Content uploaded by Bryce Currey
Author content
All content in this area was uploaded by Bryce Currey on Jun 03, 2022
Content may be subject to copyright.
Journal of Biogeography. 2022;00:1–12. wileyonlinelibrary.com/journal/jbi
|
1© 2022 John Wiley & Sons Ltd.
Received: 8 December 2021
|
Revised: 15 April 2 022
|
Accepted: 19 April 2022
DOI: 10.1111/jbi.14391
RESEARCH ARTICLE
Large contribution of woody plant expansion to recent
vegetative greening of the Northern Great Plains
Bryce Currey1 | David B. McWethy2 | Nicholas R. Fox1 | E. N. Jack Brookshire1
Strapl ine: We find that la rge climatic cha nges and the abs ence of fi re have driven lar ge woody plant ex pansio n in the No rthe rn Great Plains , and that this ex pansio n of woody cover
compri se a large porti on of recent veget ative greening a cross this temp erate grassla nd ecosystem .
1Depar tment of Land Resources and
Environmental S cience s, Mont ana State
University, Bozeman, Mont ana, USA
2Depar tment of Eart h Sciences, Montana
State Universit y, Bozeman, Montana, USA
Correspondence
Bryce Currey, Department of Lan d
Resources and Environmental Sciences ,
Montan a State University, Bozeman ,
Montan a, USA .
Email: brycecurrey93@gmail.com
Funding information
U.S. Bureau of Land Management ;
Montan a Agricultur al Exp eriment Station;
Nationa l Science Foundation
Handling Editor: Simon Scheiter
Abstract
Aim: Extensive portions of high- latitude grasslands worldwide have recently experi-
enced increased vegetative productivity (i.e., greening) and have undergone a rapid
transition towards woody plant dominance via the process of woody plant expansion
(WPE). This raises the underlying question: To what degree are WPE and greening
spatiotemporally linked? Given that these vegetative changes are predicted to con-
tinue, we seek to understand how recent changes in vegetation extent and productiv-
ity have interacted under recent climate change and anthropogenic disturbance to
provide insights surrounding the future trajectory of temperate grasslands broadly.
Location: Northern Great Plains (NGP), North America.
Tax o n: Woody plants.
Methods: Greening was measured as the significant increase in three metrics between
2000 and 2019: leaf area index (LAI), annual maximum normalized difference vegeta-
tive index (NDVI) and annual mean NDVI. WPE was measured as the significant propor-
tional increase in percent tree cover change between 2000 and 2019 in grasslands. We
then examined these variables across a host of 26 potential driving variables.
Results: We show that average proportional greening increased by 0.2– 1.3% year−1
(depending on metric), and proportional WPE increased by 6.9% year−1 since 2000
acros s the NGP. Bot h change s ar e la rgel y dr iven by the abse nce of wildfire and chang-
ing climate. Furthermore, WPE was spatially coherent and positively associated with
a large component of recent greening, as revealed by their coupling across 34.1%–
40.6% of grassland area and as evidenced by the 9.7%– 19.7% of the variability in
greening explained by WPE.
Main conclusions: WPE and greening are spatiotemporally coupled across large por-
tions of the NGP. Under continued climate change and wildfire suppression, WPE and
greening are likely to continue across large swathes of grasslands globally. Furthermore,
our results show that using a single greening metric may be insufficient to capture the
large- scale vegetative changes such as the expansion of woody vegetation.
KEYWORDS
greening, leaf area index, normalized difference vegetation index, Northern Great Plains,
temperate grasslands, woody plant encroachment, woody plant expansion
2
|
CU RREY E t al.
1 | INTRODUC TION
Globally, large areas of temperate grasslands are undergoing vege-
tative shifts towards an increased prevalence of shrubs and trees,
hereafter referred to as “woody plant expansion” (WPE; Barger
et al., 2011; Criado et al., 2020; Stevens et al., 2017). Simultaneously,
large increases in vegetative photosynthetic activity or productiv-
ity (“greening”) have been documented across temperate grasslands
(Chen, Park, et al., 2 019; Chen, Parton, et al., 2019). Independently,
these two vegetative changes have large consequences for grassland
biophysical processes, biodiversity and biogeochemistry (Andersen
& Steidl, 2019; Brookshire et al., 2020; Eldridge et al., 2011; Ratajczak
et al., 2012; Welti et al., 2020). Furthermore, dynamic global vege-
tation models (DGVMs) have projected that both changes will con-
tinue in the future (Hufkens et al., 2016; Klemm et al., 2020; Reeves
et al., 2014; Shafer et al., 2015). Increased woody plant cover has
been hypothesized to be responsible for recent greening trends
across grassland and savanna ecosystems (Chen, Park, et al., 2019;
Deng et al., 2021; Xiao & Moody, 2005), but to what extent WPE
contributes to greening and to what degree these two phenomena
are spatiotemporally coupled in extent and driving mechanisms at
fine resolution remains unresolved.
The most common approaches for measuring greening use the
no rmali ze d dif fer enc e ve ge t at ion ind ex (N DV I) or lea f ar ea in dex (LAI ;
Piao et al., 2020). LA I pro vi des inf orm ati on on pho to syn the tic sur face
area (i.e., canopy structural complexity), whereas NDVI is biomass-
independent and a more direct proxy of photosynthetic activity.
NDVI is oft en me a su r ed us ing the ma xim u m va l ue ac ros s a tim e ser ies
(peak NDVI) or the mean value from a time series (mean NDVI). Peak
NDVI highlights the maximum photosynthetic active radiation and
subsequent maximum vegetative productivity (Myneni et al., 19 95),
whereas mean NDVI integrates annual vegetative production. Peak
NDVI is often preferred over other metrics as it is best at minimizing
error from cloud, reflectance, shadow and aerosol effects (Myneni
et al., 1995). However, changes in peak NDVI may be insufficient to
distin guish bet ween WPE dynamics and back ground greenness if the
expanding shrubs and trees are primarily evergreen (thus, maintain a
relatively constant NDVI year- roun d). In this cas e, mean annu al NDVI
wou ld capture th e el on ga te d NDVI signal across the year unde r WPE,
whereas the maximum values may not change. Furthermore, both
NDVI metr ics may be inadequate compared with L AI, which captures
the higher canopy complexit y of trees than herbaceous vegetation.
Although LAI and aboveground net primary productivity (ANPP) have
both been associated with higher woody cover (Barger et al., 2011;
Knapp et al., 2008), and NDVI and ANPP have been linked at coarse
scales (Reeves et al., 2020), how the se ass oc iati on s an d th ei r un de rly-
ing assumptions apply to WPE is uncertain.
That WPE and greening are coupled at local scales is reasonable,
but how both phenomena and their drivers are linked in space and
across time at the landsc ape scale remains largely unknown. WPE and
greening are most commonly attributed to increasing CO2, changing
climate, and shifts in disturbance regimes (Archer et al., 2017; Zhu
et al., 2016), yet disentangling these drivers and their interactions have
proven challenging. For example, elevated CO2 can promote greening
via directly stimulating photosynthesis and indirectly via increased
water- and/or nutrient- use efficiency (WUE/NUE), but the strength
of this effect can vary greatly among plant functional types (Feng
et al., 2015; Terrer et al., 2019) and may be constrained by climate
and nutrient availability (Brookshire et al., 2020; Frank et al., 2015).
Precipitation dictates the bioclimatic limits of woody cover (Scholtz
et al., 2018), and higher temperatures are strongly correlated with
increased shrub cover (Criado et al., 2020). However, increases in
temperature and precipitation also drive enhanced vegetative pro-
ductivity, particularly in semi- arid environments, irrespective of WPE
(Knapp et al., 2017; Zhu et al., 2016). Similarly, fire and herbivory can
provide a check on expanding tree cover (Fuhlendorf et al., 2009), yet
these disturbances may have negative, neutral or positive ef fect s on
greening (Frank et al., 1998; Geremia et al., 2 019). Therefore, it is cru-
cial to disentangle regionally specific climatic and disturbance drivers
of WPE and greening at fine spatial and temporal resolution to under-
stand the degree to which WPE contributes to greening.
