论文信息
论文解读
AI 辅助整理- 方法
- Sentinel-2 NDVI/NDMI时序、LOESS分解、无人机树冠覆盖分层
- 发现
- 较高树冠覆盖提升NDVI/NDMI与干旱期恢复力,但极端干旱下缓冲效应减弱。
- 机会点
- 可探索极端气候下树冠缓冲阈值及多源遥感对林牧系统韧性的早期预警指标。
论文摘要
摘要 林牧复合系统对于提升热带牧场可持续性、缓解退化至关重要,但其在不同气候条件下的时间动态仍缺乏充分的定量研究。本研究评估了2016至2025年间树木覆盖度对亚马孙林牧复合系统植被动态的影响。基于Sentinel-2影像获取的归一化植被指数(NDVI)与归一化水分指数(NDMI)时间序列,在110个牧场地块上进行分析,采用LOESS季节-趋势分解方法评估长期趋势与季节格局。利用无人机影像提取的树木覆盖百分比对牧场地块进行分层。结果表明,较高的树木覆盖度始终与较高的NDVI和NDMI值相关。在降水较少的时期(2019—2022年),树木覆盖度较高的牧场地块表现出较慢的下降和较快的恢复,表明其对水文胁迫的抵抗力增强且年内变异性降低。统计模型显示,在特定干旱年份(2019、2020和2022年),树木覆盖度与NDVI振幅呈显著负相关,表明季节光谱稳定性更高,同时旱季NDVI整体升高。然而,在2024年极端干旱条件下,各类别之间的差异缩小,表明树木的缓冲能力存在限度。这些发现表明,在中等气候胁迫下,较高的树木覆盖度与研究区牧场系统的更高稳定性和恢复力相关,但极端条件可能限制这些效应。总之,本研究凸显了散生树木作为关键结构组分的作用,提示其可能在增强亚马孙农业前沿林牧复合系统恢复力的策略中发挥潜在作用。
全文
Agroforestry SystemsVol.: (0123456789)
Agroforest Syst (2027) 101:25
https://doi.org/10.1007/s10457-026-01730-1
The role of tree cover in enhancing paddock resilience:
insights from remote sensing in amazonian silvopastoral
systems
Guilherme Honório Fernandez · Adriana Cavalieri Sais ·
Leonardo Pinto de Magalhães · Alexandre de Azevedo Olival
Received: 2 May 2026 / Accepted: 7 November 2026
© The Author(s) 2026
resistance to hydrological stress and reduced intra-
annual variability. Statistical modeling revealed a sig-
nificant negative association between tree cover and
NDVI amplitude during specific drier years (2019,
2020, and 2022), indicating greater seasonal spectral
stability alongside an overall increase in dry-season
NDVI. However, during the extreme 2024 drier con-
dition, differences among classes diminished, sug-
gesting limits to the trees’ buffering capacity. These
findings indicate that higher tree cover was associated
with greater stability and resilience in the studied pas-
ture system under moderate climatic stress, though
extreme conditions can constrain these effects. Ulti-
mately, this study highlights scattered trees as a key
structural component, suggesting their potential role
in strategies to bolster silvopasture resilience across
the Amazonian agricultural frontier.
Keywords Normalized vegetation index · Drought
mitigation · Time series analysis · Agricultural
frontier · Sentinel-2
Introduction
Agriculture is a primary driver of land use expansion
and land cover change globally, and managed grazing
is historically recognized as the single most extensive
form of land use globally (Asner et al. 2004), cur -
rently occupying approximately 25% of the terrestrial
surface (FAO 2025). In Brazil, the livestock sector is
Abstract Silvopastoral systems are vital for enhanc-
ing tropical pasture sustainability and mitigating deg-
radation, yet their temporal dynamics under varying
climates remain insufficiently quantified. This study
evaluated the influence of tree cover on vegetation
dynamics in an Amazonian silvopastoral system
from 2016 to 2025. Sentinel-2 derived Normalized
Difference Vegetation Index (NDVI) and Normal-
ized Difference Moisture Index (NDMI) time series
were analyzed across 110 paddocks, utilizing Sea-
sonal-Trend decomposition using LOESS to assess
long-term trends and seasonal patterns. Tree cover
percentage, extracted from Unmanned Aerial Vehi-
cle imagery, was used to stratify paddocks. Results
demonstrated that higher tree cover consistently cor -
related with elevated NDVI and NDMI values. Dur -
ing periods of lower precipitation (2019–2022),
paddocks with greater tree cover exhibited slower
declines and faster recoveries, indicating enhanced
Supplementary Information The online version
contains supplementary material available at https:// doi.
org/ 10. 1007/ s10457- 026- 01730-1.
G. H. Fernandez · A. C. Sais (*) · L. P. de Magalhães
Rural Development Department, Federal University of São
Carlos, Araras, São Paulo, Brazil
e-mail: acsais@ufscar.br
A. de Azevedo Olival
Faculty of Biological and Agricultural Sciences, Mato
Grosso State University (UNEMAT), Alta Floresta,
Mato Grosso, Brazil
Agroforest Syst (2027) 101:25
25 Page 2 of 15
Vol:. (1234567890)
particularly prominent, predominantly characterized
by extensive pasture management. This sector occu-
pies approximately 1.65 million km2 (MapBiomas
Project 2026) and supports a herd of around 238 mil-
lion heads (IBGE 2024). Consequently, it is a major
contributor to climate change (Ribeiro et al. 2023),
largely due to deforestation trends driven by the deg-
radation of existing pastures, which forces expan-
sion into new areas (Feltran-Barbieri and Féres 2021;
Santos et al. 2024). This dynamic makes mitigation
strategies a critical topic of discussion (Santos et al.
2024).
Within this context, the Amazon forest has faced
continuous conversion to pasture. Parente et al.
(2019) noted that the largest increase in pasture area
in Brazil over the last 33 years occurred in Mato
Grosso, in the southeast Amazon region, while the
biome as a whole has lost around 520,000 km2 in
the last 40 years, with around 440,000 km2 being
converted to pastures (MapBiomas Project 2026).
Despite having extensive areas being converted from
forests to pastures, Oliveira et al. (2013) project that
the productivity of pasture will decrease in all scenar-
ios involving continued Amazon deforestation. Fur -
thermore, habitat loss for pasture implementation is
considered a leading cause of global biodiversity loss
(Foley et al. 2005), even though biodiversity itself
is intrinsically linked to sustainable food production
(Muluneh 2021).
Under these conditions, the livestock sector is
poised to be one of the most severely affected by
climate change, due to changes in precipitation pat-
terns and the increase in droughts and temperatures
(IPCC 2023). In face of these challenges, multifunc-
tional landscapes emerge as alternatives to traditional
models. By incorporating native vegetation along-
side different land uses, these landscapes allow for
the maintenance of ecosystem services, biodiversity,
and social well-being (Ros-Cuéllar et al. 2019). Sil-
vopastoral systems (SPS) are a type of agroforestry,
integrating trees, pasture, and livestock (Cubbage
et al. 2012), and are key examples of such landscapes,
regarded as a key solution to the conflict between
agricultural production and conservation of natural
ecosystems (Peri et al. 2024).
