Environmental Challenges 10 (2023) 100664

Contents lists available at ScienceDirect

Environmental Challenges
journal homepage: www.elsevier.com/locate/envc

Land use and land cover change detection and prediction based on
CA-Markov chain in the savannah ecological zone of Ghana
Philip Aniah a,∗, Simon Bawakyillenuo b, Samuel Nii Ardey Codjoe c, Fred Mawunyo Dzanku b
a
Department of Sustainable Development Studies, University for Development Studies, P. O. Box TL 1350, Tamale, Northern Region, Ghana
b
Institute of Statistical, Social and Economic Research, University of Ghana, Box LG 74, Legon, Accra, Ghana
c
Regional Institute for Population Studies, University of Ghana, Legon, Ghana

a r t i c l e i n f o a b s t r a c t

Keywords: Environmental problems have accompanied the accelerated land use and land cover change (LULCC), yet few
LULCC local level studies make an attempt to assess the dynamics of LULCC. This work employed GIS and remote sensing
Dynamics to quantify the past and predict future dynamics of LULCC based on the synergy Cellular Automata (CA) - Markov
CA-Markov model
Chain Model (MCM). The results revealed that agricultural land in the Bongo district witnessed the greatest expan-
Remote sensing
sion from 10.03% to 27.17% of total area from 1990 to 2019, while wooded savannah area witnessed the greatest
Savannah zone
Ghana decline from a share of 42.26% to 15.51% of total area from 1990 to 2019. In the Kassena-Nankana West (KNW)
district, shrub and tree savannah and agricultural land expanded from 32.91% to 54.2% and 9.44% to 18.16%
of the total area, respectively, at the expense of wooded savannah area (-32.9% of total area) between 1990 and
2019. Future predictions based on prevailing socio-economic development demonstrate that the observed trend
would continue till the 2050 period. In the Bongo district, the settlement area will witness the highest proportion
of net increase in total area (5.63 km2 ) at the expense of wooded savannah (-11.26 km2 ) between 2019 and
2050. Conversely, in the KNW district, the shrub and tree savannah area will experience the highest proportion
of net gain in total area (156.02 km2 ) at the expense of wooded savannah area (-111.49 km2 ) between 2019 and
2050. This result is an indication that the synergy CA-MCM have effectively captured the spatiotemporal trend
in LULCC in this study.

1. Introduction ica, Southeast Asia) are currently witnessing an exponential shift in land-
scape and associated ecosystem depletion (Ellis et al., 2013; Peng et al.,
For a very long time, humanity has transformed landscapes in their 2017; Wang et al., 2021).
efforts to enhance the quantity, quality, and security of environmental The accelerated landscape changes have been accompanied by lo-
resources vital to human well-being, such as medicinal products, food, cal and global environmental problems (Meyfroidt et al., 2013), raising
fibre and freshwater (Potts et al., 2016; Wang et al., 2021). Through the fears that the sustainable limits of the planet earth to support the human
increased use of technology and innovation, human populations have, population without causing irreparable damage has reached its tipping
slowly at first, and at a progressively rapid pace, later on, increased their point and threshold (Meadows & Randers, 2012). Various reports of the
capacity to obtain diverse services from the ecosystem and expand its Millennium Ecosystem Assessment team have added further credibil-
territory (Song et al., 2018). ity to the indications of apparent global landscape changes (MA, 2003,
The magnitude and rate of anthropogenic activities on the land sur- 2005). The key underlying trigger of past and recent global environ-
face have accelerated during the last 300 years (Potts et al., 2016). More mental changes has been anthropgenic activities on the land surface
land has been converted for human use than before, and already con- (Meyfroidt et al., 2013).
verted land in the earlier times was managed more intensively to in- In the context of this paper, land cover denotes the physical and
crease the yields of agricultural and forest products. The phenomenon biological attributes of the land surface. Land use on the other hand de-
of landscape change has also shifted spatially over time. While Europe, notes any human management activity associated with the land. Land-
Asia and America experienced significant landscape change in time past use change therefore implies a shift from one land use to another or the
(around the 20th century) and have now witnessed a steady trend in for- intensification of the existing land use (Kleeman et al., 2017). Land use
est land cover, Africa and other developing countries (e.g. Latin Amer- and land cover change (LULCC) directly mirror the utilisation of land

