LETTER • OPEN ACCESS You may also like
Patterns and drivers of disturbance in tropical - Assessment of land use change in theThuma forest reserve region of Malawi,
Africa
forest reserves of southern Ghana Mihla Phiri and Harrington Nyirenda
- Selection of agroforestry tree-base
participatory and ecological approach in
To cite this article: Dan Wanyama et al 2023 Environ. Res. Lett. 18 064022 Central Kalimantan, Indonesia
E I Purnawan, R Jemi, H Kasim et al.
 - Untangling methodological and scale
considerations in growth and productivity
trend estimates of Canada’s forests
View the article online for updates and enhancements. William Marchand, Martin P Girardin,
Sylvie Gauthier et al.
This content was downloaded from IP address 197.255.69.32 on 08/06/2023 at 12:33
Environ. Res. Lett. 18 (2023) 064022 https://doi.org/10.1088/1748-9326/acd399
LETTER
Patterns and drivers of disturbance in tropical forest reserves of
OPEN ACCESS southern Ghana
RECEIVED
19 October 2022 DanWanyama1, Michael C Wimberly1,∗ and Foster Mensah2
REVISED 1
3 May 2023 Department of Geography and Environmental Sustainability, University of Oklahoma, Norman, OK, United States of America
2 Centre for Remote Sensing and Geographic Information Services, University of Ghana, Accra, Ghana
ACCEPTED FOR PUBLICATION ∗ Author to whom any correspondence should be addressed.
9 May 2023
PUBLISHED E-mail: mcwimberly@ou.edu
22 May 2023
Keywords:Upper Guinean Forest region, boosted regression trees, VIIRS, Landsat, wildfire, land use and land cover change, climate
Original content from Supplementary material for this article is available online
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence. Abstract
Any further distribution Ghana has retained a substantial area of tropical forests in an extensive network of protected
of this work must
maintain attribution to reserves. These forests are impacted by land uses such as logging, mining, and agriculture as well as
the author(s) and the title
of the work, journal wildfires. We studied forest disturbance and recovery from 2013 to 2020 using annual maps of
citation and DOI. forest cover derived from Landsat imagery. Fire-associated disturbance was distinguished using
VIIRS active fire data. We used boosted regression trees to model disturbances in closed and open
forests as a function of climate variability, human accessibility, and landscape structure. A total of
3562 km2 of forest reserves were disturbed, of which 17% (615 km2) were fire disturbances and
83% (2946 km2) were non-fire disturbances. Of the total disturbed area, 68% was degradation
(change from closed to open forest), 28% was open forest loss, and only 4% was closed forest loss.
Over the same period, 2702 km2 of forest reserves recovered, with 1948 km2 of these recovering to
closed-canopy forests. Fire disturbances were strongly associated with precipitation anomalies and
occurred mostly in drier years, whereas non-fire disturbances had weaker relationships with
precipitation. Disturbances in closed forests occurred in landscapes where closed forest cover was
already low. In contrast, disturbances in open forests were most common in locations with
intermediate levels of population pressure from nearby cities and proximity to non-forest land
cover. The results support the idea that forest disturbance in Ghana is a multi-stage process
involving degradation of closed forests followed by loss of the resulting open forests. Although
non-fire disturbance rates are consistent from year to year, sharp increases in fire disturbance occur
in drought years. Locations with the highest disturbance risk are associated with measurable
indicators of climate, human pressure, and fragmentation, which can be used to identify these
areas for conservation and forest restoration activities.
1. Introduction disturbances are driven by human activities, such
as agriculture, logging, and mining, due to pressure
Tropical forests sequester 1.7 Gt of carbon per year from fast-increasing human populations (Gibbs et al
(Harris et al 2021), are home to about half of 2010, Malhi et al 2014). These dynamics are also
Earth’s biodiversity, and provide essential ecosystem influenced by climate change, including stronger and
goods and services for more than 1.2 billion people more frequent droughts (Malhi et al 2014, Edwards
(Malhi et al 2014, Lewis et al 2015). However, these et al 2019) that cause vegetation dieback and increase
forests are experiencing significant ecological disturb- the risk of wildfires (Brando et al 2019). In recent
ances, including loss of forests to other land uses years, wildfires have become a common occurrence
and degradation that changes forest structure, spe- in tropical forests, despite the high moisture in most
cies composition, and biomass within intact forests of these ecosystems (Brando et al 2019, Edwards et al
(Shapiro et al 2016, Matricardi et al 2020). Forest 2019). Most of these fires have been observed during
© 2023 The Author(s). Published by IOP Publishing Ltd
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
El Nino-related droughts (Aragão et al 2007, 2008) Fires are most common in drier tropical forest types
with the largest ones occurring in disturbed forests (Hall and Swaine 1981). Severe drought is associated
(Cochrane et al 1999, Cochrane and Laurance 2002, with increased tree mortality throughout the tropics
de Faria et al 2017, Dwomoh et al 2019). Yet, there (Phillips et al 2010), but also affects fire behavior by
is still a need to identify the geographic factors that altering the availability of understory fuels (Cochrane
make locations vulnerable to future forest loss and et al 1999). Second, we assessed human accessibility
degradation, and to expand our limited knowledge as a measure of the effect of land use pressure, includ-
of how tropical fire regimes may respond to increas- ing agricultural encroachment, logging, and mining
ing human populations and changing environments. as well as fire ignitions. Reserves located close to large
We addressed these knowledge gaps by conducting a human populations are more likely to be disturbed
study of historical disturbances in the Upper Guinean and this risk is expected to decline with decreas-
Forest (UGF) region ofWest Africa and using the res- ing population sizes and increasing distance from
ults to highlight susceptible areas where conservation settlements (Güneralp et al 2013, Herrmann et al
and restoration efforts can be targeted. 2020). Finally, we incorporated landscape structure to
The UGF region is a globally significant biod- assess how the legacies of past change influence forest
iversity hotspot (Myers et al 2000) but is also among disturbance. Strong positive feedbacks exist between
the most climatically marginal (Malhi and Wright forest structure and disturbance risk (Flores and Staal
2004) and human-modified (Norris et al 2010) trop- 2022). Forest degradation and loss thin and fragment
ical ecosystems in the world. Persistent and severe the forest canopy and affect vegetation, fuels, and
droughts have occurred in recent decades and are microclimate, rendering forests more susceptible to
expected to become more common with intensify- fire (Laurance and Williamson 2001). Historical dis-
ing climate change (Sylla et al 2016). The popula- turbance also increases the risk of non-fire disturb-
tion of West Africa increased almost six-fold between ance, as disturbed forests are preferred over intact
1950 and 2020 (72–420 million) and is projected to forests for land use activities such as crop cultiva-
reach 801 million by 2050 (UN DESA Population tion and grazing (Carvalho et al 2019, Herrmann et al
Division 2022). In Ghana, almost all the remaining 2020, Wang et al 2020).