Scattered trees have been considered ‘keystone
structures’ since they have a disproportional effect
on ecosystem services when compared to the area
occupied by each individual tree (Manning et al.
2006). Among other benefits, these systems increase
above- and below-ground carbon storage, provide
microclimatic regulation, and reduce heat stress on
cattle (Castillo et al. 2020). Complementary to these
benefits, the inclusion of trees on rural properties
can act as stepping stones or corridors (Metzger and
Brancalion 2016), thereby facilitating landscape con-
nectivity. Furthermore, as concluded by Santos et al.
(2024) for the Cerrado biome, and applicable to other
regions, these mitigation strategies for pasture deg-
radation have the potential to mitigate deforestation
pressure for new areas, enhancing overall agricultural
sustainability (Cubbage et al. 2012; Peri et al. 2024;
Piao et al. 2021), creating a scenario where farmers
improve their incomes both directly and indirectly
(Peri et al. 2024).
However, despite broad implementation in coun-
tries across Latin America, Africa, Europe, and
Asia, and noticeable ecosystem service benefits, the
integration of tree and forage species also negatively
affects herbage mass by reducing the photosyntheti-
cally active radiation reaching the pasture. This can
occur due to high tree density, species selection, or
management practices—such as pruning and trim-
ming regimes (Benavides et al. 2008; Carvalho et al.
2024). Other studies, however, have emphasized the
role of tree canopy as enhancers of pasture yield
through facilitation processes such as reduced evapo-
transpiration, soil nutrient enrichment and hydrau-
lic lift (Rivest et al. 2013). Therefore, it is crucial to
monitor the effect of tree cover on overall pasture
systems production and evaluate different tree densi-
ties. Despite the need for this assessment, it remains
rare in different biomes, especially in the Amazon
deforestation frontier where forests are being cleared
because of this degradation dynamic. Filling this
knowledge gap, thus, may prove useful to decision-
making regarding the value of scattered trees in agri-
cultural landscapes and as a sustainable agriculture
strategy.
It should also be noted that, despite the benefits for
ecosystem services and other factors, some authors
point out that there is a lack of research on the impacts
of planting trees in silvopastoral systems on aspects
such as biological and chemical indicators and their
interactions (Silva-Olaya et al. 2021). For Amazonian
regions, understanding how silvopastoral systems influ-
ence the provision of multiple ecosystem services is
essential for developing management strategies that
Agroforest Syst (2027) 101:25
Page 3 of 15 25
Vol.: (0123456789)
promote more sustainable agroecosystems (Moreno-
Perez et al. 2025). Monitoring forage biomass is essen-
tial for the sustainable management of pastures, par -
ticularly in tropical regions undergoing agricultural
intensification (Barbosa et al. 2026), as is the case in
the area analyzed here.
In addition to in-situ analyses for assessing these
productive systems, remote sensing has gained sig-
nificant prominence due to the increased availability of
free data and advances in spatial, temporal, and spectral
resolutions (Ma et al. 2014; Parente et al. 2019). These
techniques have assumed a central role in various types
of analysis—such as climate variation, moisture, and
vegetation dynamics—driven not only by rapid data
acquisition but also by substantial time and financial
savings. A prominent example of such applications is
the Normalized Difference Vegetation Index (NDVI),
a widely used metric that provides information ranging
from plant vigor to its complete absence (Huang et al.
2020; Parente et al. 2019). Another example of an index
used is the Normalized Difference Moisture Index
(NDMI), which is related to soil moisture (Hosseini
Chamani et al. 2019); because its equation includes the
Short-Wave Infrared (SWIR) wavelength, this index
is highly sensitive to variations in soil water content
(Lobell and Asner 2002).
Finally, the availability of imagery over long periods
enables the construction of time series to analyze vari-
ations within the same area, a capability exemplified by
the Sentinel-2 A satellite, launched in 2015. This con-
tinuous monitoring permits the observation of trends
and seasonal variations in perennial agricultural areas,
effectively facilitating the identification of degradation
or improvement pathways.
With this in mind, the primary objective of this
study was to evaluate the temporal dynamics of a sil-
vopastoral system in the Amazon. Specifically, we
aimed to: (1) assess the influence of tree cover percent-
age on long-term NDVI trends; (2) quantify the effect
of trees on the intra-annual variability (seasonality) of
NDVI; and (3) model the predicted response of NDVI
to varying tree densities to support decision-making for
future implementation.
Materials and methods
Study area.
The study area is shown in Fig. 1 and is located in
Nova Guarita, Mato Grosso, Brazil. The municipal-
ity has a territory of 1,121.928 km2 and a population
of 4,590 inhabitants as of 2022 (IBGE 2023). The
municipality is crossed by the Teles Pires and Peix-
oto de Azevedo rivers, which are part of the Amazon
Basin, and its natural vegetation is classified as Open
Ombrophilous Forest (IBGE 2004). The region’s soil
is predominantly Red-Yellow Argisol, with pasture as
the main land cover (Mapbiomas 2025). The climate
is classified as Am (tropical monsoon) according to
Köppen, with well-defined rainy, ranging from Octo-
ber to March, and dry seasons.
Following the removal of native vegetation, the
farmer began an assisted natural regeneration of trees
in 1999, and after 2013, the husbandry took the form
of rotational grazing. The farmer managed the regen-
eration focusing on species diversity and aiming at
15% tree cover in each paddock.
Characterized by the integration of pasture, for -
age and trees, known as a silvopasture system, with
a total area of 166 ha, the property currently has 122
paddocks of approximately 1 ha each. Of these, only
110 are used for rotational grazing, with an occupa-
tion of approximately 3.0 to 5.0 AU/ha, depending on
the season. The other 12 paddocks are used for forage
production and thus were not analysed in this study.
Remote sensing data.
Google Earth Engine is a platform designed to make
planetary-scale remote sensing analysis easy by mak -
ing high-performance computational resources avail-
able and coupled with an array of collections of geo-
spatial data (Gorelick et al. 2017). Exploiting this
foundation, we collected multi-source datasets rang-
ing from May of 2016 to April of 2025, including
NDVI, NDMI and Precipitation. Information about
the datasets used in this study is presented in Table 1,
including data source, temporal and spatial resolution
and GEE catalog link.
