
cally significant at α< 0.01. Agreement between optimal
clustering solutions, as defined by classification evaluators,
was 73%, between Ward’s (eight clusters) and k-means
(six clusters) solutions for the bio-geo-physical dataset, and
87% between Ward’s method (five clusters) and k-means
(five clusters) solutions for the social system dataset.
Percent agreement in these comparisons was also statisti-
cally significant at α<0.05.
Spatially overlaying solutions from social-
bio-geo-physical classifications resulted in 31 SEZ com-
posite classes from Ward’s classifications and 24 SEZ
composite classes for the k-means classifications
(Figure 7). Unlike simpler individual bio-geo-physical
and social system classifications, SEZ composite classes
for k-means and Ward’s were no longer clear spatial
re-representations of each other (maximum
agreement ≈50%; Figure 7). Differences in k-means and
Ward’s SEZs were largely due to additional classes in
optimal Ward’s classifications of bio-geo-physical and
social system datasets (Figures 5and 6). The Ward’s and
k-means frameworks generally agreed with respect to the
characteristics of dominant SEZs. For example, the two
largest SEZs under each method were as follows: arid
mountains/rural public lands (Ward’s=27,200 km
k-means =40,400 km
) and shrub steppe/dryland agri-
culture and public land (Ward’s=40,625 km
FIGURE 5 Overlay of the optimal Ward and k-means classification results for the social system dataset: (a) five-cluster Ward’s solution
and (b) four-cluster k-means solution.
10 of 21 AHO ET AL.
k-means =28,925 km
) (Figure 7). The characteristics
of Ward’s and k-means SEZs are summarized in
Partitioning SEZ variability in multinomial
We quantified the explanatory contributions of
bio-geo-physical, social predictors to SEZs, while controlling
for spatial autocorrelation, using multinomial GLMs. For
both Ward’sandk-means SEZs, the optimal approximating
multinomial model (low AIC model) included only
bio-geo-physical and socialsystempredictors(Tables4and
5). Levels of explained variation remained high for
bio-geo-physical and social system predictors after controlling
for spatial effects for both Ward’slinkage(bio-geo-physical:
devpartial ¼0:77, social system: R2
devpartial ¼0:58) and
k-means analyses (bio-geo-physical: R2
social system: R2
devpartial ¼0.64) of SEZs (Tables 4and 5).
Conversely, spatial effects were extremely small after
controlling for both bio-geo-physical and social predictors
devpartial ¼0.39, k-means: R2
devpartial ≈0). Thus,
with respect to individual explanatory variable subsets,
bio-geo-physical variables had greater explanatory power
for SEZs than social system variables, and both of these
had greater explanatory power than spatial autocorrela-
tion considered alone.
In multinomial models that considered SEZ compos-
ites as a function of only individual scales of spatial auto-
correlation (dbMEMs), the largest levels of explained
variation occurred for the third dbMEM in both Ward’s
pseudo-adj ¼0.1147; Appendix S1: Figure S19)
and k-means analysis (R2
pseudo-adj ¼0.1153; Appendix S1:
Figure S20) of SEZs. Loadings for this dbMEM demon-
strated large-scale patterns in spatial proximity that sepa-
rated central Idaho from regions in north-central and
southwest Idaho (Figure 2).
Multinomial predicted types had high correspondence
with mapped SEZs in cross-validation checks (Tables 6and 7).
For Ward’s SEZs, the bio-geo-physical +social
system model demonstrated 93.6% agreement (b
mean user accuracy =95%, mean producer accuracy =
96%), while for the (simpler) k-mean classifications,
this model resulted in 100% agreement between
model predictions and mapped composite SEZs (b
mean user accuracy =100%, mean producer accuracy =
100%). Better correspondence between multinomial
predictions and mapped composite SEZs occurred for
bio-geo-physical variables alone (Ward’s=75% agree-
ment; k-means =81%), than for social system variables
alone (Ward’s=56% agreement; k-means =63%).
User (commission) error rates were consistently
lower than producer (omission) error rates across multi-
nomial models for both Ward’sandk-means SEZs
(Tables 6and 7). When using both bio-geo-physical and
social predictors, the lowest levels of user and producer
accuracy for Ward’sSEZsoccurredforshrub
steppe/rural public lands (92.3% user, 96.8% producer),
maritime mountains/dryland agriculture_public lands
(92.4% user, 95.5% producer), and arid foothills/rural
public lands (93.8% user, 93.4% producer). As noted
above, the simpler k-means SEZs had 100% user and
producer accuracy for all types when using both
bio-geo-physical and social predictors.
The delineation and mapping of SESs boundaries have
been garnering attention from SES researchers and land
FIGURE 6 Maximum agreement of Ward’s and k-means
classification of grids, for 2–10 cluster solutions, for the
(a) bio-geo-physical data and (b) social systems data.
ECOSPHERE 11 of 21
FIGURE 7 Bio-geo-physical and social system composite SEZs for (a) Ward’s and (b) k-means classifications. See Appendices S2–S5 for
additional details, including percentages of SEZs under different land stewardship categories. SEZ, socio-ecological zone.
12 of 21 AHO ET AL.
managers as a way to consider and quantify the societal
dependence on ecological life-support systems (Castro
et al., 2011; Maass et al., 2016; Quintas-Soriano
et al., 2018). Few studies, however, have explicitly identi-
fied multiple, overlapping SESs boundaries for a region
of interest (Jones et al., 2019; Martín-L
opez et al., 2017).