forest is found in protected reserves located in the The overarching goal of this study was to char-
southern third of the country. In this region, annual acterize spatial patterns and drivers of fire and non-
precipitation ranges from more than 2000 mm in fire disturbances within tropical forests of Ghana.
the southwest to less than 750 mm at the northern Specific objectives were to; (1) map spatiotemporal
edge of the forest zone (Amissah et al 2014), influ- patterns of fire and non-fire disturbances in protec-
encing the distribution of forest types along a gradi- ted reserves of southern Ghana from 2013 to 2020,
ent from wet evergreen (WE) to dry semi-deciduous and (2) identify the main drivers related to climate
(Hall and Swaine 1976). These reserves are heav- variability, human accessibility, and landscape struc-
ily impacted by agricultural encroachment, logging, ture influencing fire and non-fire disturbances in
mining, and wildfires (Acheampong et al 2016, Boadi open and closed forests. We combined annual maps
et al 2016, Kouassi et al 2021). Historically, forest of forest cover derived from Landsat imagery with
fires were rare, with occasional low-intensity burns in active fire detections from the Visible Infrared Ima-
the dry semi-deciduous (fire subtype) zone (Hall and ging Radiometer Suite (VIIRS) to distinguish fire
Swaine 1981). However, forest fires were widespread from non-fire disturbance and used machine learn-
during the severe drought of the 1980s and more ing techniques to quantify the effects of climate vari-
recently in 2016, especially in the dry andmoist forest ability, human accessibility, and landscape structure
types (Swaine 1992, Dwomoh et al 2019). Strong pos- variables. Understanding these relationships is essen-
itive feedbacks between fires, land use, and forest tial for targeting conservation and forest restora-
structure have caused permanent shifts from forest tion activities in Ghana and similar tropical forest
to non-forest vegetation (Dwomoh and Wimberly regions.
2017). These dynamics offer an excellent opportun-
ity to study the effects of climate and human popula- 2. Materials andmethods
tion pressure on fire and non-fire disturbances within
tropical forests. 2.1. Data sources and preprocessing
Forest disturbance is the outcome of complex 2.1.1. Forest change data
interactions between human decisions and actions Annual forest canopy cover estimates from 2013
and ecological and biophysical processes (Flores and to 2020 for all protected forest reserves in south-
Staal 2022). We aimed to identify measurable pre- ern Ghana were generated using Landsat imagery
dictors that could be used to delineate locations combined with training and validation data from
where forest disturbance ismost likely.We considered very high-resolution satellite imagery (Wimberly et al
three groups of variables that were hypothesized to 2022). The random forests algorithm was used to
influence forest disturbance. First, we characterized predict annual canopy cover and the LandTrendr
climate variability by using data on precipitation. algorithm (Kennedy et al 2018) was applied to
2
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Table 1. Definitions of the five change categories used to classify disturbance types and recovery in Ghana.
Change type Description Disturbance type
Closed forest loss Closed forest to low tree cover CFD
Degradation Closed forest to open forest CFD
Open forest loss Open forest to low tree cover OFD
Closed forest recovery Open forest to closed forest N/A
Open forest recovery Low tree cover to open forest N/A
Table 2. List of variables used in boosted regression tree models.
Description Name
1. Forest dynamics
Closed forest fire disturbancea CFD fire
Closed forest non-fire disturbancea CFD non-fire
Open forest fire disturbancea OFD fire
Open forest non-fire disturbancea OFD non-fire
2. Climate variability
Mean annual precipitationb Precipitation (mm)
Annual standardized precipitation anomaliesb PrecipAnom
3. Human accessibility
Population gravity indexc PopulationIndex
Distance from non-forest areasd DistNonForest (km)
Distance from roadse DistRoads (km)
4. Landscape structure
Fragmentation typed FragmentationType
Percent closed forestd %ClosedForest (%)
Slopef Slope (degrees)
a Landsat-derived forest change data (Wimberly et al 2022) and VIIRS active fires (Schroeder
et al 2014).
b CHIRPS pentad data (Funk et al 2015).