Through the GEE platform, we accessed the
‘COPERNICUS/S2_HARMONIZED’ dataset
from May 2016 to April 2025. To ensure high-
quality observations, we implemented a rigorous,
Agroforest Syst (2027) 101:25
25 Page 4 of 15
Vol:. (1234567890)
Fig. 1 Location of the study area in Nova Guarita, Mato Grosso, Brazil, showing the spatial distribution of the 122 paddocks within
the silvopastoral system, with 110 rotational-grazing paddocks included in the analysis
Table 1 Summary of remote sensing, climate, and UAV datasets used in this study, including spectral characteristics and spatial–
temporal resolution
Name Data source Bands Wavelengths Temporal and spatial
resolution
GEE catalog
NDVI Sentinel-2 MSI B8 and B4 835.1 nm (S2A) /
833 nm (S2B) and
664.5 nm (S2A) /
665 nm (S2B)
5 days; 10 m COPERNICUS_S2_
HARMONIZED
NDMI Harmonized Sentinel-2
MSI
B8 and B11 835.1 nm (S2A) /
833 nm (S2B) and
1613.7 nm (S2A) /
1610.4 nm (S2B)
5 days; 10 m COPERNICUS_S2_
HARMONIZED
Precipitation CHIRPS 2.0 Not applicable Not applicable Daily; 0.05º UCSB-CHG_CHIRPS_
DAILY
Tree cover UAV (Phantom 4
advanced)
Not applicable Not applicable Not applicable; 6,5 cm Not applicable
Agroforest Syst (2027) 101:25
Page 5 of 15 25
Vol.: (0123456789)
paddock-specific cloud filtering procedure. Rather
than relying on generic scene-level cloud metadata,
a custom GEE algorithm calculated the exact cloud
cover percentage strictly within the boundaries of
each paddock (Area of Interest—AOI) using the
QA60 bitmask band. Images were retained only
if the cloud cover inside the specific paddock was
below 10%. Subsequently, pixel-level cloud mask -
ing was applied using the QA60 band to remove
remaining opaque clouds (Bit 10) and cirrus clouds
(Bit 11). Digital numbers were scaled by a factor
of 10,000 to represent top-of-atmosphere (TOA)
reflectance. NDVI and NDMI were then calculated
for each valid pixel, spatially averaged across each
paddock utilizing the GEE reduceRegion function
(mean reducer), and temporally aggregated into
monthly composites.
For precipitation data, we also used GEE to
retrieve data for the last 30 years in the entire area,
since with CHIRPS (Funk et al. 2015) resolution
we couldn’t discriminate between paddocks.
Additionally, tree cover percentage was calcu-
lated based on classification done through an image
obtained from an unmanned aerial vehicle (UAV)
and is also included in Table 1. Detailed informa-
tion of these datasets and additional information
are presented below.
NDVI
The normalized difference vegetation index is a
remote sensing indicator that provides information
about vegetation growth and cover, being able to
capture 90% of the plant information through the
use of near-infrared (NIR) and red (R) bands (Li
et al. 2021). It is also one of the most used indexes
(Huang et al. 2020) as it is easy to calculate and
interpret, the formula being:
As plants absorb the light in the red region
and reflect infrared light, the index can be used to
assess plant activity, health and stress (Maseny -
ama et al. 2022), varying from 1, with ideal perfect
plant condition, to − 1, with plant absence.
NDVI = NIR − RNIR + R
NDMI
The normalized difference moisture index has been
used to assess plant hydrological stress and also soil
moisture content (Safi et al. 2021; Koohikeradeh;
Gumiere; Bonakdari 2025). In the same manner as
NDVI, it is also a normalized index, varying from − 1
to 1. The formula is:
NDMI works by the same principle as NDVI: a
healthy plant will reflect more near-infrared light
and absorb more shortwave infrared (SWIR) light
through its water content, thus NDMI will have a
higher value; if the plant is suffering from hydrologi-
cal stress, meaning its cells have less water content, it
will absorb less SWIR, having a lower NDMI value.
Precipitation
Climate Hazards Center InfraRed Precipitation With
Station Data (CHIRPS 2.0) aims to make up-to-date
precipitation data accessible, reliable, and complete,
providing daily precipitation data at a spatial reso-
lution of 0.05° (approximately 5.5 km) (Funk et al.
2015). Through GEE, daily precipitation data for the
study area were temporarily summed into monthly
totals. The data were spatially extracted by calculating
the spatial mean of the intersecting CHIRPS pixels
across the defined polygons using the reduceRegion
mean function. Because the CHIRPS coarse spatial
resolution does not permit meaningful discrimination
among 1-ha individual paddocks, the extracted pre-
cipitation was treated as a continuous study-area-level
climatic variable.
Tree cover
Canopy area assessment was performed using a
Phantom 4 Advanced unmanned aerial vehicle
(UAV) during March 2021. The UAV was equipped
with a 20-megapixel camera featuring a 1-inch sen-
sor and a mechanical shutter, a 3-axis stabilization
gimbal and with a GPS/GLONASS connection
(DJI, 2017); FOV 84º, 8.8 mm / 24 mm (35 mm
equivalent), f/2.8-f/11, autofocus 1 m—∞. Cam-
era’s specifications. The flight height was 150 m,
with a lateral and longitudinal overlap of 80%. UAV
NDMI = NIR − SWIRNIR + SWIR
Agroforest Syst (2027) 101:25
25 Page 6 of 15
Vol:. (1234567890)
average speed during the flight was 10 m/s, result-
ing in imagery with a spatial resolution of 6.5 cm,
this planning was performed through Pix4D Cap-
ture Software.
The images resulting from the flight were merged
into an orthomosaic and, combined with an in-situ
tree species inventory, were used to classify trees and
estimate their canopy area through photointerpreta-
tion using QGIS 3.40. Because the UAV flight pro-
vided a single temporal snapshot, a visual assessment
of historical high-resolution imagery from Google
Earth (Google 2024) of 2016 and 2024 was conducted
to verify canopy continuity. This assessment was fur -
ther supported by Google Earth’s change detection
tool, which indicated minimal structural shifts within
the paddocks, with alterations predominantly con-
centrated in the adjacent Legal Reserve areas rather
than the silvopastoral system itself. This analysis con-
firmed that the structural tree cover remained highly
stable across the paddocks, with no significant growth
expansion or canopy loss observed during the study
period. This structural continuity was further cor -
roborated by information provided by the farmer dur -
ing field visits, confirming that regular canopy prun-
ing is performed to manage tree growth and maintain
optimal shading targets. This active management and
temporal stability justify the use of a fixed tree cover
percentage to stratify the paddocks throughout the
time series analysis.
Data analysis
System resilience
To evaluate system resilience using remote sensing
data, we operationalized this concept through two
complementary ecological dimensions, following
Hodgson et al. (2015): (1) Resistance, defined as the
system’s capacity to withstand disturbance, which
we assessed through the varying trends of NDVI and
NDMI in response to interannual precipitation defi-
cits; and (2) Seasonal Stability, defined as the reduc-
tion in intra-annual NDVI amplitude and the main-
tenance of vegetation vigor during the dry season
(May–September). In this context, ‘resilience’ refers
to the capacity of the paddock-level vegetation to
resist and stabilize under climate-driven hydrological
stress.
Time series decomposition
For the time series analysis, the Seasonal-Trend
decomposition using Loess (STL) method was
selected due to its flexibility in handling non-lin-
ear patterns, robustness to outliers, and transparent
design (Abbes et al. 2018; Cleveland et al. 1990).
Developed based on Cleveland et al. (1990), the
STL algorithm decomposes a time series into three
components: trend ( Tt ), seasonal ( St ), and remain-
der ( Rt ). The procedure assumes an additive decom-
position model, such that:
where Yt is the observed time series NDVI or NDMI
value at time t.
The STL procedure consists of two nested loops:
an inner loop nested within an outer loop. Within
each pass of the inner loop, the seasonal and trend
components are iteratively updated through six
steps: (1) detrending, (2) cycle-subseries smooth-
ing, (3) low-pass filtering of smoothed cycle-sub-
series, (4) detrending of smoothed cycle-subseries,
(5) deseasonalizing, and (6) trend smoothing. The
outer loop repeatedly calls the inner loop and com-
putes robustness weights, which serve to reduce the
influence of outliers in the data (Abbes et al. 2018;
Cleveland et al. 1990).