TABLE 4 Results from multinomial fits of SEZ categories to spatial, bio-geo-physical, and distance-based Moran eigenvector map
(dbMEM) data explanatory subsets to Ward’s linkage SEZ composites
Model ΔAIC R
SEZ =bio-geo-phys +social +spatial 612 0.96
SEZ =bio-geo-phys +social 0 0.94
SEZ =bio-geo-phys +spatial 4034 0.86
SEZ =social +spatial 8019 0.79
SEZ =bio-geo-phys 6626 0.77 0.72 (controlling for spatial)
SEZ =social 16,038 0.58 0.53 (controlling for spatial)
SEZ =spatial 17,179 0.57 0.50 (controlling for social and
SEZ =1 42,979 0.00
Abbreviation: AIC, Akaike information criterion.
SEZ =socio-ecological zone designations based on Ward’s-linkage classifications; social =principal components 1–11 from final social PCA;
bio-geo-phys =principal components 1–14 from final bio-geo-physical PCA; spatial =MEMdbs 1–25; SEZ =1 refers to an intercept-only model.
TABLE 5 Results from multinomial fits SEZ categories to spatial, biophysical, and distance-based Moran eigenvector map (dbMEM)
explanatory data subsets to k-means classification SEZ composites
Model ΔAIC R
SEZ =bio-geo-phys +social +spatial 1100 1.00
SEZ =bio-geo-phys +social 0 1.00
SEZ =bio-geo-phys +spatial 5553 0.89
SEZ =social +spatial 7492 0.84
SEZ =bio-geo-phys 7935 0.81 0.74 (controlling for spatial)
SEZ =social 14,460 0.66 0.63 (controlling for spatial)
SEZ =spatial 18,844 0.57 ≈0 (controlling for social and
SEZ =1 43,234 0.00
Abbreviation: AIC, Akaike information criterion.
SEZ =socio-ecological zone designations, based on k-means classifications; social =principal components 1–11 from final social PCA;
bio-geo-phys =principal components 1–14 from final bio-geo-physical PCA; spatial =MEMdbs 1–25. SEZ =1 refers to an intercept-only model.
TABLE 6 Cross-validation of Ward’s mapped socio-ecological zone (SEZ) composites and multinomial predicted types for different
Model Agreement (%) b
KUser (%) Producer (%)
SEZ =bio-geo-phys +social +spatial 98.5 98.3 99.4 99.5
SEZ =bio-geo-phys +social 96.3 96.0 97.8 98.1
SEZ =bio-geo-phys +spatial 88.4 87.2 82.7 87.9
SEZ =social +spatial 79.4 77.2 86.7 87.9
SEZ =bio-geo-phys 78.5 76.2 57.0 74.1
SEZ =social 59.5 55.0 67.5 74.1
SEZ =spatial 58.5 53.8 41.9 59.3
Krefers to the kappa classification statistic. User and producer accuracy refer to mean user and producer percent accuracy.
ECOSPHERE 13 of 21
While SES boundaries are dynamic time and space, we
hope that defining current SES boundaries in Idaho will
allow the identification and tracking of socio-ecological
dynamics in one of the most rapidly growing regions in
the United States (Quintas-Soriano et al., 2022).
Our analytical approach addresses a number of defi-
ciencies in other SES studies. Specifically, it: (1) considers
the accumulated effects of clustering algorithm predilec-
tions on SEZ designations, (2) applies cross-validation
measures to SEZs, (3) quantifies the relative explanatory
contributions of bio-geo-physical and social system pre-
dictors in defining SEZs, and (4) quantifies and accounts
for spatial autocorrelation effects on SEZs. Components
of our approach may be useful in other ecological settings
when multinomial responses resulting from multivariate
analyses, for example, clustering solutions, have a spatial
In our study, Ward’s and k-means classifications of
individual bio-geo-physical and social system datasets
were quite similar (Figure 6). Indeed, these similarities
led to the recognition of the same dominant SEZs in both
classifiers, namely arid mountains/rural public lands,
shrub steppe/dryland agriculture and public land, and
arid foothills/dryland agriculture and public land
(Figure 7). In contrast, composite SEZs generated by
Ward’s linkage and k-means analysis were relatively distinct,
particularly at fine spatial scales. The Ward’sSEZsoffered
greater spatial resolution, and higher cross-validation error
rates, due to a larger number of optimal bio-geo-physical
and social system clusters (Table 5). Thus, for our applica-
concerning coarser spatial scales.
Much of Idaho’s human population occurs in
urban-to-rural networks of mid-sized cities that have
tighter urban-to-rural feedback and less inertia to develop-
ment than larger cities (Allen et al., 2016;Feltetal.,2018).
Urbanization of agricultural land is the dominant land use
change in Idaho, altering historical feedback between local
resource production and consumption (Hubbard, 2017;
Quintas-Soriano et al., 2022). Our Ward’sSEZsmaybe
particularly useful for examining these trends. For exam-
ple, in the northern portion of Ada County, which con-
tains the state capital, urban dynamics are predominant.
Within this environment, Ward’s SEZs clearly distinguish
high-density urban and suburban types (Figure 5), and
thus could be useful for addressing city planning objectives
and establishing comparative baselines for urban/exurban
trends. The southern portion of Ada County is composed
mostly of Bureau of Land Management (BLM) land with
interspaced parcels of state land. On the other hand, the
k-means SEZs often provided greater resolution of sparsely
populated areas, particularly in southern Idaho. For
instance, k-means types identify important intersections of
exurbia/private lands and forests in the south-central por-
tion of the state. In this case, discrepancies in SEZ bound-
aries could affect efforts to effectively manage BLM lands
for ranching, hunting, and other recreation purposes, given
the high demand for these ecosystem services from the
affluent populations living in the greater Boise metropolitan
area (Narducci et al., 2019; Quintas-Soriano et al., 2022).
As Idaho mid-sized cities continue to grow and agricul-
tural conversion to urban areas continues, increasing
efforts will be required to manage land resources and
water supplies. Particularly pressing are issues including
storm water runoff, aquifer vulnerability, septic system
integrity, sewage systems, and human impacts to riparian
areas. In this context, SEZs designations could guide con-
servation/development of riparian and agricultural land
for multiple ecosystem services. For example, Huang et al.