c WorldPop (Leasure et al 2020) and Africapolis urban boundaries (OECD/SWAC 2020).
d Landsat-derived canopy cover data (Wimberly et al 2022).
e Global Roads Open Access Data Set (CIESIN—Columbia University & ITOS—University of
Georgia, 2013).
f SRTM DEM (Farr et al 2007).
identify periods of relative stability, disturbance, and than the Moderate Resolution Imaging Spectrora-
recovery. The continuous canopy cover predictions diometer (MODIS) pixel area at nadir, thus VIIRS
were reclassified into low tree cover (<15% canopy is better suited for detecting small and low-intensity
cover), open forest (15%–60%), and closed forest fires (Zhang et al 2017). We removed fire detections
(>60% canopy cover). Maps of the canopy cover flagged as low confidence and used an interpola-
classes for 2013–2020 are provided in Figure A1 in tion algorithm to convert active fire observations into
supplemental materials. Five change categories were burned area estimates by grouping pixels separated
defined: closed forest loss, degradation, open forest by a maximum distance of 2000 m and a maximum
loss, closed forest recovery, and open forest recov- time interval of 2 d and converting clusters of pixels
ery (table 1). Degradation and closed forest loss were to burned patches with a convex hull algorithm (see
reclassified as closed forest disturbance (CFD) and supplemental materials).
open forest loss was reclassified as open forest dis-
turbance (OFD). The canopy cover predictions and 2.1.3. Disturbed and unchanged locations
mapped disturbances were previously validated, and To assess drivers of forest disturbance, disturbed
details and accuracy assessment results are provided locations were contrasted with unchanged locations
in supplemental materials. that did not experience forest loss, degradation, or
recovery. For each year, the binary grids of CFD,
2.1.2. VIIRS active fires OFD, and unchanged pixels were aggregated by a
We used daily VIIRS active fire detections derived factor of three to identify 90 m cells in which all
from the instrument’s I-Band with 375 m nom- nine of the smaller pixels belonged to the given class.
inal spatial resolution (Schroeder et al 2014). The This approach focused our analysis on a 0.81 ha min-
I-Band pixel area is approximately ten-fold smaller imum mapping unit and increased our confidence
3
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
that the locations were dominated by either disturbed appropriate for measuring edge effects into unfrag-
or unchanged forest. The fire data were overlaid on mented tropical forests (Shapiro et al 2016, 2021).
the disturbance grids to assign each disturbed cell Percent closed forest was generated from the 30 m
to one of four categories: (1) fire disturbances in canopy rasters using a 210 m radius circular mov-
closed forests (CFD fire), (2) fire disturbances in open ing window. Slope angle was derived from the 30 m
forests (OFD fire), (3) non-fire disturbances in closed Shuttle Radar Topography Mission (SRTM) digital
forests (CFD non-fire), and (4) non-fire disturbances elevationmodel (DEM) (Farr et al 2007). All variables
in open forests (OFD non-fire). used in the models were rescaled to match the 90 m
resolution of the aggregated forest disturbance raster
2.1.4. Predictor variables grids. All predictor variables are listed in table 2 and
Climate variability was measured using pentad pre- maps of the predictor variables are provided in figure
cipitation records obtained from the Climate Hazards A2 in the supplemental materials.
Group Infrared Precipitationwith Stations (CHIRPS)
at 5 km spatial resolution (Funk et al 2015). Mean 2.2. Data analysis
annual precipitation was calculated using precipita- Annual data on forest disturbances in all forest
tion totals for the 1991–2020 hydrological years. A reserves in southern Ghana were summarized for
hydrological year was defined as the period between 2013–2020 to describe changes and their geographic
May 1st (beginning of rainy season) and April 30th patterns. We used boosted regression tree (BRT)
of the following year (Dwomoh et al 2019). Pixelwise models to analyze the influences of climate variability,
annual standardized anomalies for each hydrological landscape structure, and human accessibility factors
year (yr) from 2014 to 2020 were calculated follow- on forest disturbance. BRT is a machine learning
ing the procedure by Saatchi et al (2013) as departures method that uses ensembles of regression trees to gen-
from the 1991 to 2020 mean, excluding the meas- erate nonparametric models that capture nonlinear
urement from that year (yr), and normalizing by the relationships and interactions among predictor vari-
standard deviation. ables and are robust to outliers andmissing data (Elith
Human accessibility was characterized using the et al 2008).We used the dismo package in R (Hijmans
population gravity index, proximity to roads, and et al 2021) for BRT modeling.
proximity to non-forest areas. The population grav- A separate BRTmodel was fitted for each disturb-
ity index accounts for the size of nearby cities as well ance type (CFD fire, OFD fire, CFD non-fire, and
as their proximity to the forest reserves (Polyakov OFD non-fire). For each model, a random sample
et al 2008). It is highest when large human popula- of 1000 disturbed and 10 000 unchanged locations
tions are located nearby and decreases when popu- (see supplemental materials) was randomly split into
lations are smaller or located further away. We cal- a training (70%) and a validation (30%) set. The fit-
culated the population gravity index for each grid ted models were used to make spatial predictions of
cell using gridded population estimates from World- the probability of each disturbance type using land-
Pop (Leasure et al 2020) and urban boundaries from scape conditions in 2020 and two climate scenarios:
the Africapolis project (OECD/SWAC 2020) (see sup- (1) average precipitation with standardized precip-
plemental materials). We extracted road information itation anomalies set to 0, and (2) extreme drought
from road features acquired fromGlobal Roads Open with standardized precipitation anomalies set to −2.