We examined potential synchrony between WPE and greening
across the Northern Great Plains (NGP) of North America. The cur-
rent extent of WPE across the NGP has not been quantified, de-
spite anecdotal (Symstad & Leis, 2017) and historical photographic
evidence (Phillips et al., 1963) of WPE. Peak photosynthetic produc-
tivity has increased in recent decades (Brookshire et al., 2020) and
is predicted to continue in the future (Hufkens et al., 2016; Reeves
et al., 2014). In addition, the climate of the NGP is predicted to get
warmer and wetter (Conant et al., 2018; Polley et al., 2013), and
DGVMs have projected extensive climate- driven increases in woody
cover across NGP grasslands in the future (Klemm et al., 2020; Shafer
et al., 2015). Importantly, these simulations have not been con-
strained by historical obser vations of WPE. It is essential to disen-
tangle the recent trends and drivers of the potentially synchronous
vegetative changes to benchmark and constrain these interac tions
for simulating future vegetative changes accurately.
Here, we ask: (1) What is the magnitude and extent of WPE
and greening across the NGP? (2) What are the drivers of WPE and
greening across large topographic and climate ranges? (3) To what
degree are WPE and productivity synchronous across the land-
scape and through time? We use products from NASA's Moderate-
Resolution Imaging Spectroradiometer (MODIS) to explore changes
in peak NDVI, mean NDVI, LAI and percent tree cover between
2000 and 2019. We examine these vegetative changes against 26
possible climate, physical and disturbance driving variables using a
combination of random forest algorithms and linear regression.
2 | MATERIALS AND METHODS
2.1 | Study area
The NGP (as defined in this study) covers roughly 744,00 0 km2,
spans 15 degrees of longitude and 11 degrees of latitude (97.85–
112.65 W, 40.69– 50.03 N), and covers portions of five U.S. states
|
3
CURRE Y Et al.
and two Canadian provinces (Figure 1). Mean annual precipitation
(MAP) spans a wide gradient from 235 mm in the Nor thcentral-
western regions to 1214 mm in the Southeastern and mountainous
areas, and mean annual temperature (MAT) generally decreases with
increasing latitude from 10.4 to 0.8°C. Elevation in the NGP varies
from 368 to >2500 m above the sea level.
Acco rdi ng to the 2010 North American Land Ch ange Monitoring
System (NALCMS) 30 m landcover dataset, 54.8% of the NGP is
grassland, 28.4% is cropland, 7.6% is shrubland, 3% is forested (ei-
ther deciduous, needleleaf, or mixed), and 6.2% is other (i.e., urban,
water, and wetland; Figure 1). For our research, we restricted our
analyses to the combined grassland and shrubland layers (hereafter
“grasslands”; 62.3% or ~463,000 km2 of the NGP) as we were only
interested in tree cover changes in non- forested areas. Across most
of our study area, the primary trees implicated in WPE are Pinus
ponderosa Douglas ex C. Lawson and numerous Juniperus species,
including J. horizontalis Moench, J. Communis Linnæus, J. scopulo-
rum Sarg. and J. virginia Linnæus (Brown & Sieg, 1999; Symstad &
Leis, 2017). Some deciduous aspen (Populus tremuloides Michx) ex-
pansion has also been documented in the northern regions (Köchy
& Wilson, 2001).
2.2 | Data collection
All variables and their respective statistics can be found in Table S1.1
and are displayed in Figures S1.1– 1.5. Table S1.2 contains informa-
tion about the merged soils dataset. All analyses were conducted
using R 4.0.3. For a list of all packages used and the respective cita-
tions, see Table S3.1, and for a list of all dataset s used, see Table S3.2.
Finally, all data and analyses are publicly available at ht t p s: //d o i .
org/10.5061/dryad.z08kp rrdj.
2.2.1 | Tree cover, NDVI and LAI
We examined a t wo- decade time series (200 0– 2019) of WPE, NDVI
and LAI data. Peak and mean NDVI were calculated from NASA's
MODIS Vegetation Indices 16- day product (Version 6) at a 500 m
spatial scale. LAI was calculated as the mean yearly value from the
MODIS 8- day L AI/FPAR product (V6) at a 500 m resolution. WPE
is defined in this study as grassland pixels significantly increasing
in percent tree cover. Percent tree cover was obtained from the
MODIS Vegetation Continuous Fields 16- day product (V6) at a
250 m spatial scale and resampled using the median pixel value to
500 m. This product implements decision tree theory on a host of
auxiliar y variables (e.g., both MODIS and LANDSAT bands), which
allows the algorithm to capture vegetation cover at lower percent
tree cover values, although this product has been found to underes-
timate tree cover (Adzhar et al., 2022).
To determine significant (p < 0.05) changes in NDVI, LAI and
tree cover over time, we used a two- sided Mann– Kendal trend
test across the 20- year period. Only for pixels showing significant
changes, percent change was calculated relative to initial values,
and the inner 99% of the dat a was kept, removing exceptionally
large (e.g., ±4σ) outliers. A uniform 1% was added across all tree
cover values so that an increase from zero would not yield an in-
finite result but was subtracted for magnitude calculations and
mapping. We also created a ‘Greening Index’ to examine areas
that were increasing in all three greening metrics. This was done
by scaling each metric between zero and one and taking the sum,
excluding all areas not greening in all three metrics. As such, the
theoretic al maximum value of t he greening index is three, and val-
ues lower than three indicate some degree of greening from all
metrics. For greening and WPE statistics, we calculated the geo-
metric mean and s tanda rd deviation (SD) as the distr ibu tio ns we re
highly right- skewed.
2.2.2 | Climate data
Climate data were obtained from NASA's Daymet (1 km resolution
resampled to 500 m) gridded daily weather estimates. We deter-
mined the difference between the recent (2000– 2019) and historical
(1980– 1999) periods for MAP, MAT and mean seasonal climatolo-
gies. Relevant seasons were defined as spring (April, May, June;
FIGURE 1 Study area schematic for the nor thern Great Plains
(NGP). (a) The extent of the NGP within North America. Dark grey
indicates the five states and two providences partially occupied
by the NGP. (b) Seven land cover types within the NGP. Grassland
and shrubland layers were combined under the umbrella term
‘grassland.’ (c) The ‘grassland’ composite layer, covering roughly
two- thirds of the study area. The largest non- grassland area is
comprised of cropland
4
|
CU RREY E t al.
AMJ), summer (July, August, September; JAS) and winter (December,
Januar y, Februar y, March; DJFM). We estimated precipitation varia-
bility by calculating the temporal coefficient of variation (CV) of MAP
over the recent period. Finally, we included annual growing degree
days (AGDD) and change in AGDD using a base temperature of 10°C
at 1 km resolution data from the Coupled Model Intercomparison
Project (Phase 5; CMIP5).
2.2.3 | Physical and disturbance variables
To examine the effect of topography, we calculated elevation,
slope, aspect and roughness from the Shuttle Radar Topography
Mission digital elevation model dataset at 30 m resolution.
Roughness is calculated as the difference between the maximum
and minimum elevation across the cell and the eight neighbouring
cells (i.e., larger values indicate rougher terrain). To examine the
four cardinal directions of the aspec t as continuous variables, we
calculated the sine and cosine of aspect (in radians) to examine
the “northness” and “eastness,” respectively. Soils were obtained
from the National Resource Conser vation Service for soils within
the United States and from the National Soils Database of C anada
for Canadian soils and merged at the generalized order to create
seven soil classifications.