(2019) found that designation and conservation of riparian
areas through the US Department of Agriculture
Conservation Reserve Program helped maintain freshwa-
ter ecosystem services and mitigate climate change effects.
Of the three major Idaho land stewards (BLM, US
Forest Service [USFS], and private), privately owned
lands are arguably the most affected by water use and
TABLE 7 Cross-validation of mapped k-means socio-ecological zone (SEZ) composites and multinomial predicted types for different
Model Agreement (%) b
KUser (%) Producer (%)
SEZ =bio-geo-phys +social +spatial 100 1.0 100 100
SEZ =bio-geo-phys +social 100 1.0 100 100
SEZ =bio-geo-phys +spatial 87.8 86.5 85.0 88.5
SEZ =social +spatial 83.9 82.2 87.7 89.5
SEZ =bio-geo-phys 79.2 76.8 57.4 78.7
SEZ =social 64.8 60.9 72.4 75.2
SEZ =spatial 59.4 54.5 38.8 58.7
Krefers to the kappa classification statistic. User and producer accuracy refer to mean user and producer percent accuracy.
14 of 21 AHO ET AL.
access regulations, and/or the Endangered Species Act of
1973, which allows some federal oversight of water on
private lands to protect imperiled species (Ruhl, 1998).
Our SEZs consisting of shrub steppe and river plain
bio-geo-physical composites with exurban-rural and agri-
culture social systems were similar for Ward’s linkage
and k-means analysis (Figure 7) and would likely prompt
similar water management perspectives.
BLM lands are mostly composed of dryland agricul-
ture and various public lands types, mixed with either
shrub steppe or river plains in our SEZ. BLM lands are
broken into several management districts across the
state with generalized goals of maintaining multiuse
landscapes including managing mines and resource
exploration, public recreation access and opportunities,
trail systems, grazing, and wildfire control activities
(https://www.blm.gov/idaho). As a practical matter,
BLM public lands are generally more available for
resource consumption/extraction via private entities,
under the Mineral Leasing Act of 1920 and other policies,
than those managed by other federal agencies. In our
application, the use of one SEZ framework over the other
would be unlikely to drive different management
approaches on BLM land, although a potential concern
highlighted by both SEZ maps is the intersection of
farming and ranching lands with federal motorized
off-road vehicle trail use (Figure 7), because conflict
between these user groups is likely.
The USFS is the largest land steward in Idaho, with
lands comprising approximately 40% of the state, and
SEZs that are primarily rural and/or federal forest with
mountains and forest (Appendix S3). USFS lands are
administered by state management districts with general
goals of maintaining healthy forest habitats, sustainable
harvest of forest products, and providing public access
and recreation opportunities (United States Forest
Service Policies and Projects, 2021). Although constitut-
ing <1 of Idaho GDP (United States Bureau of Economic
Analysis, 2021), revenues generated from USFS forest
product sales in Idaho remain an important management
concern, because 25% of this money is designated for
school programs and road maintenance in associated
counties. Within USFS lands, clustering methodology
strongly influenced SEZs, particularly in the northern
part of the state, with potential management implications
(Figure 7; Appendix S3).
By defining the overlap of bio-geo-physical and social
system boundaries, SEZs recognize the connections and
feedbacks linking human and natural systems (Castro
et al., 2014). The usefulness of these designations can
only be improved through a rigorous consideration of
relevant analytical issues. In addition to SEZ discrepan-
cies resulting from different statistical classifiers,
described above, several other analytical issues have been
largely ignored in the literature. For example, the efficacy
of our SEZs framework was demonstrated using
cross-validation of multinomial predicted values and
mapped SEZs. Our cross-validation approach can be criti-
cized on the grounds that a separate training dataset was
not used to generate SEZs. However, we justify our
approach on the grounds that the best possible SEZ maps
(using all possible data) were desired for the purpose of
practical application. We found that bio-geo-physical
information was more important than social system
information in defining SEZs, after controlling for spatial
autocorrelation. This finding is not surprising in a state
with large regions of intact natural ecosystems, which
result in the provision of particular ecosystem services,
which will, in-turn, drive particular socioeconomic
constraints (e.g., predominance of jobs in extractive
industries, prevalence of federal jobs, importance of eco-
tourism, wage distributions, etc.), and resultant SEZs.
This trend is likely to be moderated or reversed in more
urbanized states, given the same quality and quantity of
bio-geo-physical and social data.
Historical context, ecosystem services, and
land management in Idaho
Land management will determine the ecological fate of
Idaho lands and the character of their corresponding eco-
system services. As relevant historic examples, national
concern over the negative effects of timber harvests on
water supplies and soil erosion prompted the establish-
ment of the US Forest Reserve Act of 1891, and federal
forest oversight by the USFS (Gorte & Cody, 1995), and
deterioration of croplands prompted the enactment of the
Taylor Grazing Act of 1934, and the subsequent creation
of the BLM in 1946 to manage public rangelands
(Dana & Fairfax, 1980). Other environmental regulatory
policies, important to Idaho, include federal edicts like
the Clean Water Act of 1972, which protects the quality
of drinking water and has resulted in several Idaho state
policies including the Agricultural Pollution Abatement
Plan—the Clean Air Act of 1970, and the Endangered
Species Act of 1973, and state statutes for agencies like
the Idaho Department of Environmental Quality
Western states were established during coincident
shifts in US land use policy, from private dispensation to
homesteaders, to federal retention—resulting in large
reserves of federal public lands in most western states
(Jones et al., 2019; West, 1992). Legislative bodies in west-
ern states including Idaho (62% federal land), Utah (63%
federal land), and Nevada (85% federal land) have all
ECOSPHERE 15 of 21
recently considered or enacted measures to transfer fed-
eral land management and protection to state and private
control (Congressional Research Service, 2020). These
efforts are exemplified in the so-called “sagebrush rebel-
lion”of the 1970s and 1980s and the Utah Transfer of
Public Lands Act of 2012—a state law that has failed to
result in federal proceedings. While of dubious legal and
economic merit (Keiter & Ruple, 2015), these endeavors
demonstrate the ongoing tensions between economic
interests and ecosystem conservation efforts in Idaho,
which are further fueled by dramatic population growth.