Access Data Set (CIESIN—Columbia University and Disturbance risk grids for closed and open forests
ITOS—University of Georgia 2013) and computed were combined to obtain continuous surfaces of fire
Euclidean distances from the nearest road. We also and non-fire disturbance risk within the reserves.
calculated Euclidian distance from each reserve pixel The relative importance of each predictor variable
to the nearest non-forest (low tree cover) pixel for was estimated based on the number of times a vari-
each year in 2013–2019. able was selected to create a split, weighted by the
Landscape structure was measured using forest squared improvement to the model resulting from
fragmentation type, percent closed forest, and topo- these splits, and averaged over all trees (Friedman
graphic slope. Fragmentation type was calculated and Meulman 2003, Elith et al 2008). We iden-
using methods described by Vogt et al (2007a) and tified predictors with relative influence above that
implemented by Parent et al (2007). First, forest cover expected by chance (Müller et al 2013), obtained by
from 2013 to 2019 was reclassified into two groups: dividing 100 by the number of predictors (8 in this
forest (a combination of closed and open forest study). We also created partial dependence plots for
classes) and non-forest (low tree cover). We used the the five most important variables for each model
Landscape Fragmentation Tool to classify forest pixels to assess the effect of each variable after accounting
into six groups with varying degrees of fragmenta- for the average effects of all other variables used in
tion: large core (most intact), medium core, small the model (Elith et al 2008). We used the validation
core, inner edge, edge, and patch (most fragmented) datasets to compute the area under the receiver
(Vogt et al 2007a, 2007b, Shapiro et al 2016). We operating characteristic curve (AUC) for the four
used an edge distance of 300 m which is considered models.
4
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Figure 1. Forest disturbances in tropical forest reserves in southern Ghana from 2013 to 2020. (a) Spatial patterns in the
distribution of fire-related vs non-fire disturbances within protected reserves in the eight vegetation zones of southern Ghana.
The vegetation zones are coastal savanna southern marginal (CSSM), dry semi-deciduous fire zone (DSD-FZ), DSD inner zone
(IZ), ME, MSD-NW, MSD-SE, upland evergreen (UE), and WE. (b) Fire and non-fire disturbances zoomed in to the MSD-NW
vegetation zone. (c) Types of disturbances in the MSD-NW vegetation zone. (d) Year of latest disturbance in the MSD-NW
vegetation zone. (e) Total number of disturbances in the MSD-NW vegetation zone. Figures (b)–(e) highlight patterns in
disturbance characteristics in the most disturbed forest zone in southern Ghana. Maps showing results for the whole study area
are available in figure B1 in supplemental materials.
3. Results Forest disturbance was mostly degradation and
open forest loss with less closed forest loss. From
3.1. Patterns of forest change 2013 to 2020, 50% (305 km2) of all fire disturb-
A total of 3562 km2 of forest reserves in southern ances resulted in forest degradation while another
Ghana were disturbed from 2013 to 2020. We estim- 34% (210 km2) led to open forest loss. Over the
ated 17% (615 km2) of this area to be fire disturbances same period, 71% (2102 km2) of all non-fire dis-
and 83% (2946 km2) to be non-fire disturbances. turbances led to forest degradation and another 26%
Most of these disturbances were in the moist semi- (772 km2) resulted in open forest loss. Over the eight-
deciduous northwest (MSD-NW), moist evergreen year period, 184 km2 within the protected reserves
(ME), and MSD southeast (SE) vegetation zones, were disturbed more than once (figure B1, supple-
amounting to 1419 km2, 1110 km2, and 584 km2, mental materials). These were mostly locations that
respectively over the 8 year period (figures 1, 2(c), were first degraded and later experienced open forest
(d), and B1). There was significant fire activity in loss. During that period, 2702 km2 of forest reserves
2016, during which fire disturbances were detected recovered,with 1948 km2 of these recovering to closed
in 449 km2 of the reserves, accounting for 45% of forests (figure B2, supplemental materials). Of all
all disturbances that year and 73% of all fire dis- recovered closed-canopy forests, 85% (1647 km2)
turbances over the 2013–2020 period (figure 2(a)). were found in the ME and MSD (NW and SE) veget-
In the MSD-NW vegetation zone, fires accounted ation zones. Between 2013 and 2020, closed forests
for 72% (327 km2) of the disturbed area in 2016 declined from 9075 km2 to 8374 km2, open forests
and 28% (396 km2) over the entire 8 year period increased from 3517 km2 to 3856 km2 and areas with
(figures 1 and 2(c)). During other years, fire disturb- low tree cover increased from 4177 km2 to 4538 km2
ances were less than 22 km2 per year. (figure B3, supplemental materials).
5
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Figure 2. Distribution of fire vs non-fire disturbances from 2013 to 2020. Areas of fire disturbances (a) and non-fire disturbances
(b) by year. Areas of fire disturbances (c) and non-fire disturbances (d) by vegetation zone. The vegetation zones are CSSM,
DSD-FZ, DSD-IZ, ME, MSD-NW, MSD-SE, UE, and WE. Labels above each bar represent the percentage of the total reserve area
(16 770 km2).