∗
Corresponding author.
E-mail address: [email protected] (P. Aniah).

https://doi.org/10.1016/j.envc.2022.100664
Received 14 November 2022; Received in revised form 22 November 2022; Accepted 6 December 2022
2667-0100/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Fig. 1. Location of the study areas.

resources in an area and is an important component of the environmen- paper. Section 3 provides results and discussion. Section 4 presents con-
tal systems (Kleeman et al., 2017). LULCC has emerged as a threat to clusion of the study.
social and ecological systems (Lambin and Meyfroidt, 2010) and can
cause many ecological crises at various spatial scales such as shortages
in natural resources, widespread and irreversible losses of biodiversity 2. Methodological approaches
(Tolessa et al., 2017; Karki et al., 2018; Bounoua et al., 2018; Lu et al.,
2019). 2.1. Study area
Generally, several studies on LULCC dynamics in Africa have been
carried out (Brandt et al., 2016; Greiner, 2016; Koranteng et al., 2016; This study was conducted in the Upper East Region (UER) in the
Tolessa et al., 2017; Timm-Hoffman et al., 2018). For instance, docu- savannah ecological zone of Ghana. The extension of the study area
mentation of LULCC using satellite data have been carried out in Africa lies between longitude 0° and 1° West, and latitudes 10° 30′N and 11°N
by Mohajane et al. (2018) and Tavares et al. (2019). Despite the many (GSS, 2021). Two case study districts (Bongo and Kassena-Nankana West
studies on LULCC dynamics in Africa, information about LULCC dynam- (KNW) districts) were selected (Fig. 1). Due to the semi-arid and physical
ics in the savannah ecological zone of Ghana remains scanty. Other features of the zone, it is extremely vulnerable to climate and ecolog-
studies on LULCC in Ghana (e.g. Basommi et al., 2015; Kleeman et al., ical changes (GSS, 2021). The increasing effects of environmental and
2017; Antwi-Agyei et al., 2019; Osumanu et al., 2019; Larbi et al., 2019; socio-economic factors, such as LULCC, climate and ecological change,
Tuffour-Mills et al., 2020; Osumanu & Akomgbangre, 2020) have been drought, floods, and bushfires (Klutse et al., 2020) presents severe ef-
conducted on a broader scale with a larger spatial recommendation do- fects on the livelihoods of rural dwellers most especially rain-fed agri-
main, and usually do not differentiate between vulnerable and resilient culture dependant households (Boafo et al., 2020). The primary reason
zones. Thus, there is the need to conduct location-specific and local level underlying the selection of the study region (UER) and districts is the
landscape change assessments to fill this gap. Such information would depiction of typical cases of LULCC through a Normalized Difference
be useful in managing landscapes and resources and inform future lo- Vegetation Index (NDVI) analysis of the UER. Bongo District is selected
cal level policies required for addressing peculiar local socio-ecological because it is the most vulnerable area in the UER to severe LULCC since
needs. the 1980s. Conversely, KNW District is the area in the UER that largely
Therefore, this paper seeks to fill this research gap by analysing and contains very high amounts of vegetated cover in the form of gallery
predicting LULCC dynamics from 1990 and 2050 in the two distinct and riparian forest, and was therefore selected as a resilient district to
study sites within the Savanna ecological zone. Following the presenta- LULCC (Fig. 1). KNW District’s population density of 97.64/km2 is the
tion of the introduction in Section 1, Section 2 provides a brief profile lowest compared to Bongo District’s density of 289.8/km2 and the re-
of the study area and the methodological approaches employed in this gional (UER) density of 129/km2 .