The table generated in the GEE environment was
utilized in Python 3.10, null values were filtered.
To apply the STL method, several parameters were
defined. Due to the variable number of satellite
images per year after cloud filtering, the time series
was first aggregated into monthly means. Conse-
quently, the frequency of observation, or period,
was set to 12. Following the recommendations
of Abbes et al. (2018), the seasonal component’s
smoothing parameter was set to 13 to ensure a stable
seasonal pattern, allowing for gradual interannual
shifts in phenology while preventing the seasonal
component from overfitting to random intra-annual
noise. Finally, robustness iterations were enabled to
mitigate the impact of potential outliers. To address
the sensitivity of the STL decomposition to these
smoothing parameters, we performed a sensitivity
evaluation by varying the seasonal window (11 and
15) and observing the automatically adjusted trend
smoothers. The long-term temporal trajectories and
Yt = Tt + St + Rt
Agroforest Syst (2027) 101:25
Page 7 of 15 25
Vol.: (0123456789)
the hierarchical differentiation among the tree cover
classes remained consistent across these variations,
demonstrating that the observed ecological signals
are robust and not artifacts of the specific param-
eterization of the STL algorithm.
Finally, to evaluate the influence of tree density
on NDVI and NDMI trends, the 110 paddocks were
stratified into three distinct classes based on the dis-
tribution of tree cover percentage. The High Cover
(HC) and Low Cover (LC) groups represented the
upper and lower deciles of the dataset (n = 11 each,
representing 10% of the sample), respectively. The
remaining paddocks were categorized as Intermedi-
ate Cover (IC) (n = 88). The selection of the upper
and lower deciles provides an objective, data-driven
threshold to isolate the absolute extremes of the struc-
tural gradient, avoiding arbitrary percentage cut-offs.
Contrasting the top and bottom 10% against the pre-
dominant farm matrix (the intermediate 80%) maxi-
mizes the visual detection of potential tree-mediated
ecological signals.
Statistical modeling
The monthly variation of NDVI and its annual ampli-
tude were evaluated over time using Generalized
Linear Mixed Models (GLMM), utilizing the lme4
(Bates et al. 2015) and glmmTMB packages (Brooks
et al. 2017) in the R software (R Core Team 2026).
The analysis was divided into two main components:
monthly NDVI variation during the dry season and
annual NDVI amplitude. While NDMI was main-
tained as a complementary descriptive indicator to
visually assess long-term moisture status trends (via
STL decomposition), the inferential statistical mode-
ling focused exclusively on NDVI to quantify vegeta-
tion seasonal stability and amplitude.
For the monthly variation of NDVI, the model was
fitted including the effects of tree cover percentage
(numerical factor), months of the year (categorical
variable representing collection dates), and the inter -
action between them. The paddocks where the trees
were located were included as a random effect to
account for the dependence between repeated obser -
vations within the same experimental unit. Due to the
temporal nature of the sampling, a first-order autore-
gressive correlation structure was specified among
measurements within each paddock to model the
residual autocorrelation between successive periods.
It is important to note that for NDVI, only data
from months with available information across all
evaluated years were selected. Consequently, the
months in which modeling was possible corresponded
to the dry season, from May to September (months
with lower cloud cover). Additionally, paddocks that
didn’t have data in all of these months were removed.
In total, 103 paddocks remained for the evaluation of
monthly variation of NDVI.
The annual NDVI amplitude was defined as the
difference between the maximum and minimum
monthly NDVI values observed for each paddock
within a given year. Because NDVI amplitude rep-
resents a proportion (values bounded between 0 and
1), its variation over time was modeled using a Beta
distribution with a logit link function. The model
included the interaction between year and tree cover
percentage as fixed effects, paddock as a random
effect, and a dispersion formula varying by year. To
explicitly address the role of precipitation, we tested
an alternative model with continuous ‘Precipitation ’.
However, model selection demonstrated that the year-
based model vastly outperformed the precipitation-
based model (ΔAIC = 595.93). This indicates that
the categorical year variable better captures complex
Table 2 Generalized
Linear Mixed Model
estimates for the association
between tree canopy cover
(%) and NDVI amplitude
by year
Year β (Tree canopy cover) SE 95% CI p-value
2017 − 0.00764 0.00768 − 0.02269 to 0.00742 0.320
2018 − 0.00155 0.00524 − 0.01182 to 0.00873 0.768
2019 − 0.00970 0.00378 − 0.01711 to − 0.00229 0.010
2020 − 0.02067 0.00587 − 0.03217 to − 0.00917 < 0.001
2021 − 0.00339 0.00447 − 0.01215 to 0.00537 0.448
2022 − 0.01781 0.00535 − 0.02830 to − 0.00733 < 0.001
2023 0.00674 0.00484 − 0.00276 to 0.01623 0.164
2024 0.00738 0.00419 − 0.00084 to 0.01560 0.078
Agroforest Syst (2027) 101:25
25 Page 8 of 15
Vol:. (1234567890)
interannual climatic variations (e.g., rainfall distribu-
tion or maximum temperatures) than aggregated total
precipitation alone (Table 2).
The selection of the most parsimonious model was
based on the Akaike Information Criterion (AIC)
using the model.sel() function from the MuMIn pack -
age (Barton 2024). The standardized residuals of
the selected model were inspected for normality and
homoscedasticity. The presence of residual autocorre-
lation was assessed using the ACF() function applied
to the normalized residuals. The significance of the
fixed effects was tested using the anova() function,
and the goodness-of-fit was expressed by the mar -
ginal and conditional coefficients of determination
( R2m and R2c ), alongside prediction error metrics such
as Root Mean Square Error (RMSE) and Mean Abso-
lute Error (MAE).
Predictions of the fitted NDVI values and their
annual amplitude were generated as a function of
tree cover levels, considering the fixed effects of
the selected model, and visualized using plots con-
structed with the ggplot2 package (Wickham 2016).
Results
Climatic context.
Precipitation during the study period exhibited a
clear seasonal pattern, with a distinct wet and dry
phase throughout the year (Fig. 2). Given the strong
dependence of pasture productivity on water availa-
bility in tropical systems, characterizing rainfall vari-
ability provides essential context for interpreting the
temporal dynamics of the vegetation indices analyzed
in this study (Amaral et al. 2025; Huang et al. 2025;
Tian et al. 2025).
As illustrated in Fig. 2, monthly rainfall followed
a consistent seasonal cycle, with higher values gen-
erally occurring between October and April and
reduced precipitation during the dry months from
May to September. Nevertheless, considerable inter -
annual variability was observed. The years 2022,
2023, and 2024 exhibited comparatively lower rain-
fall levels, indicating a sequence of drier conditions,
whereas 2018 stood out as a wetter year within the
series.
This recent reduction in precipitation indicates
comparatively drier conditions during the latter part
of the study period. Such conditions are consistent
with the regional climatic effects associated with
El Niño–Southern Oscillation (ENSO), which can
Fig. 2 Monthly precipitation in the study area from May 2016 to April 2025, illustrating the seasonal pattern and interannual vari-
ability. Data derived from the CHIRPS 2.0 dataset accessed via Google Earth Engine
Agroforest Syst (2027) 101:25
Page 9 of 15 25
Vol.: (0123456789)
contribute to reduced precipitation and increased
temperatures in tropical South America (Hsiang
and Meng 2015), and provide an important climatic
context for interpreting the vegetation responses
described in the following sections.