These changes were not linear and there were from non-forest areas, and population gravity index
periods when net changes differed from the overall were among the top five most influential predict-
trend. For example, the areas of forest recovery were ors in all four models, with varying levels of influ-
larger than those of degradation and loss in 2013 and ence (figure 3). Fire disturbances in both open and
2018–2020, resulting in net increases of 472 km2 in closed forests were strongly influenced by both long-
closed forest and 168 km2 in open forest during these term precipitation averages and annual precipita-
years. However, the total area of recovered forests was tion anomalies. In addition, percent closed forest
less than that of degraded and lost forests in 2014– and population gravity index were important pre-
2017, resulting in net losses of 1104 km2 of closed dictors of fire disturbances within closed forests and
forest and 396 km2 of open forest. open forests, respectively. For non-fire disturbances
in closed forests, percent closed forest was the single
3.2. Drivers of forest disturbance most important predictor with a relative importance
BRT models of fire disturbances within closed forests of 70%. Non-fire disturbances in open forests were
had higher accuracy (AUC= 0.97) compared to those influenced by a wider range of variables includ-
in open forests (AUC = 0.91). Models of non-fire ing precipitation anomalies, distance from non-forest
disturbances also had higher accuracy within closed areas, and fragmentation type.
forests (AUC = 0.90) compared to open forests Disturbance risk was highest in drier years (neg-
(AUC = 0.84). In both closed and open forests, ative precipitation anomalies) and in locations with
fire-related disturbances were more accurately pre- mean annual precipitation averaging 1300–1400 mm
dicted than non-fire disturbances. Different types (figure 4). The risk also increased sharply with
of disturbance were influenced by different pre- population gravity index, peaking at around 2000,
dictor variables. Precipitation anomalies, distance above which it quickly diminished and remained low.
6
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Figure 3. Relative importance of the five most influential variables used in predicting disturbance risk in protected reserves of
southern Ghana. (a) and (b) Most important variables predicting fire and non-fire disturbance risk respectively in closed forests.
(c) and (d) Most important variables predicting fire and non-fire disturbance risk respectively in open forests. The dotted line
represents relative influence expected by chance (Müller et al 2013).
Figure 4. Partial dependence of forest disturbance risk on the top five predictors from models of four disturbance types: (a) fire
disturbance within closed forests, (b) non-fire disturbance within closed forests, (c) fire disturbance in open forests, and (d)
non-fire disturbance in open forests. Thin black lines represent the fitted function for predictors and relative forest disturbance
risk. Thick lines represent a smoothed approximation and lines with the same color indicate variables from the same model.
7
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Figure 5. Fire and non-fire disturbance risk surfaces within the MSD-NW region. (a) and (b) Spatial patterns of fire disturbance
risk in the average precipitation and extreme drought scenarios, respectively. (c) and (d) Spatial patterns of non-fire disturbance
risk in the average precipitation and extreme drought scenarios, respectively.
Disturbance risk increased with decreasing closed disturbances of closed and open forests from 2013
canopy forest in the surrounding landscape and was to 2020 were not directly caused by fire, and instead
highest at values less than 25%. Forests within 1 kmof reflected the direct effects of overstory tree removal
non-forest areasweremost likely to be disturbed, with from logging, mining, or agriculture. However, fire
the highest risk observed for those directly adjacent to was a significant disturbance at certain times and
non-forest areas. Fragmented forests had a higher risk locations. During the El Niño–Southern Oscilla-
of being disturbed than more intact forests. tion associated drought of 2016, fires accounted
Disturbances were predicted to be more wide- for 45% (449 km2) of the total disturbed area of
spread in the extreme drought than the average pre- open and closed forests across all forest reserves.
cipitation scenario, reflecting the negative precipit- In the MSD-NW vegetation zone, fires accoun-
ation anomalies during drought events (figure 5). ted for 72% (327 km2) of the disturbed area in
There was also considerable spatial variation in both 2016 and 28% (396 km2) over the entire eight-
scenarios that reflected the effects of human access- year period. Although direct tree mortality result-
ibility and landscape structure. High disturbance risk ing from moisture stress is well documented dur-
occurred in areas where degradation and forest loss ing droughts in the tropics (Phillips et al 2009,
had already occurred, and atmore accessible locations 2010), most of the additional forest disturbance in
along the reserve boundaries. Ghana during the 2016 drought was associated with
fires.
4. Discussion Climate was the strongest driver of fire disturb-
ance with the highest risk observed in years with
Disturbances of closed canopy forests primarily res- negative rainfall anomalies and locations with low
ulted in degradation rather than forest loss, whereas mean annual rainfall. In tropical forests, fuel mois-
forest loss occurred primarily in open forests. Most ture is typically too high to support combustion
8
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
and sustained drought is necessary for forests to factors that influence the rate of disturbance in partic-
burn.Higher fire detections have been associatedwith ular locations. For instance, degradation in theMSD-
reduced precipitation in tropical forests of South- NW forests is related to the abundance of valuable
east Asia (Sloan et al 2017, Sze and Lee 2019) and commercial timber species in this zone (Adam et al
Amazonia (Aragão et al 2008). The MSD-NW zone 2006), and the lower proportions of closed-canopy
of Ghana, where most of the fires occurred, was drier forest in this area may reflect differences in species
than themoist andWE zones further south. Although composition that have made these forests desirable
the 2016 rainfall anomalies were not as extreme in this for logging in the past and in the future. Other stud-
zone as in other parts of Ghana (Dwomoh et al 2019), ies in the Amazon and Southeast Asia have also con-
the reduction in fuel moisture was sufficient to allow cluded that previously disturbed forests are likely to
widespread burning. If droughts become more com- be disturbed again (Cochrane et al 1999, Adrianto
mon because of climate change, increased fire occur- et al 2020,Wang et al 2020,Qin et al 2021) and contin-
rence has the potential to increase forest degrada- ued disturbance can lead forests to shift permanently
tion and loss. Negative precipitation anomalies had a to non-forest states (de Dantas et al 2016).