2
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Table 1 2022). We employed fixed mtry. The preference of avoiding the possi-
Details of remote sensing data used in this study. bility of overfitting the trained model was the key guiding principle that
Bongo informed our choice of the hyperparameters. The trained model was sub-
sequently employed to predict LULCC classes in the test datasets. The
Year Satellite Sensor Date Bands used Spatial Resolution
RFA performance was examined by employing the user accuracy values
1990 Landsat 4 MSS 30/11/1990 2,3,4 30m and the kappa coefficients (Congalton, 1991; Hayes et al., 2014). A sep-
2000 Landsat 7 ETM 09/11/2000 2,3,4 30m arate RFA classification was undertaken on the same dataset to further
2010 Landsat 5 TM 29/11/2010 2,3,4 30m
2019 Landsat 8 OLI 22/11/2019 3,4,5 30m
assess the performance of the RFA classifier. Several trials with different
mtry (e.g. 3, 6, 9) and ntree (e.g. 200, 700, 900) parameterizations were
KNW
undertaken to asses out-of-bag (OOB) error levels.
Year Satellite Date Bands used Spatial Resolution Lastly, the RFA was trained at ntree value of 600 with a default set-
1990 Landsat 4 TM 13/11/1990 2,3,4 30m ting for mtry and this helped produce a low OOB error level. We then
TM 22/11/1990 2,3,4 30m employed the trained model to predict the 6 LULCC classes based on the
2000 Landsat 7 ETM 09/11/2000 2,3,4 30m test datasets for every study year. To determine genuine LULCC from
ETM 09/11/2000 2,3,4 30m likely inaccurate changes due to errors in classification, per class accu-
2010 Landsat 5 TM 29/11/2010 2,3,4 30m
racy indices and error matrices were calculated. We employed Google
TM 20/11/2010 2,3,4 30m
2019 Landsat 8 OLI 22/11/2019 3,4,5 30m Earth images to randomly sample 150 points for every study year and
OLI 29/11/2019 3,4,5 30m their equivalent LULCC classes were determined. The validated classes
were employed to assess producer and user accuracy (Abdullah et al.,
2019; Santos et al., 2022). The following LULCC classes were identified:
2.2. Acquisition of satellite imagery settlement, agriculture, bare soil, water bodies, shrub and tree savan-
nah, and wooded savannah. Four LULCC maps were then generated for
Landsat Multispectral Scanner (MSS), Thematic Mapper (TM), En- each of the 2 study sites with a spatial resolution of 30 m.
hanced Thematic Mapper (ETM+) and Operational Land Imager (OLI)
satellite data were used in this study. These were downloaded from the 2.4. Accuracy assessment
USGS Global Visualization Viewer website (Table 1). To better visualise
roads and rivers, Google Earth was used to create an overlapping vector It is important to perform an accuracy assessment between reference
layer of existing roads and rivers for use in the display of the results. data and classified data.The accuracy of 1990, 2000, 2010 and 2019
LULCC maps were assessed by independent datasets. The kappa coeffi-
2.3. Data pre-processing and classification cient, overall accuracy and producer’s and user’s accuracy were calcu-
lated from the error matrix. The dynamics of LULCC for each target year
To achieve accurate surface information from satellite data, it is es- 1990, 2000, 2010 and 2019 was determined by calculating the quantity
sential to apply radiometric and atmospheric corrections. Radiomet- of each specific class category per time window (Butt et al., 2015) after
ric and atmospheric corrections were performed using the ENVI 5.2 we performed the accuracy assessment for the generated LULCC maps
platform prior to the classification. The imagery scenes were then mo- between 1990 and 2019. Cross-tabulation matrices were performed for
saicked, and the study area extracted. Our study combined various bands change detection.
such as RGB 5, 4, 3 for OLI and RGB 4, 3, 2 for TM to facilitate the Table 2 illustrate the overall accuracy, and the kappa statistics for
gathering of training data required for the classification of the images. the years 1990, 2000, 2010 and 2019 for the Bongo and KNW districts.
Also, in describing the feature classes, we made use of historical and ex- The results revealed that the overall accuracy and kappa statistics for the
isting expert knowledge on the physical patterns of the study sites and 1990, 2000, 2010 and 2019 images were 85% (0.7984), 78% (0.7098),
other relevant supplementary data we deemed useful were incorporated 84% (0.7990) and 91% (0.8838), respectively for the Bongo district
(Khwarahm et al., 2021). (Table 2). The overall accuracy and kappa statistics for the KNW dis-
The Random Forest Algorithm (RFA), a non-parametric supervised trict for the years 1990, 2000, 2010 and 2019 were 82% (0.7569), 87%
classification method was employed in our study. The RFA produces (0.8317), 87% (0.8356) and 90% (0.8664), respectively (Table 2).
several classification trees that makes contribution through a voting
system for classification of data. RFA can produce moderate accuracy 2.5. Markov chain model, cellular automata (CA) model and CA-Markov
land cover maps even when employed with coarse data (Breiman, 2001; models
Abdullah et al., 2019; Santos et al., 2022). Our study employed the RFA
package in R to classify LULCC maps for 4 years (1990, 2000, 2010 and The CA methods are tools for imitating complex spatial processes
2019) for the 2 study sites. Six (6) LULCC classes were identified based founded on simple decision rules. The Markov Chain model (MCM) ex-
on researchers’ prior local knowledge, physiographical knowledge of plains and reveals the quantification of conversion states and the trans-
the study areas, visual interpretation using google earth historical func- fer rate between different land-use types (Khwarahm et al., 2021). The
tion and supportive supplementary data. The Google Earth images were MCM, which depends on the stochastic process, has been widely ac-
employed to extract reference data and a random sampling technique cepted to stimulate and predict LULCC over time (Guan et al., 2011).
was used to gather 150 ground control points. The data were divided LULCC prediction normally depend on the transition probability and the
into test (70%) and training (30%) datasets. The RFA was fitted on the transition area matrixes from time one (t0) to time two (t1) as well as
training data for every study year using the number of decision trees the generated LULCC maps (Guan et al., 2011; Khwarahm et al., 2021).
(ntree) = 600 and the number of input features (mtry) = 3 (Santos et al., The CA model, which consist of even grid-cells with each single grid-cell