Vegetation dynamics (NDVI and NDMI).
The temporal dynamics of vegetation were evaluated
through the trend components of NDVI and NDMI,
allowing comparison among paddocks with contrast-
ing levels of tree cover. Figure 3 shows the decom-
posed NDVI trend for 110 paddocks stratified by
tree cover classes. A clear and consistent hierarchy is
observed across the study period, with the high-cover
group exhibiting the highest NDVI values, followed
by the intermediate-cover and low-cover groups.
During the decline phase between 2019 and 2022,
the HC group demonstrated a more gradual drop in
NDVI compared to the steeper decline observed in
the LC group. Similarly, following the 2023 decline
in precipitation, the HC group recovered to a higher
NDVI peak and maintained this advantage until the
onset of the next major reduction in rainfall in 2024.
The exceptional decline in precipitation of 2024,
however, marked a shift in the dynamic between these
groups as in the final period the trends of the three
groups converged. This convergence was driven by
a continued decline in the HC group’s NDVI, while
the LC group showed a slight recovery from its lower
baseline.
Similarly, NDMI exhibited a declining trend
between 2019 and 2022 (Fig. 4), with a steeper
decline in the low-cover group. In contrast, the inter -
mediate-cover group showed a more gradual decline,
reaching values similar to the high-cover group
between 2020 and 2021.
Effects of tree cover on seasonal stability.
Regarding the interannual variation of NDVI (ampli-
tude), the model showed a significant interaction
Fig. 3 Decomposed NDVI trend from 2016 to 2025 strati-
fied by tree cover classes. Paddocks were grouped into low-
cover (LC), intermediate-cover (IC), and high-cover (HC)
classes. Colored lines represent the mean trend for each class
(red = LC, black = IC, green = HC), calculated from 110 pad-
docks, while grey lines represent individual paddocks. The
trend component was derived using STL decomposition. The
figure shows a consistent hierarchy among classes, with higher
tree cover associated with higher NDVI values over time
Agroforest Syst (2027) 101:25
25 Page 10 of 15
Vol:. (1234567890)
between tree cover and specific years. The model
showed a R2m of 0.885 and a R2c of 1.000. The RMSE
and MAE were 0.0329 and 0.0250, respectively, indi-
cating relatively small deviations between observed
and model-predicted NDVI amplitude values on the
original response scale. On the logit scale, significant
negative slopes were detected in 2019 (β = − 0.0097,
p = 0.010), 2020 (β = − 0.0206, p < 0.001), and 2022
(β = − 0.0178, p < 0.001) (Table 2). Tree cover was
associated, in these years, with a reduction in NDVI
amplitude, indicating greater seasonal stability. Pair -
wise comparisons further confirmed that the slopes
Fig. 4 Decomposed NDMI trend from 2016 to 2025 strati-
fied by tree cover classes. Paddocks were grouped into low-
cover (LC), intermediate-cover (IC), and high-cover (HC)
classes. Colored lines represent the mean trend for each class
(red = LC, black = IC, blue = HC), calculated from 110 pad-
docks, while grey lines indicate individual paddocks. The trend
component was derived using STL decomposition. A similar
pattern to NDVI is observed, with higher tree cover gener -
ally associated with higher NDMI values and a more gradual
decline during drier periods
Fig. 5 Model-predicted annual amplitude of NDVI by tree
cover and year. Lines represent predicted NDVI range for dif-
ferent levels of tree cover (4.3, 16.5, and 21.5%). Shaded areas
indicate the 95% confidence intervals for the predictions. Grey
dots represent the raw observed NDVI amplitude data for the
individual paddocks
Agroforest Syst (2027) 101:25
Page 11 of 15 25
Vol.: (0123456789)
in 2019, 2020, and 2022 were significantly different
from those observed in subsequent years like 2023
and 2024 (p < 0.05). To visualize this effect in align-
ment with our previously established canopy classes,
predicted values were extracted using the exact
boundaries of the tree cover deciles (4.3, 16.5, and
21.5%) (Fig. 5).
For the analysis of NDVI monthly variation dur -
ing dry seasons, the ANOVA revealed a significant
positive effect of tree cover (p = 0.01). The temporal
factor (month of the data) did not yield a significant
main effect in this restricted seasonal selection. The
model demonstrated high explanatory power, show -
ing a 90% correlation between the observed NDVI
in dry months and the values predicted by tree cover.
Quantitatively, the model indicates that for every 1%
increase in tree cover, the paddock exhibits an aver -
age NDVI increase of 0.0029 during the dry months
(Fig. 6).
Discussion
Climate variability and long-term paddock resilience.
The temporal fluctuations observed in NDVI and
NDMI closely followed interannual precipitation
variability, particularly during the sequence of drier
conditions after 2018 and the severe reduction in pre-
cipitation experienced in 2024, associated with the El
Niño–Southern Oscillation (Hsiang and Meng 2015).
While climate variability explains the overall trajec-
tory of decline and recovery, it does not fully account
for the persistent differences observed among tree
cover classes, indicating an additional structural influ-
ence within the system.
Our results indicate that the presence of scat-
tered trees in Amazonian pastures is associated with
reduced intra-annual variability and increased veg-
etation stability. Over the nearly decade-long time
series, paddocks with higher tree cover consistently
exhibited higher NDVI and NDMI values, as well
as a more gradual decline during drier periods. In
addition, these paddocks maintained higher veg-
etation activity during the dry season, suggesting
a buffering effect of tree cover under water-limited
conditions. In keeping with De Macêdo Carvalho
et al. (2024), one possible explanation for this pat-
tern is the effect of tree shade on evapotranspiration
and soil moisture retention, potentially mediated by
lower soil and air temperatures. These mechanisms,
however, were not directly measured in the present
study.
However, this effect was not constant across all cli-
matic conditions. The convergence observed among
tree cover classes during the severe drier conditions
of 2024 indicates that extreme climatic stress may
limit the stabilizing influence of trees. This suggests
that while tree cover enhances system resilience
under moderate stress, its buffering capacity may be
reduced under extreme drier conditions.
In addition to climatic variability, local manage-
ment factors may have contributed to the observed
fluctuations in vegetation indices. According to infor-
mation provided by the farmer during field visits,
periods of NDVI decline were associated with pest
outbreaks, particularly spittlebugs and caterpillars,
which are known to reduce pasture vigor. Further -
more, changes in herd composition, from growing
cattle to breeding cows, may have increased grazing
pressure, as adult cows typically exhibit higher forage
intake than younger animals.
Fig. 6 Predicted NDVI
as a function of tree cover
(%) during the dry season
(May–September). The
solid blue line represents
the model estimate, and the
shaded area indicates the
95% confidence interval.
Grey dots represent the raw
observed NDVI data for the
individual paddocks
Agroforest Syst (2027) 101:25
25 Page 12 of 15
Vol:. (1234567890)
The subsequent increase in NDVI observed in
2024 coincided with pest control measures and a
reduction in stocking rates, suggesting a recovery of
vegetation conditions. Although these factors were
not quantitatively measured in this study, they provide
important contextual information that helps explain
deviations from purely climate-driven patterns.