weaker association with non-fire disturbances, which Amajor strength of this study was the use of high-
likely captured direct effects of drought on tree mor- quality, annual disturbance maps calibrated and val-
tality as well as possible misclassification of burned idated within the study area (Wimberly et al 2022),
areas not captured by the VIIRS active fire data. which were combined with burned area estimates
Human accessibility affects forest disturbance from 375 m VIIRS active fire data to identify fire dis-
through multiple pathways. In Ghana, forests close turbances. Although the scale mismatch likely resul-
to non-forest land cover were at a higher risk of ted in some misclassification of fire and non-fire dis-
being disturbed than forests located further away. turbances, we were able to identify distinctive sets of
The removal of forests is typically associated with drivers for each disturbance type. We focused on pre-
land uses such as agriculture and mining, and nearby dictors that were measurable using geospatial data
locations are therefore susceptible to further human and did not include variables on forest governance
disturbance and spread of fire used for land clear- systems, policies, and logging histories because these
ing. Disturbance risk was also highest at intermedi- data were not accessible.We did not also consider dis-
ate levels of the population gravity index. Proxim- turbances that do not have an instantaneous effect on
ity to dense human populations is associated with canopy density such as the long-term cultivation of
higher demands for natural resources and agricul- crops in forest understories. Nevertheless, ourmodels
tural products, and research in Southeast Asia and classified forest disturbance accurately (AUCs 0.84–
Africa has attributed increased forest disturbance to 0.97). Although we cannot elucidate the proximal
larger or closer settlements (van Khuc et al 2018, Sze causes of forest disturbance, the models do have the
and Lee 2019, Gou et al 2022). However, locations in capability to highlight the locations and climatic con-
our study with the highest population gravity index ditions under which disturbances are most likely.
were very close to large cities, and theymay experience Our results provide new insights into the dis-
less disturbance if there is less use of fire andmore sur- turbance regimes within the forest reserves of Ghana.
veillance for illegal activities. These results emphas- The extent of non-fire disturbance is relatively con-
ize that forest disturbance risk is likely to intensify in stant from year to year. Fire typically affects less area
Ghana due to increasing human populations, rapid than non-fire disturbance but can increase sharply
urbanization, and associated land use and land cover in response to drought. Areas where previous dis-
changes. turbances opened the forest canopy and caused frag-
Past disturbances influence forest and landscape mentation were more susceptible to disturbance than
structure, which in turn affects the likelihood of intact forests, supporting the hypothesis that pos-
future disturbances (Vieira et al 2004). Degrada- itive feedbacks are driving forest degradation and
tion reduces canopy cover, tree density, and biomass, loss (Dwomoh and Wimberly 2017). Understanding
while forest loss alters vegetation structure andmicro- these dynamics is important for conservation and
climate at forest edges. These changes increase the forest restoration activities inGhana and similar trop-
probability of fire by allowing more solar radiation ical forest regions. Models based on climate variabil-
into the forest understory, which can increase surface ity, human accessibility, and landscape structure can
fuel loads and decrease fuel moisture (Cochrane et al identify where and when disturbance risk is highest
1999, Laurance and Williamson 2001). Open forests, and help target these actions accordingly.
forest edges, and flat terrain can be preferred for land
uses such as farming and logging (Busch and Ferretti- Data availability statement
Gallon 2017, Edwards et al 2019) because less effort is
required for land clearing. The forest structure vari- The data that support the findings of this study are
ables thatwe usedmay also be proxies for unmeasured available upon reasonable request from the authors.
9
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
Funding patterns of fire-induced forest degradation in Amazonia
Environ. Res. Lett. 12 1–12
This work was supported by the National Aeronaut- Dwomoh F K and Wimberly M C 2017 Fire regimes and forest
resilience: alternative vegetation states in the West African
ics and Space Administration Carbon Cycle Science tropics Landsc. Ecol. 32 1849–65
Program (Grant 80NSSC21K1714). Dwomoh F K, Wimberly M C, Cochrane M A and Numata I 2019
Forest degradation promotes fire during drought in moist
ORCID iDs tropical forests of Ghana For. Ecol. Manage. 440 158–68
Edwards D P, Socolar J B, Mills S C, Burivalova Z, Koh L P and
Wilcove D S 2019 Conservation of tropical forests in the
Dan Wanyama https://orcid.org/0000-0002-4844- Anthropocene Curr. Biology 29 R1008–20
8803 Elith J, Leathwick J R and Hastie T 2008 A working guide to
Michael C Wimberly https://orcid.org/0000-0003- boosted regression trees J. Anim. Ecol. 77 802–13
1549-3891 Farr T G et al 2007 The shuttle radar topography mission Rev.
Geophys. 45 1–33
Foster Mensah https://orcid.org/0000-0002-2839-
Flores B M and Staal A 2022 Feedback in tropical forests of the
1782 Anthropocene Glob. Change Biol. 28 5041–61
Friedman J H and Meulman J J 2003 Multiple additive regression
References trees with application in epidemiology Stat. Med.