We examined three disturbance variables: nitrogen (N) deposi-
tion, cattle per km2 (a proxy for herbivory) and bur ned area. N depo-
sition (kg N ha−1 year−1) was obtained from the 1993 global product.
Ca ttl e per km2 was obt a in ed fo r 20 10 fro m the area l- we igh te d cat tle
density global product. Finally, total burned area across the study
period was obtained from MODIS and was converted into a binary
variable of burned/unburned.
2.3 | Random forest and statistical analyses
We examined the variable importance metric from random forest
models to determine the relative importance of each climate, topo-
graphic and disturbance variable on greening and tree cover changes.
Each variable's relative importance is calculated by randomly permu-
tating the data to induce randomness to each variable and compar-
ing these simulations to the observation- driven simulations (known
as ‘decreased accuracy’). The most significant variables have the
largest differences between observation- driven versus randomly
permutated runs. The error rate of random forest models is calcu-
lated as the average of all individual trees' misclassification rates
across the forest. Before modelling, feature selection determined
only the aspect variables (‘northness’ and ‘eastness’) non- relevant
and were thus removed from all analyses. We examined the rela-
tive importance of all 24 remaining drivers of L AI change, peak and
mean NDVI change, tree cover change, and 2019 tree cover. We also
examined the drivers of 2000 tree cover for variables with histori-
cal data (1980– 1999, n = 15). We also assumed constant cattle den-
sity, N deposition, and soil type across time. Finally, we calculated
the variance explained by the predictor variables for all models
(i.e., pseudo- R2). All variables were resampled using median values
to the resolution of the greening and tree cover variables (500 m
resolution), on which we used 500 classification trees and four
variables at each candidate node. These analyses were performed
on the Hyalite High- Performance Computing System operated by
University Information Technology Research Cyberinfrastructure at
the Montana State Universit y.
Because random forest models cannot determine how individual
variables interact with the response variables (i.e., the relationship's
direction), we examined all 24 variable's responses against WPE,
peak NDVI, mean NDVI and LAI using ordinary least squares (OLS)
regression models. Before modelling, all variables were standard-
ized (z- score) to examine the effect size on a uniform scale. Effect
sizes (slopes from the bivariate OLS models) were combined with
the random forest rank importance analysis to examine each vari-
able's importance score and direction concurrently. As such, a higher
variable rank indicates higher importance, and a larger z- score in-
dicates a steeper directional slope. It should be noted that a steep
relationship (z- score far from the zero line) does not necessitate high
variable importance, nor does higher rank importance necessitate a
steep slope. For instance, a high importance value but a small z- score
indicates a strong spatial coupling but a less steep relationship.
Similarly, we conducted bivariate analyses examining all greening
metrics versus tree cover change to examine the variance explained
within each greening variable by WPE. We then examined the cor-
relation of WPE and greening variables in two ways: first, we exam-
ined each variable's spatial correlation using Tjøstheim's coefficient
(Α) and, second, we examined Spearman's rank correlation (ρ) be-
tween the drivers of each response variable (both bounded between
1 and −1). Tjøstheim's coefficient examines the spatial similarity of
the response variables, whereas Spearman's coefficient examined
the correlation in both z- score and importance of all 24 driving vari-
ables of each response variable. Said in another way, Tjøstheim's
A examines the degree of co- occurrence, whereas Spearman's
ρ examines the similarity between each vegetative change's driving
variables. For the rank correlation analysis between tree cover in
the year 20 00 and other response variables, we only examined the
correlation bet ween the 15 variables used in the 2000 tree cover
random forest model.
3 | RESULTS
3.1 | Climate, topography and disturbance
3.1.1 | Climate
Recent (2000– 2019) MAP increased by 24.5 mm on average over
the entire study area compared with the 1980– 1999 climatology
(Table S1.1 and Figure S1.2) but was regionally heterogeneous.
For instance, some areas experienced large declines (decreased
<100 mm; primarily in the northern and central regions), yet others
|
5
CURRE Y Et al.
experienced substantial increases (increased >200 mm; primarily in
the southern region). The greatest seasonal precipitation increases
were in the spring, with winter and summer mean precipitation
barely changing on average. MAT also increased on average by 0.1°C
(Table S1.1 and Figure S1.3), with the largest average increases in the
summ er mont hs (~0.7° C) . Conse qu ent ly, AGD D inc re ased ac ro ss th e
entirety of the NGP, with the largest increases in the southwestern
areas (Figure S1.4).
3.1.2 | Disturbance
Over the 20- year study period, ~2.7% of grassland pixels experi-
enced fire (0.135% probability yr−1 across all pixels), translating to
~20,000 km2 of burned area (Table S1.1 and Figure S1.5). N deposi-
tion across the region was relatively homogenous after being up-
scaled yet with an evident east– west gradient from a high of 7.4 to a
low of 1.8 kg N ha−1 year−1, and well within the range of in situ stud-
ies from the region (Symstad et al., 2 019). Finally, the average distri-
bution of cattle across the NGP was 12.5 head km−2 , but regions in
the southern NGP showed densities as high as ~137 head km−2.
3.2 | Greening and WPE
3.2.1 | Extent of greening
We document extensive but variable greening across most NGP
grasslands (Table 1; Figure 2). LAI increases were the most exten-
sive, occurring across 76.0% of grassland pixels. L AI- based greening
also had the largest proportional 20- year changes of 26.8 (2.1)% on
average, or ~1.3% year−1 . Nex t, peak NDVI increased significantly
across 62.7% of grasslands, with a magnitude of 9.9 (1.7)%, or ~0.5%
year−1. Finally, mean NDVI increased in over half (57.3%) of grassland
pixels by 4.2 (1.9)%, or ~0.2% year−1. Synchronous increases in all
three metrics (i.e., the ex tent of the greening index) occurred across
30.6% of NGP grassland area, and all three metrics had a moderately
high Tjøstheim's coefficient (Α ≥ 0.42; Table S2.1). We note slightly
higher greening in shrubland versus grassland pixels, although the
magnitude of changes was similar (Table S2.2).
3.2.2 | Driving variables of greening
All random forest classification models for greening per formed ex-
ceptionally well with error rates for LAI, peak NDVI and mean NDVI
changes of 1.43%, 0.44% and 1.49%, respectively. Despite strong
model performance, the 24 predictor variables only explained 32.7%,
24.4% and 53.9% of each greening metric's respective variance.
TAB LE 1 Vegetative change metric statistics for NGP grasslands.
Inc is the frac tion of total area undergoing respec tive vegetative
increase, decrease (Dec) represents the fraction decreasing.
x
inc
(sd) is the geometric average (geometric sd) magnitude of increases
in the respective increasing proportion.
x
dec (sd) is the geometric
average (geometric sd) magnitude of decreases in the respective
decreasing proportion
Proportion
of NGP
grasslands Inc
xinc
(SD
)
Dec
xdec
(SD
)
Tree Cover Δ48.0 46.8 137.8 (2.3) 1.2 −17.7 (3.5)
LAI Δ77.6 76.0 26.8 (2.1) 1.6 −3.9 (4.3)
Peak NDVI Δ63.0 62.7 9.9 (1.7) 0.3 −1.9 (4.6)
Mean NDVI Δ58.9 5 7. 3 4.2 (1.9) 1.6 −1.7 (4.1)
FIGURE 2 NGP grassland (shaded grey) proportional greening changes from 20 00 to 2019 measured as (a) L AI changes, (b) peak NDVI
changes, and (c) mean NDVI changes. White areas represent non- grasslands (i.e., forested, cropland, urban, wetland)
6
|
CU RREY E t al.