Given this context, SEZ mapping in Idaho is particularly
important because it allows the recognition and tracking
of socio-ecological dynamics in Idaho, facilitating their
Idaho indigenous tribes
We considered Idaho indigenous social systems data by
including indigenous reservations as a land management
type, and indigenous population numbers in social
system classifications. These efforts, however, clearly
remain inadequate for conveying the historical impor-
tance of traditional indigenous social systems to Idaho
social–ecological zones over thousands of years, and the
radical shifts in anthropogenic impacts and ecosystem
service perspectives as those social systems have been
Cessations of indigenous territories in the western
United States have involved the official transfer of hun-
dreds of square miles of land from indigenous to federal
and private ownership, often in violation of prior treaties
(e.g., the treaty of Fort Laramie 1868; National Archives
and Records Administration, 2016). For regions now
encompassing the state of Idaho, cessations of tribal
lands are detailed in no fewer than 18 Acts of Congress
enacted in the late 19th century (United States Library of
Idaho native tribes include the Kootenai, Kalispel,
Coeur d’Alene, Palouse, Nimi’ipuu (Nez Perce),
Northern Paiute in northern and central Idaho, and
the Shoshoni/Bannock, including the Warraeekas,
Tukuerukas (sheepeaters), Agaidikas (salmon eaters),
and Pohugue (people of the sage), which occupied
large expanses of southern Idaho on the Snake River
plain. Boundaries for Idaho tribal occupation were not
incorporated into base layers for analysis as these
delineations are overlapping, approximate, and histori-
cally fluid, and often fall along natural boundaries
TABLE 8 Environmental policies and agencies important to the management of Idaho lands.
Policy name Year enacted Managing agency Asset
Clean Water Act 1972 EPA Wetlands and streams
Safe Drinking Water Act 1974 EPA Drinking water resources
Clean Air Act 1970 EPA Atmosphere
Resource Conservation and Recovery Act 1976 EPA Regions affected by human waste disposal
Comprehensive Environmental Response,
Compensation, and Liability Act
1980 EPA Cleanup of chemically contaminated sites
Endangered Species Act 1973 USFWS Imperiled species/habitats
National Forest Management Act 1976 USFS Federal forest resources
Federal Land Policy and Management Act 1976 BLM Primarily federal rangeland resources
Reclamation Act 1902 BOR Water resources for irrigation
Idaho Irrigation and Drainage, Code 42 1895 IDWS Watersheds (water rights)
Idaho State Policy on Environmental
Protection, Code 39-1
1972 IDEQ Atmosphere/Watersheds
Idaho Department of Lands, Code 58-5
1937 Idaho state lands
Idaho Fish and Game, Code 36-101 through
1976 IDFG Wildlife (primarily game species)
Abbreviations: BOR, Bureau of Reclamation; EPA, US Environmental Protection Agency; IDEQ, Idaho Department of Environmental Quality; IDFG, Idaho
Department of Fish and Game; IDWS, Idaho Department of Water Resources; USFS, US Forest Service; USFWS, US Fish and Wildlife Service.
16 of 21 AHO ET AL.
coincident with bio-geo-physical data (Digital Atlas of
Idaho, 2022; Ray et al., 1938). Nonetheless, we empha-
size the importance of future consideration of these
historic patterns in the context of socio-ecological
research in Idaho.
The character of natural ecosystems fundamentally
and uniquely underly the traditional cultural and
spiritual perspectives of Idaho indigenous tribes. For
instance, Burger (2012) found that Bannock-Shoshoni
tribal members were more likely to value complex eco-
system characteristics (e.g., food chains) and emergent
processes (e.g., species diversity) than local Caucasians
in Idaho. Tribal perspectives, however, have become
increasingly incorporated into management decisions
concerning resources on western public lands, and may
be further influenced and informed by reliable SEZs
for these regions. For instance, several recent court rul-
ings have supported tribal efforts to restore indigenous
social–ecological systems by requiring that state gov-
ernments co-manage Pacific salmon (Oncorhynchus
sp.) populations with tribalrepresentatives,andthe
federal Northwest Forest Plan has specifically consid-
ered effective consultation with tribes and protections
afforded by tribal treaty rights (Long & Lake, 2018).
The US Department of the Interior (which includes the
BLM) is the primary federal agency charged with
maintaining relationships with federally recognized
tribes, and promoting tribal self-determination
(DOI, 2022). Detailed SEZ designations would be par-
ticularly useful for ongoing BLM-tribal consultations
in western states regarding coal leases and other
resource decisions (BLM, 2022).
Our methodological approach for delineating and map-
ping SEZs addresses both the strong need for description
of current SES spatial patterns in one of the most rapidly
growing regions in the United States, and the general need
for analytical refinement in SES mapping. With regard to
analytical considerations, we: (1) demonstrate the poten-
tial distinctiveness of SEZ types resulting from different
clustering algorithms, (2) provide cross-validation mea-
sures of SEZ efficacy, (3) quantify the individual contribu-
tions of bio-geo-physical and social system predictors in
defining SEZs, and (4) explicitly account for spatial auto-
correlation effects on SEZs. Our analytical approaches are
broadly applicable to ecological settings in which a multi-
nomial response variable (e.g., cluster types) has a spatial
component. Our SEZ composites identify hotspots for
management consideration in Idaho, including pivotal
regions with high water management concerns and
“Exurban to Rural”proximal to relatively undisturbed
bio-geo-physical types that are optimal for development.