22 1365–81
Acheampong E Insaidoo T F G and Ros-Tonen M A F 2016 Funk C et al 2015 The climate hazards infrared precipitation with
Management of Ghana’s modified taungya system: stations—A new environmental record for monitoring
challenges and strategies for improvement Agrofor. Syst. extremes Sci. Data 2 1–21
90 659–74 Gibbs H K, Ruesch A S, Achard F, Clayton M K, Holmgren P,
Adam K A, Pinard M A and Swaine M D 2006 Nine decades of Ramankutty N and Foley J A 2010 Tropical forests were the
regulating timber harvest from forest reserves and the status primary sources of new agricultural land in the 1980s and
of residual forests in Ghana Int. For. Rev. 8 280–96 1990s Proc. Natl Acad. Sci. USA 107 16732–7
Adrianto H A, Spracklen D V, Arnold S R, Sitanggang I S and Gou Y, Balling J, De Sy V, Herold M and de Keersmaecker W 2022
Syaufina L 2020 Forest and land fires are mainly associated Intra-annual relationship between precipitation and forest
with deforestation in Riau Province, Indonesia Remote Sens. disturbance in the African rainforest Environ. Res. Lett.
12 1–12 17 1–13
Amissah L, Mohren G M J, Bongers F, Hawthorne W D and Güneralp B, Seto K C and Ramachandran M 2013 Evidence of
Poorter L 2014 Rainfall and temperature affect tree species urban land teleconnections and impacts on hinterlands
distribution in Ghana J. Trop. Ecol. 30 435–46 Curr. Opin. Environ. Sustain. 5 445–51
Aragão L E O C, Malhi Y, Barbier N, Lima A, Shimabukuro Y, Hall J B and Swaine M D 1976 Classification and ecology of
Anderson L and Saatchi S 2008 Interactions between closed-canopy forest in Ghana J. Ecol. 64 913–51
rainfall, deforestation and fires during recent years in the Hall J B and Swaine M D 1981 Distribution and Ecology of Vascular
Brazilian Amazonia Phil. Trans. R. Soc. B 363 1779–85 Plants in a Tropical Rain Forest: Forest Vegetation in Ghana
Aragão L E O C, Malhi Y, Roman-Cuesta R M, Saatchi S, vol 1, ed M J A Werger (Dordrecht: Springer)
Anderson L O and Shimabukuro Y E 2007 Spatial patterns Harris N L et al 2021 Global maps of twenty-first century forest
and fire response of recent Amazonian droughts Geophys. carbon fluxes Nat. Clim. Change 11 234–40
Res. Lett. 34 1–5 Herrmann S M, Brandt M, Rasmussen K and Fensholt R 2020
Boadi S, Nsor C A, Antobre O O and Acquah E 2016 An analysis Accelerating land cover change in West Africa over four
of illegal mining on the Offin shelterbelt forest reserve, decades as population pressure increased Commun. Earth
Ghana: implications on community livelihood J. Sustain. Environ. 1 1–10
Min. 15 115–9 Hijmans R J, Phillips S, Leathwick J and Elith J 2021 Package
Brando P M, Paolucci L, Ummenhofer C C, Ordway E M, dismo: species distribution modeling (available at: https://
Hartmann H, Cattau M E, Rattis L, Medjibe V, Coe M T and mirror.linux.duke.edu/cran/web/packages/dismo/dismo.
Balch J 2019 Droughts, wildfires, and forest carbon cycling: a pdf)
pantropical synthesis Annu. Rev. Earth Planet. Sci. 47 555–81 Kennedy R E, Yang Z, Gorelick N, Braaten J, Cavalcante L,
Busch J and Ferretti-Gallon K 2017 What drives deforestation and Cohen W B and Healey S 2018 Implementation of the
what stops it? A meta-analysis Rev. Environ. Econ. Policy LandTrendr algorithm on Google earth engine Remote Sens.
11 3–23 10 1–10
Carvalho R, Adami M, Amaral S, Bezerra F G and de Kouassi J L, Gyau A, Diby L, Bene Y and Kouamé C 2021
Aguiar A P D 2019 Changes in secondary vegetation Assessing land use and land cover change and farmers’
dynamics in a context of decreasing deforestation rates in perceptions of deforestation and land degradation in
Pará, Brazilian Amazon Appl. Geogr. 106 40–49 south-west Côte d’Ivoire, West Africa Land 10 1–25
CIESIN—Columbia University and ITOS—University of Georgia Laurance W F and Williamson B G 2001 Positive feedbacks
2013 Global roads open access data set, version 1 among forest fragmentation, drought, and climate change in
(gROADSv1) (Palisades, NY: NASA Socioeconomic Data the Amazon Biol. Conserv. 15 1529–35
and Applications Center (SEDAC)) (https://doi.org/ Leasure D and Tatem A (Worldpop) 2020 Bayesian gridded
10.7927/H4VD6WCT) population estimates for Ghana 2018, version 1.0 (available
Cochrane M A, Alencar A, Schulze M D, Souza Jr C M, at: https://eprints.soton.ac.uk/443566/)
Nepstad D C, Lefebvre P and Davidson E A 1999 Positive Lewis S L, Edwards D P and Galbraith D 2015 Increasing human
feedbacks in the fire dynamic of closed canopy tropical dominance of tropical forests Science 349 827–32
forests Science 284 1832–5 Malhi Y, Gardner T A, Goldsmith G R, Silman M R and
Cochrane M A and Laurance W F 2002 Fire as a large-scale edge Zelazowski P 2014 Tropical forests in the Anthropocene
effect in Amazonian forests J. Trop. Ecol. 18 311–25 Annu. Rev. Environ. Resour. 39 125–59
de Dantas V, Hirota L, Oliveira M, S R and Pausas J G 2016 Malhi Y and Wright J 2004 Spatial patterns and recent trends in
Disturbance maintains alternative biome states Ecol. Lett. the climate of tropical rainforest regions Phil. Trans. R. Soc.