Roughness, slope and elevation were negatively associated with
increasing LAI (Figure 3a), indicating that L AI- based greening was
largest on smooth, flat topographies. High precipitation variability
and absence of fire were also strong predictors. Precipitation driv-
ers of LAI greening were seasonally dependent, although increas-
ing MAP was correlated with increasing LAI. Warming summer
temperatures were also positively associated with LAI. For peak
NDVI (Figure 3b), greening was positively associated with higher
precipitation variability and increasing precipitation (save for winter
precipitation). Peak NDVI- based greening was strongly associated
with smoother topographies and low cattle density. Interestingly,
fire was a poor predictor of peak NDVI- based greening. Conversely,
the absence of fire was the strongest predictor of mean NDVI in-
creases (Figure 3c). The highest mean NDVI- based greening was as-
sociated with smoother topographies and lower elevations. Finally,
year- round wetting, albeit variably, and warming temperatures were
strong driving variables of higher mean NDVI greening.
Overall, smoother topographies and the absence of fire (save
for peak NDVI) were strong predic tors of greening. Seasonally de-
pendent changes in precipitation and temperature were apparent, in
which warmer and wetter summers, but cooler and drier winters, led
to the highest proportional greening, apart from mean NDVI, which
was driven by a year- round increase in precipitation. Higher precip-
itation variability was also associated with the largest proportional
changes of all three metrics. Lower cattle densities (i.e., herbivory)
were a moderately strong predictor of NDVI- based greening. N
deposition, AGDD, baseline climatology and soil variables were less
important driving variables for greening metrics. Despite climatic
differences, the driving variables of all greening metrics had high
Spearman correlations (ρ ≥ 0.75; Table S2.1).
3.2.3 | Extent of tree cover and WPE
Initial (year 2000) median percent tree cover in NGP grasslands
was 1.6% and generally very low across the entire region (Table 2,
Figure 4a). Nonetheless, close to half (46.8%) of grassland pixels
observed a significant increase in percent tree cover over the two
decades (Table 1, Figure 4b). As a result, proportional tree cover
changes were large, with the average increase at 137.8 (2.3)%, or
~6.9% year−1, compared to tree cover in 20 00. These large propor-
tional changes increased the median percent tree cover to 3.4% by
2019, a 1.8% cover increase over the 20- year period (0.09% cover
year−1; Figure S2.1a). Interestingly, the inner quartile range in 20 00
FIGURE 3 Relative importance from random forest models in descending order of importance for the three greening indices: (a) LAI
increases, (b) peak NDVI increases, and (c) mean NDVI increases. On the x- axis, standardized ( z- score) effect size coefficients from ordinary
least- squares regression. Positive values indicate a significant positive relationship with the greening metric; negative values indicate a
significant negative relationship. Dots far ther away from the zero line indicate a steeper slope. The fire product was converted to a numeric
variable, with positive one indicating fire presence and zero indicating absence. Soil is on the zero line as it was kept a categorical variable.
Grey circles signif y topographic variables, gold circles signify disturbance variables, and blue circles signify climate variables
Year 2000 Year 2019
Q1
x
Q3 IQR Q1
x
Q3 IQR
Tree Cover 1.0 1.6 2.7 1.7 1.8 3.4 6.5 4.7
LAI 0.28 0.38 0 . 51 0.23 0.48 0.50 0.65 0 .17
Peak NDVI 0.37 0.46 0.58 0.21 0. 51 0.60 0.71 0.20
Mean NDVI 0.22 0.26 0. 31 0.09 0.26 0.30 0.34 0.08
TAB LE 2 Vegetative metrics for NGP
grasslands. Q1 and Q3 are the inner
quartiles for the respective vegetative
metric, IQR is the inner quartile range, and
x
is the geometric average
|
7
CURRE Y Et al.
was 1.7% but increased to 4.7% over the two- decade period, sig-
nalling larger changes at higher tree cover densities (Table 2).
Nonetheless, 2000 and 2019 tree cover had a large degree of spatial
similarity (A = 0.79).
3.2.4 | Driving variables of tree cover and WPE
Models of both static percent tree cover and changing percent tree
cover performed well, with an error of 1.6% for the year 2000 tree
cover model, 0.9% for the 2019 tree cover model and 1.8% error for
the changing tree cover model. Based on the random forest models,
the 15 driving variables of 2000 tree cover explained 54.5% of the
variation, and the 24 predic tors for 2019 tree cover and tree cover
change explained 39.3% and 32.8% of the variability, respectively.
Initial tree cover was concentrated in areas with high topographic
roughness and steeper slopes, with low precipitation variabilit y and
high winter precipitation (Figure 5a). Drivers of tree cover in 2000
and 2019 were highly correlated (ρ = 0.79), although many of the
changing climate variables became important to tree cover in 2019
(Figure S2 .1b).
In contrast to year 2000 tree cover, tree cover change was
primarily driven by high precipitation variability, the absence of
fire and other disturbances, and occurred on smoother topog-
raphies with lower slopes (Figure 5b). All changing precipitation
variables were positively associated with WPE, indicating that
FIGURE 4 NGP grassland (shaded grey) tree cover and tree
cover changes. (a) Percent tree cover from MODIS for the year
2000 and (b) the statistically signific ant proportional change
between 2000 and 2019. White areas represent non- grasslands
(i.e., forested, cropland, urban, wetland)
FIGURE 5 Relative importance from random forest models
in descending order of importance for (a) percent tree cover
(year 200 0) and (b) tree cover percent increase. On the x- a x i s ,
standardized (z- score) effect size coefficients from ordinary least-
squares regression. Positive values indicate a significant positive
relationship with the tree cover metric; negative values indicate a
significant negative relationship. Dots far ther away from the zero
line indic ate a stronger directional relationship. The fire product
was converted to a numeric variable, with positive values indicating
fire presence. Soil is on the zero line as it was kept a categorical
variable. Grey circles signify topographic variables, gold circles
signify disturbance variables, and blue circles signify climate
variables. Percent tree cover (a) has only variables that are relevant
to the year 20 00
8
|
CU RREY E t al.
the most significant proportional changes occurred where it has
gotten wetter year- round. Finally, tree cover in 2000 and tree
cover change had a low Tjøstheim coefficient (A = −0.11), and
their respective drivers had a ρ = −0.52 Spearman correlation
(Figure S2.1). This indicates that areas with the highest rates of
WPE were not where tree cover was initially highest, nor were
they associated with the same driving variables, emphasizing the
point that tree cover change is occurring primarily as expansion
rather than infilling.
3.3 | Concurrent greening and WPE
Both WPE and greening occurred across vast areas of NGP grass-
lands. Here, we document that these changes were often concurrent
and influenced by similar drivers. While 6.3%– 12.7% of WPE and
23.0%– 35.5% of greening occurred independent of one another (de-
pending on which greening metric is examined in conjunction with
WPE; i.e., WPE and L AI, WPE and peak NDVI, or WPE and mean
NDVI), the largest proportion of vegetative changes were concur-
rent (Table 3).
This was particularly true for LAI and percent tree cover, which
increased simultaneously across 40.6% of grasslands (Table 3,
Figure S2.2a). A doubling of percent tree cover corresponded with a
5.2 ± 0.04% increase in LAI (p < 0.0 01), and WPE explained 11.1% of
the variability in L AI greening. Indeed, we found a relatively strong
spatial correlation (A = 0.49) and a ver y strong correlation bet ween
drivers of WPE and LAI (ρ = 0.67; Table S2.1).
We found similar results for NDVI variables; 34.1% of all grass-
lands experienced increases in peak NDVI and WPE (Table 3,
Figure S2.2b), and 34.2% underwent mean NDVI- based greening
and WPE ( Table 3, Figure S2.2c). For every doubling of percent tree
cover, peak NDVI increased by 1.28 ± 0.01% (p < 0.001) and mean
NDVI increased by 0.91 ± 0.01% (p < 0.001). WPE explained 9.7% of
the variation in peak NDVI- based greening and 19.7% of mean NDVI
increases. Interestingly, the Spearman correlation between drivers
of WPE and peak NDVI was the lowest (ρ = 0.56), suggesting that
even though WPE and peak NDVI co- occurred across large areas,
the primary drivers of each differed. Conversely, drivers of mean
NDVI and WPE were highly correlated (ρ = 0.76), and both peak
and mean NDVI had similar Tjøstheim's coefficients (A = 0.43 and
A = 0.44, respectively, Table S2.1).