In general, SEZs can be developed to guide particular eco-
logical concerns, including threatened species, and cul-
tural and societal concerns including those of historically
This publication was made possible by the National
Science Foundation (NSF) Idaho Established Program to
Stimulate Competitive Research (EPSCoR) under NSF
award number IIA-1301792. Support for this research
was also provided by the NSF via RC CZO Cooperative
agreement NSF EAR-1331872. We acknowledge that
Idaho State University (in Pocatello Idaho) is located
within the boundaries of the original Fort Hall
Reservation on the traditional lands of the Shoshone and
Bannock peoples. The state of Idaho is located on the tra-
ditional lands of the Kootenai, Kalispel, Coeur d’Alene,
Palouse, Nimi’ipuu (Nez Perce), Northern Paiute, and
CONFLICT OF INTEREST
The authors declare no conflict of interest.
DATA AVAILABILITY STATEMENT
Datasets used for this research are publicly available and
are as follows:
1. National Elevational Dataset (NED, 2018): https://
lta.cr.usgs.gov/NED (see Table 1for variables
2. Panchromatic Remote-sensing Instrument for Stereo
Mapping Data (PRISM, 2015): https://prism.jpl.nasa.
gov/prism_data.html (see Table 1for variables used).
3. Moderate Resolution Imaging Spectroradiometer
Data (MODIS, 2022): https://modis.gsfc.nasa.gov/
data/ (see Table 1for variables used).
4. National Land Cover Database (NLCD, 2022):
land-cover-database (see Table 1for variables used).
5. Mineral Resources Online Spatial Data (MRData, 2022):
https://mrdata.usgs.gov/general/map-us.html (see Table
1for variables used).
6. Soil Survey Geographic Database (SSURGO, 2022):
http://websoilsurvey.nrcs.usda.gov/ (see Table 1for
7. Digital General Soil Map of the United States
gov/ (see Table 1for variables used).
8. BiodiversityMapping.org (Jenkins, 2022; Jenkins
et al., 2013): https://biodiversitymapping.org/ (see
Table 2for variables used).
ECOSPHERE 17 of 21
9. American Community Survey (ACS, 2022): https://
www.census.gov/programs-surveys/acs/ (see Table 2
for variables used).
10. Idaho Surface Management Agency Data (ISMA,
(see Table 2for variables used).
11. Idaho Traffic Density Data (ITD, 2022): https://itd.
idaho.gov/road-data/ (see Table 2for variables used).
12. Global Human Modification dataset (Kennedy
et al., 2019) (see Table 2).
All code for this paper (Aho, 2022), including novel
code, is available from Zenodo: https://zenodo.org/
Ken Aho https://orcid.org/0000-0001-5998-2916
Susan Parsons https://orcid.org/0000-0003-4317-1427
Antonio J. Castro https://orcid.org/0000-0003-1587-
Kathleen A. Lohse https://orcid.org/0000-0003-1779-
ACS. 2022 “United States Census Bureau.”https://www.census.
Aho, K. 2013. “Foundational and Applied Statistics for Biologists
Using R.”In Introduction to R. Boca Raton, FL: CRC Press.
Taylor & Francis Group.
Aho, K. 2020. “asbio: A Collection of Statistical Tools for
Biologists.”R Package Version 1.6-8. https://cran.r-project.org/
Aho, K. 2022. “R Code for the Article: ‘Mapping Socio-Ecological
Systems in Idaho: Spatial Patterns and Analytical
Aho, K., D. W. Roberts, and T. Weaver. 2008. “Using Geometric
and Non-geometric Internal Evaluators to Compare Eight
Vegetation Classification Methods.”Journal of Vegetation
Science 19: 549–62.
Alessa, L. N., A. A. Kliskey, and G. Brown. 2008. “Social-Ecological
Hotspots Mapping: A Spatial Approach for Identifying Coupled
Social-Ecological Space.”Landscape and Urban Planning 85:
Allen, C. R., H. E. Birge, S. Bartelt-Hunt, R. A. Bevans, J. L.
Burnett, B. A. Cosens, X. Cai, et al. 2016. “Avoiding Decline:
Fostering Resilience and Sustainability in Midsize Cities.”
Sustainability 8(9): 844.
Bailey, R. G. 2009. Ecosystem Geography from Ecoregions to Sites.
New York: Springer.
Balch, J. K., B. A. Bradley, C. M. D’Antonio, and J. G
2013. “Introduced Annual Grass Increases Regional Fire
Activity across the Arid Western USA (1980–2009).”Global
Change Biology 1: 173–83.
Berkes, F., and C. Folke. 1998. Linking Social and Ecological
Systems: Management Practices and Social Mechanisms for
Building Resilience. New York: Cambridge University Press.
BLM. 2022. “Tribal Consultations.”https://www.blm.gov/services/
Borcard, D., P. Legendre, C. Avois-Jacquet, and H. Tuomisto. 2004.
“Dissecting the Spatial Structure of Ecological Data at
Multiple Scales.”Ecology 85(7): 1826–32.
G. Midgley. 2016. “A Socio-Ecological Approach for
Identifying and Contextualizing Spatial Ecosystem-Based
Adaptation Priorities at the Sub-National Level.”PLoS One
Burger, J. 2012. “Perceptions of Goods, Services and Eco-Cultural
Attributes of Native Americans and Caucasians in Idaho.”
Remediation Journal 22(3): 105–21.
Calinski, T., and J. Harabasz. 1974. “A Dendrite Method for Cluster
Analysis.”Communications in Statistics 3(1): 1–27.
Castellarini, F., C. Siebe, E. Lazos, B. de la Tejera, H. Cotler,
C. Pacheco, E. Boege, et al. 2014. “A Social-Ecological Spatial
Framework for Policy Design Towards Sustainability: Mexico
as a Study Case.”Investigaci
on Ambiental 6: 45–59.