Table 2
Accuracy assessment for Bongo and KNW Districts for the years 1990, 2000, 2010 and 2019 classification.

Bongo district KNW district

1990 2000 2010 2019 1990 2000 2010 2019

Overall accuracy 85.0 78.0 84.0 91.0 82.0 87.0 87.0 90.0
Kappa Statistic 0.7984 0.7098 0.7990 0.8838 0.7569 0.8317 0.8356 0.8664

3
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Fig. 2. Simulated and actual LULCC maps for Bongo District.

in a fixed number of positions, is capable of characterizing non-linear 2.6. Model validation

spatially distributed LULCC class categories (Mishra & Rai, 2016). The
main principle of the CA model is that changes in cell condition can Model validation is a critical step in evaluating the accuracy of pre-
be described by current condition and changes in adjacent cells based dicted data. In this paper, the kappa coefficient was calculated, and the
on the previous condition (Mitsova et al., 2011; Liping et al., 2018). model was validated after simulating the 2019 LULCC conditions using
Combining the MCM with the CA methods solves the limitation of the 1990, 2000, 2010 and 2019 LULCC maps. The accuracy of the simu-
MCM by adding a spatial dimension to the CA model (Mitsova et al., lated 2019 LULCC map was assessed by estimating the degree of agree-
2011; Khwarahm et al., 2021). The CA–Markov chain model is an effec- ment between modelled and reference map of 2019 using the Kappa In-
tive combination between the MCM and the CA model for prediction dex of Agreement. The Kappa indices (Kappa for no information (Kno),
of spatial LULCC. Therefore, the integrated CA–Markov chain model Kappa for location (Klocation), and Kappa for standard (Kstandard) as
can be used to effectively predict spatiotemporal changes in LULCC expressed in Eqs. (1) to (3) determine the overall success rate of the
(Halmy et al., 2015). Transition suitability images and Markov tran- results. A Kappa coefficient of between 0.75 and 1 implies a high agree-
sition parameters which determines probability of change and the fil- ment while a Kappa coefficient greater than 0.5 but less than 0.75 falls
ter contiguity description (kernel size 4 4, 6 6, 8 8) as well as number within the medium agreement range. However, if Kappa coefficient is
of iterations (about 600 iterations) were used (Arsanjani et al., 2011; less than 0.5, it implies there is rarely any agreement. After obtaining
Guan et al., 2011). Many categories of filter contiguity and several it- successful Kappa values, the CA–Markov model was employed to sim-
erations were assessed to obtain the best kernel dimension (6 6) and ulate the LULCC maps of 2030, 2040, and 2050. The equations below
the iterations number (400 iterations). The simulated LULCC map for express the summary statistics for the Kappa variations according to
2019 was compared with the actual LULCC map of 2019. Kappa index Omar et al. (2014).
was employed to validate the accuracy of the simulated maps. The CA-
Markov model was employed to predict the LULCC in the years 2030, (
2040 and 2050 for the Bongo and KNW districts based on 1990, 2000, M(𝑚)N(𝑛)
Kno = − 𝑁 (𝑛) (1)
2010 and 2019 LULCC maps. Р (𝑝 )