Finally, we document that 19.0% of NGP grasslands have simul-
taneously undergone increases in all three greening metrics and
WPE (Figure 6a; Table 3). Green ing and WPE were strongl y and pos-
itively associated; a doubling of woody plant cover corresponded
to a proportional 15.0% increase in the greening index (p < 0.0 01;
Figure 6b). Although WPE and the greening index were mutually ex-
clusive across large portions of grasslands (Table 3), WPE explained
17.0% of the variabilit y in the greening index across the NGP.
TABLE 3 Percent of NGP grasslands undergoing synchronous
versus mutually exclusive WPE and greening categorized by
greening metric. As an example, 40.6% of NGP grasslands
increased in LAI and tree cover synchronously. Conversely, 35.5%
of the NGP experienced only L AI- based greening and 6.3% only
WPE. Examining the greening index and WPE, we see that 19% of
NGP grasslands increased in all three greening metrics and WPE
Greening metric with WPE Synchronous
Mutually
exclusive
Greening
only
WPE
only
LAI and WPE 40.6 35.5 6.3
Peak NDVI and WPE 34.1 28 .7 12.7
Mean NDVI and WPE 34.2 23.0 12.6
Greening Index and WPE 19.0 11.6 13 .5
FIGURE 6 Greening and WPE across NGP grasslands (shaded
grey; white areas represent non- grasslands). (a) Mapped changes.
WPE only represents areas where only tree cover has increased.
Greening only represents pixels where only greening has occurred.
Greening is represented here as the ‘greening index,’ where
peak NDVI, mean NDVI and LAI have increased simultaneously.
Concurrent Δ represents where greening and WPE have
co- occurred. (b) Relationship between cells undergoing greening
and concurrent tree cover increase. (c) Distribution of greening
index values
|
9
CURRE Y Et al.
4 | DISCUSSION
We document a strong coupling between WPE and vegetative
greening at high- resolution (~500 m) across broad disturbance, cli-
mate and topographic gradients within one of the last remaining rel-
atively intact high latitude grasslands globally. Significant increases
in percent tree cover occur on smooth topographies and are primar-
ily driven by the absence of fire and increasing temperature and
precipitation. We demonstrate that a doubling in proportional tree
cover drives a concurrent 15% proportional increase in our integra-
tive greening index and that these synchronous vegetative changes
occur across roughly one- fifth (roughly 90,00 0 km2) of all NGP
grasslands. Indeed, WPE alone account s for 17% of the variability in
the greening in dex. Comp arati vely, all 24 other climatic, to pographic
and disturbance variables combined explained 17.4%– 50.7% of the
variance, depending on the greening metric. These results highlight
that WPE is a large component of vegetative greening in the NGP
and imply that it may be an underappreciated component of green-
ing trends in other semi- arid biomes.
Our results provide necessary observations to benchmark and
parameterize models of vegetation change over the nex t century.
While our results point to a more tree- covered NGP in the future,
observed rates of change are much lower than required to result in
a fully forested NGP by 2100 suggested by some DGVM projections
(Shafer et al., 2015). Despite large proportional changes in percent
tree cover, absolute changes of percent tree cover across the NGP
were small. Median percent tree cover increased by 1.8% over the
two- decade period (~0.1% year−1), with the upper quartile showing
a slightly larger increase of 3.8% (~0.2% year−1). A linear ex trapola-
tion of these trends suggests that WPE rates would have to increase
more than fivefold to obtain full woody cover by end- of- century
across the NGP (i.e., median tree cover of at least 50%).
Our analyses of recent WPE point to critical interactions be-
tween topography, fire and climate change in explaining the spatial
distribution of this vegetative change. Early studies of vegetation
distributions in the NGP observed that P. ponderosa and Juniperus
spp. were historically restricted to fire- protected rocky scarps and
uplands (Wells, 1965). In alignment with these findings, we show
that initial (year 200 0) tree cover was highest in rough terrain, yet
the highest rates of WPE occurred on smooth terrain in the absence
of fire. Furthermore, drivers of established tree cover in 2000 and
subsequent vegetative changes differed, indicating that what once
allowed for the establishment of trees is distinct from what is as-
sociated with their expansion and greening today. Although stand
level stem dynamics cannot be determined at the spatial resolution
used in this study, we find evidence that past tree cover and pres-
ent WPE are negatively associated across space. This indicates that
the majority of increasing percent tree cover is not where percent
tree cover is currently the highest, providing evidence that trees are
expanding. Facilitating this expansion was very low fire occurrence
across the NGP, as less than 3% of NGP grasslands experienced fire
over the 20- year period. Similar to the result of active fire suppres-
sion, the dominant fire regime across much of the NGP has shifted
from one with a relatively frequent fire return inter val (FRI) of <30–
50 years to one of infrequent and small fires (e.g., > 50- year FRI;
(Reid & Fuhlendorf, 2011; Umbanhowar, 1996). This tren d could sus-
tain WPE and greening further into the 21st century. It is also possi-
ble that, under a warming climate, higher woody fuel loads resulting
from continued WPE could trigger novel veget ation- fire feedbacks
characterized by more frequent high- intensity fire activity (Moritz
et al., 2012).
Enhanced vegetative growth and WPE were strongly associated
with spring and winter precipitation increases, with the highest
proportional increases occurring in more arid areas characterized
by high long- term precipitation variability and lower baseline pro-
ductivity and tree cover. In fact, precipitation CV ranked as a more
important driver than any other climate variables for all greening
and tree cover change measures. These results are consistent with
high precipitation sensitivity and higher marginal gains in vegetation
growth in response to increasing water supply in more arid areas
(Hoover et al., 2021; Hufkens et al., 2016; Ritter et al., 2020). The
NGP has been getting wetter over the last few decades (Bromley
et al., 2020; Polley et al., 2013) and is predicted to continue to get
wetter in the future (Conant et al., 2018), a trend that may sustain
increases in vegetative activity. At the same time, enhanced mois-
ture stress associated with higher evaporative demand and con-
tinued declines in summer precipitation could potentially dampen
greening and WPE trends.
Ou r d oc u me nte d gree nin g r at es of 0. 2%– 1.3 % yea r−1 align with
climate- driven model simulations showing 20%– 100% increases
in NGP vegetative productivity over the next century (Hufkens
et al., 2016; Reeves et al., 2014), but our results further suggest
that shif ts in vegetation composition and phenology may under-
lie some of these changes. While WPE has contributed dispro-
portionately to greening relative to its spatial extent, enhanced
vegetative productivity occurred across 23.0%– 35.5% of NGP
grasslands in the absence of WPE. This is likely in par t due to an
elongated growing season, consistent with the warming tempera-
tures we obser ve in this study. Nonetheless, the fact that LAI and
mean annual NDVI were strongly correlated with WPE suggests
that inc reases in evergreen co nifer cover enhance greening due to
year- long photosynthetic activity and increased biomass. In con-
trast, peak NDVI had the lowest correlation with WPE, suggest-
ing peak NDVI is largely driven by herbaceous- driven greening
(Brookshire et al., 2020). Similarly, that peak NDVI had the lowest
variance explained by climatological changes in our models may
indicate t hat peak NDVI is mo re sensitive to inter an nu al - to- an nu al
climate variability than mean NDVI or LAI, both of which were
better explained by our random forest models. These differences
in model responses across greening variables highlight the impor-
tance of examining multiple greening metrics to capture the var-
ious aspects of phenology when examining greening phenomena
of large- scale vegetative changes (Wood et al., 2021).