Castro, A. J., B. Martín-L
opez, M. García-Llorente, P. A. Aguilera,
opez, and J. Cabello. 2011. “Social Preferences Regarding
the Delivery of Ecosystem Services in a Semiarid Mediterranean
Region.”Journal of Arid Environments 75: 1201–8.
Castro, A. J., C. Quintas-Soriano, J. Brandt, C. L. Atkinson, C. V.
Baxter, M. Burnham, B. N. Egoh, et al. 2018. “Applying
Place-Based Social-Ecological Research to Address Water
Scarcity: Insights for Future Research.”Sustainability 10: 1516.
Castro, A. J., P. H. Verburg, B. Martín-L
opez, M. Garcia-Llorente,
J. Cabello, C. C. Vaughn, and E. Lopez. 2014. “Ecosystem
Service Trade-Offs from Supply to Social Demand:
A Landscape-Scale Spatial Analysis.”Landscape and Urban
Planning 132: 102–10.
Cervantes-Jiménez, M., C. A. Mastachi-Loza, C. Díaz-Delgado,
omez-Albores, and E. Gonz
“Socio-Ecological Regionalization of the Urban Sub-Basins in
Mexico.”Water 9(1): 14.
Cherkasskii, B. L. 1988. “The System of the Epidemic Process.”
Journal of Hygiene, Epidemiology, Microbiology, and Immunology
Congalton, R. G., and R. A. Mead. 1983. “A Quantitative Method
to Test for Consistency and Correctness in
Photointerpretation.”Photogrammetric Engineering and
Remote Sensing 49(1): 69–74.
Congressional Research Service. 2020. “Federal Land Ownership:
Overview and Data.”CRS Report R42346. https://crsreports.
Cornell, J. D., C. Quintas-Soriano, K. Running, and A. J. Castro.
2019. “Examining Concern about Climate Change and Local
Environmental Changes from an Ecosystem Service
Perspective in the Western U.S.”Environmental Science and
Policy 101: 221–31.
Cruz-Cardenas, G., J. T. Silva, S. Ochoa-Estrada, F. Estrada-Godoy,
and J. Nava-Velazquez. 2017. “Delineation of Environmental
Units by Multivariate Techniques in the Duero River
Watershed, Michoacan, Mexico.”Environmental Modeling and
Assessment 22: 257–66.
18 of 21 AHO ET AL.
Dana, S. T., and S. K. Fairfax. 1980. Forest and Range Policy: Its
Development in the United States, 2nd ed. 158–64. New York:
Díaz, S., S. Demissew, J. Carabias, C. Joly, M. Lonsdale, N. Ash,
A. Larigauderie, et al. 2015. “The IPBES Conceptual
Framework-Connecting Nature and People.”Environmental
Sustainability 14: 1–16.
Digital Atlas of Idaho. 2022. “The Peoples of Idaho: Native Settlers.”
College of Science and Engineering Idaho State University.
DOI. 2022. “International Affairs: Tribes.”https://www.doi.gov/
Dray, S., D. Bauman, G. Blanchet, D. Borcard, S. Clappe,
G. Guenard, T. Jombart, et al. 2018. “adespatial: Multivariate
Multiscale Spatial Analysis.”R Package Version 0.3-2. https://
Dressel, S., G. Ericsson, and C. Sandström. 2018. “Mapping
Social-Ecological Systems to Understand the Challenges
Underlying Wildlife Management.”Environmental Science and
Policy 84: 105–12.
Dunn, J. 1974. “Well Separated Clusters and Optimal Fuzzy
Partitions.”Journal of Cybernetics 4: 95–104.
Efron, B. 1978. “Regression and ANOVA with Zero–One Data:
Measures of Residual Variation.”Journal of the American
Statistical Association 73: 113–21.
Felt, C., M. Fragkias, D. Larson, H. Liao, K. A. Lohse, and
D. Lybecker. 2018. “A Comparative Study of Urban
Fragmentation Patterns in Small and Mid-Sized Cities of
Idaho.”Urban Ecosystem 21: 805–16.
Forbes. 2018. “America’s Fastest-growing Cities.”https://www.
Gardner, J. W. 1991. “Detection of Vapours and Odours from a
Multisensor Array Using Patterns Recognition Part
1. Principal Component and Cluster Analysis.”Sensors and
Actuators B: Chemical 4(1–2): 109–15.
Gorte, R. W., and B. A. Cody. 1995. The Forest Service and Bureau
of Land Management: History and Analysis of Merger
Proposals. Washington, DC: Congressional Research Service,
Library of Congress.
Griffith, D. A. 1996. “Spatial Autocorrelation and Eigenfunctions of
the Geographic Weights Matrix Accompanying
Geo-Referenced Data.”Canadian Geographer 40: 351–67.
Hamann, M., R. Biggs, and B. Reyers. 2015. “Mapping
Social-Ecological Systems: Identifying ‘Green-Loop’and
‘Red-Loop’Dynamics Based on Characteristic Bundles of
Ecosystem Service Use.”Global Environmental Change 34:
Hanspach, J., J. Loos, I. Dorresteijn, D. J. Abson, and J. Fischer.
2016. “Characterizing Social-Ecological Units to Inform
Biodiversity Conservation in Cultural Landscapes.”Diversity
and Distributions 22(8): 853–64.
Hartigan, J. A., and M. A. Wong. 1979. “Ak-Means Clustering
Algorithm.”Applied Statistics 28: 100–8.
Hennig, C. 2020. “fpc: Flexible Procedures for Clustering.”R Package
Version 2.2-9. https://CRAN.R-project.org/package=fpc.