4
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Fig. 3. Simulated and actual LULCC maps for KNW District.

(
M(𝑚)𝑁 (𝑛) 3.2. Dynamics in LULCC in Bongo and KNW Districts
Klocation = − 𝑁 (𝑛) (2)
𝑃 (𝑝 )
The total land area of Bongo District is about 41,647.53 Ha. Around
(
𝑀 (𝑚)𝑁 (𝑛) the year of 1990, shrub and tree savannah (45.88%) and wooded sa-
AndKstandard = − 𝑁 (𝑛) (3) vannah (42.26%) areas occupied the greatest portion of the total land
𝑃 (𝑝 )
area (Fig. 4). Agriculture land was the third greatest land use and cover
where no information is defined by N(n), medium grid cell-level infor- form (10.03% of total area). By the turn of 2019 as shown in Fig. 4, agri-
mation by M(m), and perfect grid cell-level information across the land- culture land witnessed the greatest expansion (27.17%) at the expense
scape by P(p). of wooded savannah (15.51% of total area) between 1990 and 2019.
Settlement, bare soil, and water bodies increased by about three-fold.
3. Results and discussion Conversely, the total land area of KNW District is approximately
83,992.85 Ha. Wooded savannah and shrub and tree savannah consti-
3.1. Classification accuracy tuted 56.16% and 32.91% of total area, respectively in 1990 (Fig. 5).
Agriculture area occupied 9.44% (7,931.61 Ha) of the total land area
Figs. 2 and 3 show a comparison of the simulated and actual maps while settlement, bare soil, and water bodies (0.35%, 0.45% and 0.69%,
in 2019 for both Bongo and KNW Districts. In the simulated map for the respectfully) each occupied less than 1% of the total land area in 1990.
Bongo District, the kappa values of Kno = 0.7794, Klocation = 0.7903, However, by the 2019 period, Wooded savannah area significantly de-
and Kstandard = 0.7409 illustrate a good performance of the model. clined to 23.26% of total land area (net loss of 32.9%). Conversely,
Similarly, In the simulated map for the KNW District, the Kappa coef- shrub and tree savannah area expanded to 54.2% of the total area,
ficients of Kno = 0.7331, Klocation = 0.6823, and Kstandard = 0.6247 while agricultural area expanded to 18.16% of total land area. Similar
also demonstare good performance of the model. The good agreement trends have been observed in previous studies (Osumanu et al., 2019;
of all Kappa values confirm that the model has enough ability to forecast Larbi et al., 2019; Tuffour-Mills et al., 2020; Osumanu and Akomgban-
future LULCC dynamics soundly. gre, 2020). Declines in wooded savannah areas is more noticeable in

5
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Fig. 4. LULCC map of Bongo District (1990-2019).

Table 3
Class percentage change for 1990-2019 in Bongo and KNW districts.

Bongo District KNW District

1990-2000 2000-2010 2010-2019 1990-2000 2000-2010 2010-2019

LULCC Type
Change Percent Change Percent Change Percent Change Percent Change Percent Change Percent
(Ha) Change (Ha) Change (Ha) Change (Ha) Change (Ha) Change (Ha) Change