In addition to the elongation of photosynthetic activity associ-
ated with WPE, the increasing abundance of conifers raises import-
ant questions about potential shifts in plant resource acquisition
10
|
CU RREY E t al.
and use- efficiency that may affect future greening and WPE. Foliar
isotope analysis shows that WUE of grassland vegetation across the
NGP has increased by 37% on average since the 1960s in response
to atmospheric CO2 enrichment, while N availability has declined
(Brookshire et al., 2020). Although declining N availabilit y suggest s
a potential bottleneck to future greening, it also indicates enhanced
NUE given sustained productivity. How coniferous WPE may af-
fect these trajec tories is less certain. For example, P. ponderosa and
Juniperus spp. may have advantages over grasses in access to soil
nutrients via dif ferences in rooting depth or mycorrhizal nutrient
acquisition strategies (Policelli et al., 2019) that could sustain tree
expansion and greening, particularly under elevated CO2 (Terrer
et al., 2019).
Our machine learning approach highlights the challenges of
using empirical modelling to benchmark process- based DGVM
projections of vegetative changes. A not able disadvantage is a re-
liance on accurate and heterogeneous spatial data. For inst ance,
atmospheric CO2 enrichment likely drives some of the patterns
we obser ve through direc t stimulation of photosynthesis and/or
indirect enhancement of WUE and NUE (Brookshire et al., 2020),
but we did not account for CO2 due to its spatial homogeneity. In
addition, although the MODIS VCF tree cover product has been
validated across 442 WPE sites (Deng et al., 2021), this product
has been shown to consistently underestimate tree cover by 0%–
18% (Adzhar et al., 2022). Therefore, our estimates of WPE are
conservative as we cannot capture trees too small or young to
be included in the algorithm. Indeed, the WPE rate we document
(~0.1% cover year−1) is lower than field estimates in other western
North American grasslands (0.5%– 0.6%, Barger et al., 2011), but
closer to global estimates that use the same product (0.3% cover
year−1; Deng et al., 2021).
Our findings document the pace of vegetation change and its
dominant controls across one of the last remaining relatively intact
grasslands worldwide and provide insight into alternative future
ecosystem states. We have shown that the greening of the NGP
is driven by interac tions between large- scale vegetative shift s and
cooccurring photosynthetic enhancement from established herba-
ceous veget ation. While recent WPE has contributed to enhanced
C sequestration, WPE is also associated with a host of ecosystem
and socio- economic tradeoffs, including loss of grassland biodi-
versity, transformation of biogeochemical cycles, and changes to
wildlife habitat (Barger et al., 2011; Epstein et al., 2021; Symstad
& Leis, 2017; Wilcox et al., 2018). Fur thermore, the combination of
higher conifer densities and a warmer climate dramatically increases
the probability of large wildfires (McWethy et al., 2019; Morit z
et al., 2012). As such, assessment of these large- scale and interac-
tive vegetative changes should be disassociated from values- based
ideologies regarding whether woody encroachment and greening
are entirely beneficial or detrimental processes. Instead, it is im-
portant to examine the complex tradeoffs between these vegeta-
tive changes to guide understanding, evaluation and management
of future vegetative states and functions under projec ted climate
change.
ACKNOWLEDGEMENTS
We thank Simon Scheiter, Anping Chen and an anonymous re-
viewer for their insightful improvements to this manuscript. This
work was supported by the National Science Foundation, Award
Numbers: OIA- 1632810 (ENJB) and BCS 1832486 (DBM), the
Montana Agricultural Experiment Station and the Department
of Interior, Bureau of Land Management award L16AS0 0082 to
ENJB. No permit s were required for this research.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
DATA AVAIL AB I LI T Y STATE MEN T
All large datasets are in .csv format and are available on datad ryad.
org at 10.5061/dryad.z08kprrdj. All other datasets, code, and fig-
ures are publicly available on this project's repository on github.com
at 10.5281/zenodo.6473246.
ORCID
Bryce Currey https://orcid.org/0000-0001-9794-9906
David B. McWethy https://orcid.org/0000-0003-3879-4865
E. N. Jack Brookshire https://orcid.org/0000-0002-0412-7696
REFERENCES
Adzhar, R., Kelley, D. I., Dong, N., George, C., Torello Raventos, M.,
Veenendaal, E., Feldpausch, T. R., Phillips, O. L., Lewis, S. L., Sonké,
B., Taedoumg, H., Schwantes Marimon, B., Domingues, T., Arroyo,
L., Djagbletey, G., Saiz, G., & Gerard, F. (2022). MODIS vegetation
continuous fields tree cover needs calibrating in tropical savan-
nas. Biogeosciences, 19 (5), 1377– 1394. https://doi.org/10.5194/
b g - 1 9 - 1 3 7 7 - 2 0 2 2
Andersen, E. M., & Steidl, R. J. (2019). Woody plant encroach-
ment restructures bird communities in semiarid grasslands.
Biological Conservation, 240, 108276. http s://doi.org /10.1016/ j.
biocon.2019.108276
Archer, S. R., Andersen, E. M., Predick, K. I., Schwinning, S., Steidl, R.
J., & Woods, S. R. (2017). Woody Plant encroachment : Causes and
consequences. In D. D. Briske (Ed.), Rangeland systems: Processes,
management and challenges (pp. 25– 84). Springer International
Publishing. h t t p s : / / d o i . o r g / 1 0 . 1 0 0 7 / 9 7 8 - 3 - 3 1 9 - 4 6 7 0 9 - 2 _ 2
Barger, N. N., Archer, S. R., Campbell, J. L ., Huang, C., Morton, J. A., &
Knapp, A. K. (2011). Woody plant proliferation in north American
drylands: A synthesis of impacts on ecosystem carbon balance.
Journal of Geophysical Research Biogeosciences, 116 , G0 0K07.
htt ps://doi.org/10.10 29/2010J G 00 1506
Bromley, G. T., Gerken, T., Prein, A . F., & Stoy, P. C. (2020). Recent trends
in the near- surface climatology of the northern north American
Great Plains. Journal of Climate, 33(2), 461– 475. h t t p s :// d oi .
o r g / 1 0 . 1 1 7 5 / J C L I - D - 1 9 - 0 1 0 6 . 1
Bro ok sh ire, E. N. J. , Stoy, P. C., Cur rey, B., & Fin ney, B. (2 020). The gre en-
ing of the northern Great Plains and its biogeochemical precursors.
Global Change Biology, 26(10), 5404– 5 413. http s://doi.o rg /10.1111/
gcb.15115
Brown, P. M., & Sieg, C . H. (1999). Histori cal variability in fire at the pon-
derosa pine— Northern Great Plains prairie ecotone, southeastern
Black Hills, South Dakota. Écoscience, 6(4), 539– 547. h t tp s : //d o i.
org /10.1080/11956 86 0.199 9.11682563
Chen, C., Park, T., Wang, X., Piao, S., Xu, B., Chaturvedi, R. K., Fuchs,
R., Brovkin, V., Ciais, P., Fensholt, R., Tømmervik, H., Bala, G., Zhu,
|
11
CURRE Y Et al.
Z., Nemani, R. R., & Myneni, R. B. (2019). China and India lead in
greening of the world through land- use management. Nature
Sustainability, 2(2), 122– 129. http s://doi.org /10.1038/s4189
3 - 0 1 9 - 0 2 2 0 - 7
Chen, M., Parton, W. J., Har tman, M. D., Grosso, S. J. D., Smith, W.
K., Knapp, A. K., Lutz, S., Derner, J. D., Tucker, C . J., Ojima, D. S.,
Volesky, J. D., Stephenson, M. B., Schacht, W. H., & Gao, W. (2019).