In contrast to the widespread degradation trends
reported for open pastures in Brazil (Santos et al.
2024; Feltran-Barbieri and Féres 2021), paddocks
with intermediate and high tree cover maintained a
more stable trajectory for most of the study period.
This pattern suggests that the inclusion of trees may
contribute to slowing degradation pathways com-
monly observed under conditions comparable to
those observed in the study site.
The mechanisms underlying this pattern were not
directly measured in this study, but may involve facil-
itative ecological processes. Previous studies have
shown that trees can improve soil water dynamics,
reduce evapotranspiration, and enhance nutrient avail-
ability (Hébert et al. 2025; Olival et al. 2021). Such
processes may help explain the observed increase in
vegetation stability and reduced sensitivity to sea-
sonal water limitation in paddocks with higher tree
cover.
Mitigating seasonal deficits in tropical systems.
This facilitative role of trees becomes particularly
evident when examining the intra-annual variability
of vegetation activity. Our statistical model showed
that higher tree cover was associated with a signifi-
cant reduction in the annual amplitude of NDVI, par -
ticularly during years marked by climatic stress (such
as 2019, 2020, and 2022). A lower NDVI amplitude
indicates a more stable system throughout the dry
season.
In open pastures, vegetation dynamics typically
exhibit pronounced peaks during the rainy season
followed by marked declines during the dry months
(Skidmore 2022). In contrast, paddocks with higher
tree cover maintained a more stable baseline of veg-
etation activity. Consistently, our dry-season analysis
revealed a significant positive relationship between
tree cover percentage and NDVI from May to Sep-
tember. At the study site, these results suggest that
tree cover contributes to reducing seasonal variability
and may help maintain vegetation activity during the
dry period, a critical constraint for tropical livestock
systems (Larsen et al. 2021; Skidmore 2022).
These findings are consistent with the global meta-
analysis by Ripamonti et al. (2025), which reported
that agroforestry systems in tropical regions gener -
ally maintain comparable productivity levels to full-
light systems. In these environments, trees may offset
constraints such as high insolation, high temperatures
and water limitation through microclimatic regulation
and improved soil moisture conditions (Bretas et al.
2020), potentially compensating for reductions in
light availability.
Although these mechanisms were not directly
measured and the tree canopy inherently contributes
to the observed radiometric response, the patterns
confirm that higher tree cover promotes greater sea-
sonal spectral stability across the studied silvopas-
ture systems. In this context, the integrated paddock
vegetation, comprising both the herbaceous layer and
scattered trees, maintains a more stable radiomet-
ric baseline under seasonal water stress compared to
environments with lower tree density.
Conclusion
This study evaluated the temporal dynamics of an
Amazonian silvopastoral system using a multi-year
remote sensing approach, focusing on the role of tree
cover in shaping paddock responses under variable
climatic conditions. The results indicate that higher
tree cover is consistently associated with increased
vegetation activity (NDVI), greater moisture sta-
tus (NDMI), and reduced intra-annual variability,
particularly during dry periods. These patterns sug-
gest that tree cover contributes to the stabilization of
paddock conditions and may enhance overall system
resilience under moderate climatic stress.
However, the convergence observed among tree
cover classes during the drier conditions of 2024 indi-
cates that the buffering capacity of trees is not unlim-
ited and may be reduced under extreme climatic con-
ditions. This highlights the importance of evaluating
the potential benefits and limits of trees when design-
ing and managing silvopastoral systems in regions
subject to increasing climatic variability.
Overall, the findings reinforce the role of tree cover
as a key structural component influencing the tempo-
ral stability of the studied silvopasture systems. By
Agroforest Syst (2027) 101:25
Page 13 of 15 25
Vol.: (0123456789)
integrating long-term satellite data with field-based
characterization of tree cover, this study contributes
evidence on how a silvopastoral system responded
to climate variability in the Amazonian agricultural
frontier.
Despite these contributions, some limitations
should be acknowledged. First, tree cover was derived
from UAV imagery at a single point in time and
assumed to be stable throughout the study period,
which may not fully capture temporal changes in can-
opy structure. Second, the use of NDVI and NDMI
as proxies does not allow direct inference about pas-
ture biomass, soil moisture, or microclimatic condi-
tions. Third, the study was conducted on a single sil-
vopastoral farm, which may limit the generalization
of the results to other regions or management sys-
tems. Finally, the ecological mechanisms underlying
the observed patterns—such as changes in soil water
dynamics, nutrient availability, or microclimate—
were not directly measured and therefore remain
inferred.
Future research integrating field-based measure-
ments of productivity, soil properties, and micro-
climatic conditions across different silvopastoral
contexts will be essential to further elucidate the
mechanisms driving the relationships observed here
and to support the development of more resilient pas-
ture systems in face of future climate change.
Acknowledgements The authors would like to thank the
São Paulo Research Foundation (FAPESP) for the undegradu-
ate scholarship awarded to Guilherme Honório Fernandez
(Grant nº2024/22345-4) and the Brazilian Coordination for the
Improvement of Higher Education Personnel (CAPES) for the
postdoctoral scholarship granted to Leonardo Pinto de Magal-
hães (Process nº 88887.691467/2022-00). This project was also
supported by the Center of Agricultural Sciences (CCA) at
UFSCar through the project FAI RTI-CCA.
Author contributions GHF and ACS conceptualized the
study, performed the literature collection, drafted the manu-
script, and provided overall supervision. GHF retrieved the sat-
ellite data, while AAO performed the experimental modeling.
ACS, LPM, and AAO revised the manuscript. GHF reviewed
and edited the final version for coherence and handled the sub-
mission process. All authors have read and approved the final
manuscript.
Funding The Article Processing Charge (APC) for the
publication of this research was funded by the Coordenação
de Aperfeiçoamento de Pessoal de Nível Superior - Brasil
(CAPES) (ROR identifier: 00x0ma614). Guilherme Honório
Fernandez received an undergraduate scholarship during the
course of this research (Grant nº2024/22345–4) and Leonardo
Pinto de Magalhães received a postdoctoral scholarship from
Brazilian Coordination for the Improvement of Higher Educa-
tion Personnel (CAPES) (Process nº 88887.691467/2022–00).
This project was also supported by the Center of Agricultural
Sciences (CCA) at UFSCar through the project FAI RTI-CCA.
Data availability Data supporting Fig. 2-6, are publicly
available in a Github repository, as part of this study: https://
github. com/ Guilh ermeF ernan dez/ The- Role- of- Tree- Cover- in-
Enhan cing- Pastu re- Resil ience
Declarations
Conflicts of interests The authors declare no competing in-
terests.
Open Access This article is licensed under a Creative Com-
mons Attribution 4.0 International License, which permits
use, sharing, adaptation, distribution and reproduction in any
medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Crea-
tive Commons licence, and indicate if changes were made. The
images or other third party material in this article are included
in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not
included in the article’s Creative Commons licence and your
intended use is not permitted by statutory regulation or exceeds
the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/.
References
Abbes AB, Bounouh O, Farah IR et al (2018) Comparative
study of three satellite image time-series decomposition
methods for vegetation change detection. Eur J Remote
Sens 51:607–615. https:// doi. org/ 10. 1080/ 22797 254. 2018.