19 12–19 B 359 311–29
De Faria B L, Brando P M, Macedo M N, Panday P K, Matricardi E A T, Skole D L, Costa O B, Pedlowski M A,
Soares-Filho B S and Coe M T 2017 Current and future Samek J H and Miguel E P 2020 Long-term forest
10
Environ. Res. Lett. 18 (2023) 064022 DWanyama et al
degradation surpasses deforestation in the Brazilian Forest condition in the Congo Basin for the assessment of
Amazon Science 369 1378–82 ecosystem conservation status Ecol. Indic. 122 1–16
Müller D, Leit̃ao P J and Sikor T 2013 Comparing the Sloan S, Locatelli B, Wooster M J and Gaveau D L A 2017 Fire
determinants of cropland abandonment in Albania and activity in Borneo driven by industrial land conversion and
Romania using boosted regression trees Agric. Syst. drought during El Niño periods, 1982–2010 Glob. Environ.
117 66–77 Change 47 95–109
Myers N, Mittermeier R A, Mittermeier C G, da Fonseca G A B Swaine M D 1992 Characteristics of dry forest in West Africa and
and Kent J 2000 Biodiversity hotspots for conservation the influence of fire J. Veg. Sci. 3 365–74
priorities Nature 403 853–8 Sylla M B, Elguindi N, Giorgi F and Wisser D 2016 Projected
Norris K, Asase A, Collen B, Gockowksi J, Mason J, Phalan B and robust shift of climate zones over West Africa in response to
Wade A 2010 Biodiversity in a forest-agriculture anthropogenic climate change for the late 21st century Clim.
mosaic—The changing face of West African rainforests Biol. Change 134 241–53
Conserv. 143 2341–50 Sze J S and Lee J S H 2019 Evaluating the social and
OECD/SWAC 2020 Africa’s Urbanisation Dynamics 2020 (OECD) environmental factors behind the 2015 extreme fire event in
(available at: www.oecd-ilibrary.org/development/africa-s- Sumatra, Indonesia Environ. Res. Lett. 14 1–14
urbanisation-dynamics-2020_b6bccb81-en) UN DESA Population Division 2022 UN population division data
Parent J, Civco D and Hurd J 2007 Simulating future forest portal: interactive access to global demographic indicators
fragmentation in a Connecticut region undergoing (available at: https://population.un.org/dataportal/home)
suburbanization ASPRS Annual Conf. 2007: Identifying van Khuc Q, Tran B Q, Meyfroidt P and Paschke MW 2018
Geospatial Solutions vol 2 (Tampa, Florida, USA) Drivers of deforestation and forest degradation in Vietnam:
Phillips O L et al 2009 Drought sensitivity of the Amazon an exploratory analysis at the national level For. Econ. Policy
rainforest Science 323 1344–7 90 128–41
Phillips O L et al 2010 Drought-mortality relationships for Vieira S et al 2004 Forest structure and carbon dynamics in
tropical forests New Phytol. 187 631–46 Amazonian tropical rain forests Oecologia 140 468–79
Polyakov M, Majumdar I and Teeter L 2008 Spatial and temporal Vogt P, Riitters K H, Estreguil C, Kozak J, Wade T G and
analysis of the anthropogenic effects on local diversity of Wickham J D 2007a Mapping spatial patterns with
forest trees For. Ecol. Manage. 255 1379–87 morphological image processing Landsc. Ecol. 22 171–7
Qin Y et al 2021 Carbon loss from forest degradation exceeds that Vogt P, Riitters K H, Iwanowski M, Estreguil C, Kozak J and
from deforestation in the Brazilian Amazon Nat. Clim. Soille P 2007b Mapping landscape corridors Ecol. Indic.
Change 11 442–8 7 481–8
Saatchi S, Asefi-Najafabady S, Malhi Y, Aragão L E O C, Wang Y, Ziv G, Adami M, de Almeida C A, Antunes J F G,
Anderson L O, Myneni R B and Nemani R 2013 Persistent Coutinho A C, Esquerdo J C D M, Gomes A R and
effects of a severe drought on Amazonian forest canopy Proc. Galbraith D 2020 Upturn in secondary forest clearing
Natl Acad. Sci. USA 110 565–70 buffers primary forest loss in the Brazilian Amazon Nat.
Schroeder W, Oliva P, Giglio L and Csiszar I A 2014 The new Sustain. 3 290–5
VIIRS 375 m active fire detection data product: algorithm Wimberly M C, Dwomoh F K, Numata I, Mensah F, Amoako J,
description and initial assessment Remote Sens. Environ. Nekorchuk D M and McMahon A 2022 Historical trends of
143 85–96 degradation, loss, and recovery in the tropical forest reserves
Shapiro A C, Aguilar-Amuchastegui N, Hostert P and Bastin J F of Ghana Int. J. Digit Earth 15 30–51
2016 Using fragmentation to assess degradation of forest Zhang T, Wooster M J and Xu W 2017 Approaches for
edges in Democratic Republic of Congo Carbon Balance synergistically exploiting VIIRS I- and M-Band data in
Manage. 11 1–15 regional active fire detection and FRP assessment: a
Shapiro A C, Grantham H S, Aguilar-Amuchastegui N, demonstration with respect to agricultural residue burning
Murray N J, Gond V, Bonfils D and Rickenbach O 2021 in Eastern China Remote Sens. Environ. 198 407–24
11