Herrero-Jauregui, C., C. Arnaiz-Schmitz, L. Herrera, S. M. Smart,
C. Montes, F. D. Pineda, and M. Fe Schmitz. 2019. “Aligning
Landscape Structure with Ecosystem Services along an
Urban–Rural Gradient. Trade-Offs and Transitions towards
Cultural Services.”Landscape Ecology 34: 1525–45.
Huang, L., F. H. Liao, K. Lohse, D. Larson, M. Fragkias,
D. Lybecker, and C. Baxter. 2019. “Land Conservation Can
Mitigate Freshwater Ecosystem Services Degradation Due to
Climate Change in a Semiarid Catchment: The Case of the
Portneuf River Catchment, Idaho, USA.”Science of the Total
Environment. 651(2): 1796–809.
Hubbard, M. 2017. “Farm Land Conversion in the American West.
The Blue Review.”https://thebluereview.org/farm-land-
Idaho Department of Agriculture. 2021. https://agri.idaho.gov/
ISMA. 2022. “BLM national surface management agency area
polygons—National Geospatial Data Assets (NGDA).”https://
ITD. 2022. “Idaho Transportation Department Road Data.”https://
Jenkins C. N. 2022. BiodiversityMapping.org.https://
Jenkins, C. N., S. L. Pimm, and L. N. Joppa. 2013. “Global Patterns
of Terrestrial Vertebrate Diversity and Conservation.”
Proceedings of the National Academy of Sciences 110(28):
Jones, K., J. Abrams, T. Belote, B. J. Beltran, J. Brandt, N. H. Carter,
A. J. Castro, et al. 2019. “The American West as a
Social-Ecological Region: Drivers, Dynamics and Implications
for Nested Social-Ecological Systems.”Environmental Research
Letters 14(11): 115008.
Keiter, R.B. and J. Ruple. 2015. “The Transfer of Public Lands
Movement: Taking the ‘Public’out of Public Lands.”Stegner
Center White Paper No. 2015-01, S. J. Quinney College of Law
Research Paper. https://ssrn.com/abstract=2555922.
Kennedy, C. M. J. R., D. M. Oakleaf, S. B.-M. Theobald, and
J. Kiesecker. 2019. “Managing the Middle: A Shift
in Conservation Priorities Based on the Global Human
Modification Gradient.”Global Change Biology 25: 811–26.
Klos, P. Z., T. E. Link, and J. T. Abatzoglou. 2014. “Extent of the
Rain-Snow Transition Zone in the Western U.S. under
Historic and Projected Climate.”Geophysical Research Letters
Kok, M., T. Sterzel, and D. Sietz. 2016. “A New Method for
Analysing Socio-Ecological Patterns of Vulnerability.”
Regional Environmental Change 16: 229–43.
Kullman, L. 2004. “The Changing Face of the Alpine World.”
Global Change Newsletter 57(1): 12–4.
Legendre, P., and L. F. Legendre. 2012. Numerical Ecology, 3rd ed.
Liu, J., T. Dietz, S. R. Carpenter, M. Alberti, C. Folke, E. Moran,
A. N. Pell, et al. 2007. “Complexity of Coupled Human and
Natural Systems.”Science 317: 1513–6.
Long, J. W., and F. K. Lake. 2018. “Escaping Social-Ecological
Traps through Tribal Stewardship on National Forest Lands in
the Pacific Northwest, United States of America.”Ecology and
Society 23(2): 10.
Maass, M., P. Balvanera, P. Bourgeron, M. Equihua, J. Baudry,
J. Dick, M. Forsius, et al. 2016. “Changes in Biodiversity and
ECOSPHERE 19 of 21
Trade-Offs among Ecosystem Services, Stakeholders, and
Components of Well-Being: The Contribution of the
International Long Term Ecological Research Network
(ILTER) to Programme on Ecosystem Change and Society
(PECS).”Ecology and Society 21(3): 31.
MacQueen, J. B. 1967. “Some Methods for Classification and
Analysis of Multivariate Observations.”In Proceedings of 5th
Berkeley Symposium on Mathematical Statistics and
Probability, edited by L. M. Le Cam, and J. Neyman, 281–97.
Berkeley, CA: University of California Berkeley Press.
Maes, J., B. Egoh, L. Willemen, C. Liquete, P. Vihervaara, J. P.
Schagner, B. Grizzetti, et al. 2012. “Mapping Ecosystem
Services for Policy Support and Decision Making in the
European Union.”Ecosystem Services 1: 31–9.
opez, B., I. Palomo, M. Garcia-Llorente, I. Iniesta-Arandia,
A. J. Castro, D. G. Del Amo, E. Gomez-Baggethun, and
C. Montes. 2017. “Delineating Boundaries of Social-Ecological
Systems for Landscape Planning: A Comprehensive Spatial
Approach.”Land Use Policy 66: 90–104.
McClain, J. O., and V. R. Rao. 1975. “CLUSTISZ: A Program to Test
for the Quality of Clustering of a Set of Objects.”Journal of
Marketing Research 12(4): 456–60.
Mimura, K., Y. Shirakawa, S. Nakamura, and M. Koshiba. 2013.
“Multivariate PCA Analysis Combined with Ward’s Clustering
for Verification of Psychological Characterization in Visually
and Acoustically Social Contexts.”Journal of Clinical
Toxicology 3(1): 10–4172.
MODIS. 2022. “National Aeronautics and Space Administration.”
MRData. 2022. “United States Geological Survey.”https://mrdata.
Narducci, J., C. Quintas-Soriano, A. J. Castro, R. Castellanos, and
J. Brandt. 2019. “Implications of Urban Growth and Farmland
Loss for Ecosystem Services in the Western United States.”
Land Use Policy 86: 1–11.
National Archives and Records Administration. 2016. “Sioux Treaty of
NED. 2018. “United State Geological Survey.”https://lta.cr.usgs.
NLCD. 2022. “United States Geological Survey.”https://www.usgs.