Assessing precipitation, evapot ranspiration, and NDVI as controls
of U.S. Great Plains plant production. Ecosphere, 10(10), e02889.
htt ps://doi.org/10.10 02 /ecs2.2889
Conant, R. T., Kluck, D., Anderson, A ., Badger, A., Boustead, B . M.,
Derner, J., Farris, L., Hayes, M. J., Livneh, B., McNeeley, S., Peck,
D., Shulski, M. D., & Small, V. A . (2018). Northern Great Plains. In
Impacts, risks, and adaptation in the United States: Fourth National
Climate Assessment, Volume II (pp. 941- 986). U.S. Global Change
Research Program. https://doi.org/10.7930/NCA4.2018.CH22
Criado, M. G., Myers- Smith , I. H., Bjorkman, A. D., Lehmann, C. E. R.,
& Stevens, N. (2020). Woody plant encroachment intensifies
under climate change across tundra and savanna biomes. Global
Ecology and Biogeography, 29(5), 925– 943. ht tp s://doi.o rg /10.1111/
geb.13 072
Deng, Y., Li, X., Shi, F., & Hu, X. (2021). Woody plant encroachment
enhanced global vegetation greening and ecosystem water- use
efficiency. Global Ecology and Biogeography, 30(12), 2337– 2353.
https ://doi.or g/10.1111/geb.13386
Eldridge, D. J., Bowker, M. A ., Maestre, F. T., Roger, E., Reynolds,
J. F., & Whitford, W. G. (2011). Impac ts of shr ub encroach-
ment on ecosystem structure and functioning: Towards a
global synthesis. Ecology Letters, 14(7 ), 709– 722. ht t p s :// do i .
org /10.1111/j .1461- 0248 .2 011. 016 30 .x
Epstein, K., Wood, D., Roemer, K., Currey, B., Duf f, H., Gay, J., Goemann,
H., Loewen, S., Milligan, M., Wendt, J., Brookshire, E. N. J., Maxwell,
B., McNew, L., McWethy, D., Stoy, P., & Hagger ty, J. (2021). Toward
an urgent yet deliberate conser vation strategy: Sustaining social-
ecological systems in rangelands of the nor ther n Great Plains,
Montana. Ecology and Society, 26(10). https://doi.org/10.5751/ES-
1 2 1 4 1 - 2 6 0 1 1 0
Feng, Z., Rütting, T., Pleijel, H., Wallin, G., Reich, P. B., Kammann, C.
I., Newton, P. C. D., Kobayashi, K., Luo, Y., & Uddling, J. (2015).
Constr aints to nitrogen acquisition of terrestrial plants under el-
evated CO2. Global Change Biology, 21(8), 3152– 3168. ht t ps : //d o i .
org /10.1111/gcb.12 938
Frank, D. A., McNaughton, S. J., & Tracy, B. F. (1998). The ecology of the
Earth's grazing ecosystems. Bioscience, 48(7), 513– 521. ht t p s : //d o i.
org/10.2307/1313313
Frank, D. C., Poulter, B., Saurer, M., Esper, J., Huntingford, C., Helle,
G., Treydte, K., Zimmermann, N. E., Schleser, G. H., Ahlström, A.,
Ciais, P., Friedlingstein, P., Levis, S., Lomas, M., Sitch, S., Viovy, N.,
Andreu- Hayles, L., Bednarz, Z., Berninger, F., & Boettger, T. (2015).
Water- use efficiency and transpiration across European forests
during the Anthropocene. Nature Climate Change, 5(6), 579– 583.
https://doi.org/10.1038/nclim ate2614
Fuhlendorf, S. D., Engle, D. M., Kerby, J., & Hamilton, R. (2009). Pyric
herbivory: Rewilding landscapes through the recoupling of fire
and grazing. Conservation Biology, 23(3), 588– 598. h t t ps : //d o i .
org /10.1111/j .1523- 1739.20 08.01139.x
Geremia, C., Merkle, J. A., Eacker, D. R., Wallen, R. L., White, P. J.,
Hebblewhite, M., & Kauf fman, M . J. (2019). Migrating bison engi-
neer the green wave. Proceedings of th e National Acad emy of Sciences
of the USA, 116(51), 25707– 25713. ht tps://doi.o rg/10.1073/
pnas.19137 83116
Hoover, D. L., Lauenroth, W. K., Milchunas, D. G ., Porensky, L. M.,
Augustine, D. J., & Derner, J. D. (2021). Sensitivity of productivity
to precipitation amount and pattern varies by topographic posi-
tion in a semiarid grassland. Ecosphere, 12(2), e03376. h t t p s: //d o i .
org /10.1002/ec s2.3376
Hufkens, K ., Keenan, T. F., Flanagan, L. B., Scott, R . L., Bernacchi, C. J.,
Joo, E., Brunsell, N. A., Verfaillie, J., & Richardson, A. D. (2016).
Productivity of north American grasslands is increased under fu-
ture climate scenarios despite rising aridity. Nature Climate Change,
6(7), 710– 714. https://doi.org/10.1038/nclim ate2942
Klemm, T., Briske, D. D., & Reeves, M. C. (2020). Potential natural vegeta-
tion and NPP responses to future climates in the U.S. Great Plains.
Ecosphere, 11(10), e03264. https://doi.org/10 .100 2/ecs2. 3264
Knapp, A. K., Briggs, J. M., Collins, S. L., Archer, S. R ., Bret- Harte, M.
S., Ewers, B. E., Peters, D. P., Young, D. R., Shaver, G. R., Pendall,
E., & Clear y, M. B. (2008). Shrub encroachment in nor th Americ an
grasslands: Shifts in growth form dominance rapidly alters control
of ecosystem carbon input s. Global Change Biology, 14(3), 615– 623.
https ://doi.or g/10.1111/ j.1365- 2486. 2007.01512.x
Knapp, A . K., Ciais, P., & Smith, M. D. (2017). Reconciling inconsis tencies
in precipitation– productivity relationships: Implications for climate
change. New Phytologist, 214(1), 41– 47. ht tps://doi .org /10.1111 /
nph .14381
Köchy, M., & Wil son , S. D. (2 001). Nit rogen deposition and fo res t exp an-
sion in the northern Great Plains. Journal of Ecology, 89(5), 807– 817.
https://doi.org/10.1046/j.0022- 0477.2001.00600.x
McWethy, D. B., Schoennagel, T., Higuera, P. E., Krawchuk, M., Har vey,
B. J., Metc alf, E. C., Schult z, C., Miller, C., Metcalf, A. L., & Buma, B.
(2019). Rethinking resilience to wildfire. Nature Sustainability, 2(9),
7 9 7 – 8 0 4 .
Moritz, M. A., Parisien, M.- A ., Batllori, E., Krawchuk, M. A., Dorn, J. V.,
Ganz, D. J., & Hayhoe, K. (2012). Climate change and disruptions to
global fire activity. Ecosphere, 3(6), art49. https://doi.org/10.1890/
E S 1 1 - 0 0 3 4 5 . 1
Myneni, R . B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The in-
terpret ation of spectr al vegetation indexes. IEEE Transactions
on Geoscience and Remote Sensing, 33(2), 481– 486. h t tp s : //d o i.
org /10.1109/36 .37 7948
Phillips , W. S ., Walter, S., Shant z, H. L ., & Homer, L. (1963). Vegetational
changes in northern Great Plains: Photographic documentation.
College of Agriculture, Universit y of Arizona. https://repos itory.
arizo na.edu/handl e/10150/ 297663
Piao, S., Wang, X., Park, T., Chen, C., Lian, X., He, Y., Bjerke, J. W., Chen,
A., Ciais, P., Tømmervik, H., Nemani, R. R., & Myneni, R. B. (2020).