14653 60
Amaral DS, De Faria Melo CC, Dalri AB et al (2025) Nutri-
tional imbalances and reduced forage production under
intra-seasonal drought are alleviated by irrigation in tropi-
cal forage grasses. Sci Rep 15:34925. https:// doi. org/ 10.
1038/ s41598- 025- 18760-x
Asner GP, Elmore AJ, Olander LP, Martín RE, Harris AT
(2004) Grazing systems, ecosystem responses, and global
change. Annu Rev Environ Resour 29:261–299. https://
doi. org/ 10. 1146/ annur ev. energy. 29. 062403. 102142
Barbosa LAS, Uliana EM, Mendes MASA et al (2026) Sus-
tainable management of tropical pastures through
dynamic forage biomass monitoring with remote sensing
and machine learning. Progress in Engineering Science
3:100290. https:// doi. org/ 10. 1016/j. pes. 2026. 100290
Bartoń K (2024) Multi-Model Inference [R package MuMIn
version 1.48.11]. https:// cran.r- proje ct. org/ packa ge=
MuMIn. Accessed 2 Mar 2026
Bates D, Mächler M, Bolker B, Walker S (2015) Fitting Linear
Mixed-Effects Models Using lme4. J Stat Softw 67:1–48.
https:// doi. org/ 10. 18637/ jss. v067. i01
Agroforest Syst (2027) 101:25
25 Page 14 of 15
Vol:. (1234567890)
Benavides R, Douglas GB, Osoro K (2008) Silvopastoralism in
New Zealand: review of effects of evergreen and decidu-
ous trees on pasture dynamics. Agrofor Syst 76:327–350.
https:// doi. org/ 10. 1007/ s10457- 008- 9186-6
Bretas IL, Paciullo DSC, Alves BJR et al (2020) Nitrous oxide,
methane, and ammonia emissions from cattle excreta on
Brachiaria decumbens growing in monoculture or sil-
vopasture with Acacia mangium and Eucalyptus grandis.
Agric Ecosyst Environ 295:106896. https:// doi. org/ 10.
1016/j. agee. 2020. 106896
Brooks ME, Kristensen K, van Benthem KJ, Magnusson A,
Berg CW, Nielsen A, Skaug HJ, Mächler M, Bolker BM
(2017) glmmTMB balances speed and flexibility among
packages for zero-inflated generalized linear mixed mod-
eling. The R J 9:378–400. https:// doi. org/ 10. 32614/
RJ- 2017- 066
Castillo MS, Tiezzi F, Franzluebbers AJ (2020) Tree species
effects on understory forage productivity and microcli-
mate in a silvopasture of the Southeastern USA. Agric
Ecosyst Environ 295:106917. https:// doi. org/ 10. 1016/j.
agee. 2020. 106917
Cleveland RB, Cleveland WS, McRae JE, Terpenning I (1990)
STL: A seasonal-trend decomposition procedure based on
loess. J Off Stat 6:3–73
Comprehensive R Archive Network (CRAN) (2025) Linear
and Nonlinear Mixed Effects Models [R package nlme
version 3.1–168]. https:// cran.r- proje ct. org/ packa ge=
nlme. Accessed 2 Mar 2026
Cubbage F, Balmelli G, Bussoni A et al (2012) Comparing sil-
vopastoral systems and prospects in eight regions of the
world. Agrofor Syst 86:303–314. https:// doi. org/ 10. 1007/
s10457- 012- 9482-z
De Macêdo Carvalho CB, De Mello ACL, Da Cunha MV et al
(2024) Ecosystem services provided by silvopastoral sys-
tems: a review. The Journal of Agricultural Science 1–16.
https:// doi. org/ 10. 1017/ s0021 85962 40005 95
De Santana Ribeiro LC, De Sousa Filho JF, Santos GFD, Da
Silva Freitas LF (2023) Structural decomposition analysis
of Brazilian greenhouse gas emissions. World Develop-
ment Sustainability 2:100067. https:// doi. org/ 10. 1016/j.
wds. 2023. 100067
FAO (2025) Land statistics 2001–2023 – Global, regional and
country trends. FAOSTAT Analytical Briefs, No. 107.
Rome.
Feltran-Barbieri R, Féres JG (2021) Degraded pastures in Bra-
zil: improving livestock production and forest restoration.
R Soc Open Sci 8:201854. https:// doi. org/ 10. 1098/ rsos.
201854
Foley JA, DeFries R, Asner GP et al (2005) Global conse-
quences of land use. Science 309:570–574. https:// doi. org/
10. 1126/ scien ce. 11117 72
Funk C, Peterson P, Landsfeld M et al (2015) The climate
hazards infrared precipitation with stations—a new envi-
ronmental record for monitoring extremes. Sci Data
2:150066. https:// doi. org/ 10. 1038/ sdata. 2015. 66
Google (2024) Google Earth Detect Change. Nova Guarita,
Mato Grosso, Brazil. Imagery dates: 2017–2025. Availa-
ble via: https:// earth. google. com/. Accessed 18 Sep 2026.
Gorelick N, Hancher M, Dixon M et al (2017) Google Earth
Engine: planetary-scale geospatial analysis for everyone.
Remote Sens Environ 202:18–27. https:// doi. org/ 10.
1016/j. rse. 2017. 06. 031
Hébert È-M, Martin-Guay M-O, Maheu A et al (2025) Micro-
climate, soil moisture and forage yield vary spatially
within a temperate tree-based intercropping system:
from competition to facilitation. Agric Water Manag
319:109782. https:// doi. org/ 10. 1016/j. agwat. 2025. 109782
Hodgson D, McDonald JL, Hosken DJ (2015) What do you
mean, ‘resilient’? Trends Ecol Evol 30(9):503–506.
https:// doi. org/ 10. 1016/j. tree. 2015. 06. 010
Hosseini Chamani F, Firouzi FA, Amerykhah H (2019)
Pedotransfer Function (PTF) for Estimation Soil moisture
using NDVI, land surface temperature (LST) and nor -
malized moisture (NDMI) indices. J. Water Soil Conserv
26:239–254
Hsiang SM, Meng KC (2015) Tropical economics. Am Econ
Rev 105:257–261. https:// doi. org/ 10. 1257/ aer. p2015 1030
Huang S, Tang L, Hupy JP et al (2020) A commentary review
on the use of normalized difference vegetation index
(NDVI) in the era of popular remote sensing. J for Res
32:1–6. https:// doi. org/ 10. 1007/ s11676- 020- 01155-1
Huang Z, Li X, Mao F et al (2025) Integrating LUCC and for -
est aging to project and attribute subtropical forest NEP in
Zhejiang Province under four SSP-RCP scenarios. Agric
for Meteorol 365:110462. https:// doi. org/ 10. 1016/j. agrfo
rmet. 2025. 110462
IBGE (2004) Mapa de vegetação do Brasil. Instituto Brasileiro
de Geografia e Estatística, Rio de Janeiro
IBGE (2023) Censo Brasileiro de 2022. Instituto Brasileiro de
Geografia e Estatística, Rio de Janeiro
IBGE (2024) Pesquisa Pecuária Municipal 2024. Instituto Bra-
sileiro de Geografia e Estatística, Rio de Janeiro
IPCC (2023) Climate Change 2023: Synthesis Report. Con-
tribution of Working Groups I, II and III to the Sixth
Assessment Report of the Intergovernmental Panel on Cli-
mate Change [Core Writing Team, H. Lee and J. Romero
(eds.)]. IPCC, Geneva, Switzerland.