Ostrom, E. 2009. “A General Framework for Analyzing Sustainability
of Social-Ecological Systems.”Science 325: 419–22.
Pimm, S. L., C. N. Jenkins, R. Abell, T. M. Brooks, J. L. Gittleman,
L. N. Joppa, P. H. Raven, C. M. Roberts, and J. O. Sexton.
2014. “The Biodiversity of Species and their Rates of
Extinction, Distribution, and Protection.”Science 344(6187):
PRISM. 2015. “Jet Propulsion Laboratory, California Institute of
Quintas-Soriano, C., J. Brandt, C. V. Baxter, J. M. Requena-Mullor,
and A. J. Castro. 2022. “A Framework for Assessing Coupling
and de-Coupling Trajectories in River Social-Ecological
Systems.”Sustainability Science 17(1): 121–34.
Quintas-Soriano, C., J. Brandt, K. Running, C. V. Baxter, D. M.
Gibson, J. Narducci, and A. J. Castro. 2018. “Social-Ecological
Systems Influence Ecosystem Service Perception: A
Programme on Ecosystem Change and Society (PECS)
Analysis.”Ecology and Society 23(3): 3.
Quintas-Soriano, C., M. A. Gibson, J. Brandt, M. L
J. Cabello, P. Aguilera, and A. J. Castro. 2021. “An
Interdisciplinary Assessment of Private Conservation Areas in
the Western United States.”Ambio 50: 150–62.
R Core Team. 2020. R: A Language and Environment for Statistical
Computing. Vienna: R Foundation for Statistical Computing.
Ray, V. F., G. P. Murdock, B. Blyth, O. C. Stewart, J. Harris, E. A.
Hoebel, and D. B. Shimkin. 1938. “Tribal Distribution in
Eastern Oregon and Adjacent Regions.”American
Anthropologist 40(3): 384–415.
Redman, C., M. J. Grove, and L. Kuby. 2004. “Integrating Social
Science into the Long Term Ecological Research (LTER)
Network: Social Dimensions of Ecological Change and Ecological
Dimensions of Social Change.”Ecosystems 7(2): 161–71.
Ruhl, J. B. 1998. “The Endangered Species Act and Private
Property: A Matter of Timing and Location.”Cornell Journal
of Law and Public Policy 8(37): 37–53.
Shwiff, S., J. Holderieath, W. Haden-Chomphosy, and A. Anderson.
2018. “Economics of Invasive Species Damage and Damage
Management.”In Ecology and Management of Terrestrial
Vertebrate Invasive Species in the United States, edited by W. C.
Pitt, J. C. Beasley, and G. W. Witmer, 35–9. Boca Raton, FL:
Sinare, H., L. J. Gordon, and E. E. Kautsky. 2016. “Assessment of
Ecosystem Services and Benefits in Village Landscapes—A
Case Study from Burkina Faso.”Ecosystem Services 21:
SSURGO. 2022. “Soil Survey Staff, Natural Resources Conservation
Service.”United States Department of Agriculture. Web Soil
STATSGO2. 2022. “Soil Survey Staff, Natural Resources Conservation
Service.”United States Department of Agriculture. Web Soil
Tang, C., B. T. Crosby, J. M. Wheaton, and T. C. Piechota. 2012.
“Assessing Streamflow Sensitivity to Temperature Increases in
the Salmon River Basin, Idaho.”Global and Planetary Change
Theobald, D. M. 2013. “A General Model to Quantify Ecological
Integrity for Landscape Assessments and US Application.”
Landscape Ecology 28(10): 1859–74.
United States Bureau of Economic Analysis. 2021. “GDP by State.”
United States Census Bureau. 2021. “State Population Totals and
Components of Change: 2010-2019.”https://www.census.gov/
United States Forest Service. 2021. “Frank Church River of No
United States Forest Service Policies and Projects. 2021. https://
United States Library of Congress. 2022. “Indian Land Cessions in
the United States, 1784–1894.”United States Serial Set,
Number 4015. https://memory.loc.gov/ammem/amlaw/lwss-
Vallejos, M., S. Aguiar, G. Baldi, M. E. Mastrangelo, F. Gallego,
“Social-Ecological Functional Types: Connecting People and
Ecosystems in the Argentine Chaco.”Ecosystems 23(3): 471–84.
20 of 21 AHO ET AL.
Venables, W. N., and B. D. Ripley. 2002. Modern Applied Statistics
with S, 4th ed. New York: Springer.
Ward, J. H., Jr. 1963. “Hierarchical Grouping to Optimize an
Objective Function.”Journal of the American Statistical
Association 58: 236–44.
West, T. L. 1992. “Centennial Mini-Series Histories of the Forest
Service.”United Department of Agriculture. Forest Service
Winter, K. B., and M. Lucas. 2017. “Spatial Modeling of
Social-Ecological Management Zones of the Ali’I Era on the
Island of Kaua’I with Implications for Large-Scale Biocultural
Conservation and Forest Restoration Efforts in Hawai’i.”
Pacific Science 71(4): 457–77.
Zhang, D. 2017. “A Coefficient of Determination for Generalized
Linear Models.”The American Statistician 71: 310–6.
Zhang, Q., F. Wu, L. Wang, L. Yuan, and L. Zhao. 2011.
“Application of PCA Integrated with CA and GIS in
Eco-Economic Regionalization of Chinese Loess Plateau.”
Ecological Economics 70: 1051–6.
Additional supporting information can be found online
in the Supporting Information section at the end of this
How to cite this article: Aho, Ken,
Susan Parsons, Antonio J. Castro, and Kathleen
A. Lohse. 2022. “Mapping Socio-Ecological Systems
in Idaho: Spatial Patterns and Analytical
Considerations.”Ecosphere 13(10): e4242. https://
ECOSPHERE 21 of 21