Characteristics, drivers and feedbacks of global greening. Nature
Reviews Earth & Environment, 1(1), 14– 27. https://doi.org/10.1038/
s 4 3 0 1 7 - 0 1 9 - 0 0 0 1 - x
Policelli, N., Bruns, T. D., Vilgalys, R., & Nuñez, M. A. (2019). Suilloid fungi
as global drivers of pine invasions. New Phytologist, 222(2), 714–
725. ht tps://doi .org /10.1111 /nph.1566 0
Polley, H. W., Briske, D. D., Morgan, J. A ., Wolter, K., Bailey, D. W., &
Brown, J. R. (2013). Climate change and nor th Americ an range-
lands: Trends, projections, and implications. Rangeland Ecology
& Management, 66(5), 493– 511. h tt ps://doi. or g/10 .2111/
R E M - D - 1 2 - 0 0 0 6 8 . 1
Ratajc zak, Z., Nipp ert, J. B., & Collins, S. L. (2012). Woody en-
croachment decreases diversity across north American grass-
lands and savannas. Ecology, 93(4), 697– 703. h t t p s: // do i .
org /10.1890/11- 1199.1
Reeves, M. C., Hanberr y, B. B., Wilmer, H., Kaplan, N. E., & Lauenroth, W.
K. (2020). An assessment of production trends on the Great Plains
from 1984 to 2017. Rangeland Ecology & Management, 76 (1), 165–
179. https://doi.org/10.1016/j.rama.2020.01.011
Reeves, M. C., Moreno, A. L., Bagne, K. E., & Running, S. W. (2014).
Estimating climate change effec ts on net primary production of
rangelands in the United States. Climatic Change, 126 (3), 429– 442.
h t t p s : / / d o i . o r g / 1 0 . 1 0 0 7 / s 1 0 5 8 4 - 0 1 4 - 1 2 3 5 - 8
Reid, A. M., & Fuhlendor f, S. D. (2011). Fire Management in the National
Wildlife Refuge System: A case study of the Charles M. Russell
National Wildlife Refuge, Montana. Rangelands, 33(2), 17– 23.
htt ps://doi.org /10.2111/1551- 501X- 33.2.17
12
|
CU RREY E t al.
Ritter, F., Berkelhammer, M., & Garcia- Eidell, C. (2020). Distinct response
of gross primary productivity in five terrestrial biomes to precip-
itation variability. Communications Earth & Environment, 1(1), 1– 8.
h t t p s : / / d o i . o r g / 1 0 . 1 0 3 8 / s 4 3 2 4 7 - 0 2 0 - 0 0 0 3 4 - 1
Scholt z, R., Fuhlendorf, S. D., & Archer, S. R. (2018). Climate– fire inter-
actions constrain potential woody plant cover and stature in north
American Great Plains grasslands. Global Ecology and Biogeography,
27(8), 936– 945.
Shafer, S. L., Bartlein, P. J., Gray, E. M., & Pelltier, R. T. (2015). Projected
future vegetation changes for the Northwest United States and
Southwest Canada at a fine spatial resolution using a dynamic
global vegetation model. PLoS One, 10(10), e0138759. h t t p s: //d o i .
org/10.1371/journ al.pone.0138759
Stevens, N., Lehmann, C. E. R., Murphy, B. P., & Durigan, G. (2017).
Savanna woody encroachment is widespread across three con-
tinents. Global Change Biology, 23(1), 235– 244. h t t p s: //d o i .
org /10.1111/gcb.13 40 9
Symstad, A. J., & Leis, S. A. (2017). Woody encroachment in northern Great
Plains grasslands: Perceptions, actions, and needs. Natural Areas
Journal, 37(1), 118– 127. https://doi.org/10.3375/043.037.0114
Symstad, A. J., Smith, A. T., Newton, W. E., & Knapp, A. K. (2019).
Experimentally derived nitrogen critical loads for northern Great
Plains vegetation. Ecological Applications, 29(5), e01915. ht t p s : //d oi .
org /10.10 02/eap.1915
Terrer, C., Jackson, R. B., Prentice, I. C., Keenan, T. F., Kaiser, C.,
Vicca, S., Fisher, J. B., Reich, P. B., Stocker, B. D., Hungate, B.
A., Peñuelas, J., McCallum, I., Soudzilovskaia, N. A., Cernusak ,
L. A., Talhelm, A. F., Sundert, K. V., Piao, S., Newton, P. C . D.,
Hovenden, M. J., … Franklin, O. (2019). Nitrogen and phos-
phorus constrain the CO2 fertilization of global plant biomass.
Nature Climate Change, 684– 689. https://doi.org/10.10 38/
s 4 1 5 5 8 - 0 1 9 - 0 5 4 5 - 2
Umbanhowar, C. E. (1996). Recent fire history of the northern Great
Plains. The American Midland Naturalist, 135(1), 115– 121. h t tp s : //
doi.org/10.2307/2426877
Wells, P. V. (1965). Scarp woodlands, transported grassland soils, and
concept of grassland climate in the Great Plains region. Science,
148 (3667), 246– 249.
Welti, E. A. R., Roeder, K. A., Beurs, K. M. de, Joern, A., & Kaspari, M.
(2020). Nutrient dilution and climate cycles underlie declines in a
dominant insect herbivore. Proceedings of the National Academy of
Sciences of the USA, 117(13), 7271– 7275. htt ps://doi.or g/10.1073/
p n a s . 1 9 2 0 0 1 2 1 1 7
Wilcox, B. P., Birt, A., Fuhlendor f, S. D., & Archer, S. R . (2018).
Emerging frameworks for unders tanding and mitigating woody
plant encroachment in grassy biomes. Current Opinion in
Environmental Sustainability, 32, 46– 52. https://doi.org/10.1016/j.
cosust.2018.04.005
Wood, D. J. A., Powell, S., Stoy, P. C ., Thurman, L. L., & Beever, E. A.
(2021). Is the grass always greener? Land surface phenology reveals
differences in peak and season- long vegetation productivity re-
sponses to climate and management. Ecology and Evolution, 11(16),
11168– 11199. https://doi.org/10.1002/ece3.7904
Xiao, J., & Moody, A. (20 05). Geographic al distribution of global green-
ing trends and their climatic correlates: 1982– 1998. International
Journal of Remote Sensing, 26(11), 2371– 2390. h t tp s : //d o i.
org/10.1080/01431 16050 0033682
Zhu, Z., Piao, S., Myneni, R . B., Huang, M., Zeng, Z., Canadell, J. G., Ciais,
P., Sitch, S., Friedlingstein, P., Arneth, A., Cao, C., Cheng, L., Kato, E.,
Koven, C ., Li, Y., Lian, X., Liu, Y., Liu, R., Mao, J., … Zeng, N. (2016).
Greening of the earth and its drivers. Nature Climate Change, 6(8),
791– 795. https://doi.org/10.1038/nclim ate3004
BIOSKETCH
Bryce Currey is a PhD candidate at the Montana State University.
Broadly, his work focuses on ecological and biogeochemical dy-
namics across multiple scales. By employing tools from classic
field ecology and remote sensing, his research examines the
properties and dynamics that emerge at larger landscape scales,
and how these dynamics scale or change when compared with
finer- scale in- situ observation. The ecosystems that he currently
conducts research in include tropical forests, temperate grass-
lands and temperate savannas.
Author contributions: B.C., E.N.J.B. and D.B.M. conceived and
designed the study. N.R.F. handled the soils’ data. B.C. conducted
all analyses and wrote the first edition of the manuscript. All au-
thors contributed to the revision of the manuscript.
SUPPORTING INFORMATION
Additional supporting information may be found in the online
version of the article at the publisher ’s website.
How to cite this article: Currey, B., McWethy, D. B., Fox, N.
R. & Brookshire, E. N. (2022). L arge contribution of woody
plant expansion to recent vegetative greening of the
Northern Great Plains. Journal of Biogeography, 00, 1–12.
https://doi .org /10.1111/j bi.14391