Koohikeradeh E, Gumiere SJ, Bonakdari H (2025) NDMI-
derived field-scale soil moisture prediction using ERA5
and LSTM for precision agriculture. Sustainability
17:2399. https:// doi. org/ 10. 3390/ su170 62399
Larsen RE, Shapero MWK, Striby K et al (2021) Forage quan-
tity and quality dynamics due to weathering over the dry
season on California annual rangelands. Rangeland Ecol
Manage 76:150–156. https:// doi. org/ 10. 1016/j. rama. 2021.
02. 010
Li S, Xu L, Jing Y et al (2021) High-quality vegetation
index product generation: a review of NDVI time series
reconstruction techniques. Int J Appl Earth Obs Geoinf
105:102640. https:// doi. org/ 10. 1016/j. jag. 2021. 102640
Lobell DB, Asner GP (2002) Moisture effects on soil reflec-
tance. Soil Sci Soc Am J 66:722–727. https:// doi. org/ 10.
2136/ sssaj 2002. 7220
Ma Y, Wu H, Wang L et al (2014) Remote sensing big data
computing: challenges and opportunities. Future Gener
Comput Syst 51:47–60. https:// doi. org/ 10. 1016/j. future.
2014. 10. 029
Manning AD, Fischer J, Lindenmayer DB (2006) Scattered
trees are keystone structures – implications for conserva-
tion. Biol Conserv 132:311–321. https:// doi. org/ 10. 1016/j.
biocon. 2006. 04. 023
Agroforest Syst (2027) 101:25
Page 15 of 15 25
Vol.: (0123456789)
MapBiomas Project - Collection 11 of the Annual Land Cover
and Land Use Maps of Brazil. https:// plata forma. brasil.
mapbi omas. org/. Accessed 2 Mar 2026.
Masenyama A, Mutanga O, Dube T et al (2022) A systematic
review on the use of remote sensing technologies in quan-
tifying grasslands ecosystem services. Gisci Remote Sens
59:1000–1025. https:// doi. org/ 10. 1080/ 15481 603. 2022.
20886 52
Metzger JP, Brancalion PHS (2016) Landscape ecology and
restoration processes. In: Palmer MA, Zedler JB, Falk DA
(eds) Foundations of restoration ecology, 2nd edn. Island
Press, Washington, DC, pp 90–120. https:// doi. org/ 10.
5822/ 978-1- 61091- 698-1_4
Moreno-Perez C, Mora-Motta D, Ortíz-Morea FA et al (2025)
Transitioning from extensive pastures to silvopastoral sys-
tems improves multiple soil ecosystem services in Colom-
bian Amazon. Sci Total Environ 974:179185. https:// doi.
org/ 10. 1016/j. scito tenv. 2025. 179185
Muluneh MG (2021) Impact of climate change on biodiversity
and food security: a global perspective—a review article.
Agriculture & Food Security 10:36. https:// doi. org/ 10.
1186/ s40066- 021- 00318-5
Olival AA, Souza SEXF, Moraes JPG, Campana M (2021)
Effect of Amazonian tree species on soil and pasture qual-
ity in silvopastoral systems. Acta Amazonica 51:281–290.
https:// doi. org/ 10. 1590/ 1809- 43922 02004 692
Oliveira LJC, Costa MH, Soares-Filho BS, Coe MT (2013)
Large-scale expansion of agriculture in Amazonia may be
a no-win scenario. Environ Res Lett 8:024021. https:// doi.
org/ 10. 1088/ 1748- 9326/8/ 2/ 024021
Parente L, Mesquita V, Miziara F et al (2019) Assessing the
pasturelands and livestock dynamics in Brazil, from 1985
to 2017: a novel approach based on high spatial resolu-
tion imagery and Google Earth Engine cloud computing.
Remote Sens Environ 232:111301. https:// doi. org/ 10.
1016/j. rse. 2019. 111301
Peri PL, Chará J, Viñoles C et al (2024) Current trends in sil-
vopastoral systems. Agrofor Syst 98:1945–1953. https://
doi. org/ 10. 1007/ s10457- 024- 01093-5
Piao RS, Silva VL, Del Aguila IN, De Burgos Jiménez J (2021)
Green growth and agriculture in Brazil. Sustainability
13:1162. https:// doi. org/ 10. 3390/ su130 31162
R Core Team (2026) R: a language and environment for statis-
tical computing. R Foundation for Statistical Computing,
Vienna
Ripamonti A, Finocchi M, Pulina A et al (2025) Effects
of tree presence on forage yield and nutritive value
in agroforestry livestock systems: a global systematic
review. Agrofor Syst 99:110. https:// doi. org/ 10. 1007/
s10457- 025- 01214-8
Rivest D, Paquette A, Moreno G, Messier C (2013) A meta-
analysis reveals mostly neutral influence of scattered
trees on pasture yield along with some contrasted effects
depending on functional groups and rainfall conditions.
Agric Ecosyst Environ 165:74–79. https:// doi. org/ 10.
1016/j. agee. 2012. 12. 010
Ros-Cuéllar J, Porter-Bolland L, Bonilla-Moheno M (2019)
Can multifunctional landscapes become effective con-
servation strategies? Challenges and opportunities from
a Mexican case study. Land 8:6. https:// doi. org/ 10. 3390/
land8 010006
Safi AR, Karimi P, Mul M et al (2021) Translating open-source
remote sensing data to crop water productivity improve-
ment actions. Agric Water Manag 261:107373. https:// doi.
org/ 10. 1016/j. agwat. 2021. 107373
Santos COD, De Siqueira Pinto A, Santos MPD et al (2024)
Livestock intensification and environmental sustainability:
an analysis based on pasture management scenarios in the
Brazilian savanna. J Environ Manage 355:120473. https://
doi. org/ 10. 1016/j. jenvm an. 2024. 120473
Silva-Olaya AM, Olaya-Montes A, Polanía-Híncapié KL et al
(2021) Silvopastoral systems enhance soil health in the
Amazon region. Sustainability. https:// doi. org/ 10. 3390/
su140 10320
Skidmore ME (2022) Outsourcing the dry season: cattle ranch-
ers’ responses to weather shocks in the Brazilian Amazon.
Am J Agric Econ 105:409–433. https:// doi. org/ 10. 1111/
ajae. 12333
Tian W, Liang F, Tu L et al (2025) Mineral fertilizer substitu-
tion and application of Bacillus velezensis SQR9 reduced
nitrogen-oxide emissions in tropical vegetable fields.
Agric Ecosyst Environ 384:109554. https:// doi. org/ 10.
1016/j. agee. 2025. 109554
Wickham H (2016) ggplot2: Elegant Graphics for Data Analy -
sis. Springer-Verlag, New York.
Publisher’s Note Springer Nature remains neutral with regard
to jurisdictional claims in published maps and institutional
affiliations.
论文原文与出版信息
Agroforestry Systems