Settlement 61.56 0.15 540.99 1.31 385.83 0.93 401.49 0.48 573.03 0.68 69.3 0.08
Agriculture 3398.94 8.16 2449.81 5.88 1289.6 3.10 3900.29 4.64 1951.3 2.32 1472.9 1.75
Bare Soil 264.24 0.64 148.05 0.36 754.11 1.81 –98.91 –0.12 –63.9 –0.08 907.83 1.08
Water Body 158.4 0.38 9.9 0.02 630 1.51 167.22 0.20 455.04 0.54 16.02 0.02
Shrub and Tree –416.7 –1.00 –1680.9 –4.04 3148.9 7.56 –3303.9 –3.93 12707.5 15.13 8479.9 10.10
Savannah
Wooded Savannah –3466.5 –8.32 –1467.8 –3.52 –6208.49 –14.91 –1067.2 –1.27 –15621.9 –18.61 –10945.9 –13.03

Bongo District than KNW District, owing to land use intensification, high tlement area increased by 0.15% between 1990 and 2000, by 1.3% from
population density and smaller total land area compared to KNW Dis- 2000 to 2010 and by 0.93% from 2010 to 2019. The settlement area in
trict (Larbi et al., 2019). The findings of this study also mirror that of KNW District increased by 0.48% between 1990 and 2000. Settlement
Wardell et al. (2003) and Kleemann et al. (2017) which disclosed de- area in KNW District continued to expand from 2000 to 2010 and 2010
clining trends in wooded savannah areas and marked expansion in agri- to 2019 by 0.68% and 0.08%, respectively. Distinct expansion in settle-
culture and settlement areas since the 1980s in the UER. The minimal ment area occurred in Bongo District than KNW District from 2000 to
regeneration of fallow areas to wooded savannah area occurred around 2019.
river blindness-endemic areas along the sissili river in the KNW District. Agricultural area in Bongo District witnessed significant expansion
of 8.16% (representing 3398.94 Ha) of total land area between 1990
3.3. Land use and land cover class percentage change for Bongo and KNW and 2000. From 2000 to 2010 and 2010 to 2019, agriculture area in
Districts between 1990 and 2019 Bongo District increased by 5.9% (2449.81 Ha) and 3.1% (1289.6 Ha) of
total land area, respectively. Similarly, in KNW District, the mainstream
Table 3 presents LULCC in hectares and percentage change for Bongo LULCC class expansion was seen in agricultural land, which expanded
and KNW Districts between 1990 and 2019. In Bongo District, the set- by 4.6% of the total land area (3900.3 Ha) from 1990 to 2000 and by

6
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Fig. 5. LULCC map of KNW District (1990-2019).

2.3% (1951.3 Ha) between 2000 and 2010. From 2010 to 2019, the to significant declines in wooded savannah -18.6% (-15621.9 Ha) from
agriculture area increased by 1.8% of the total land area. Although the 2000 to 2010 and -13% (-10945.9 Ha) from 2010 to 2019.
agriculture area witnessed the highest expansion relative to other land Analogous to this present study, Ruelland et al. (2010) observed that
use and cover types in both study districts, Bongo District witnessed a wooded savannah and shrub and tree savannah areas declined from 34%
higher expansion of agricultural land area relative to the total land area and 18% of total land area in 1967 to 11% and 6%, respectively, of to-
of Bongo District than KNW District between 1990 and 2019. tal land area in 2003 in the UER. Agriculture and degraded areas in-
Bare soil and water bodies increased by less than 1% between 1990 creased from 11% to 23% of total land area and 9% to 27% of total land
and 2010 in Bongo District. However, both bare soil and water bod- area, respectively, between 1967 and 2003 in the UER (Ruelland et al.,
ies increased by 1.8% and 1.5%, respectively, between 2010 and 2019. 2010; Larbi et al., 2019). This findings further confirm the widespread
Bare soil in KNW District decreased by -0.12% and -0.08% between deforestation and loss of forest cover in the UER (Yiran et al., 2012;
1990-2000 and 2000-2010, respectively, but increased by 1.1% between Boateng, 2017; Larbi et al., 2019). The remaining areas in the UER with
2010 and 2019. Water bodies marginally increased by less than 1% substantial forest cover are uninhabited areas along the Red, Black and
from 1990 to 2019 in KNW District. Whereas bare soil consistently in- White Volta Rivers (Yiran et al., 2012; Boateng, 2017). The relative sta-
creased between 1990 and 2019 in Bongo District, bare soil decreased in bility of the shrub & tree savannah areas in Bongo and KNW Districts in
KNW District. The increases in bare land areas in Bongo District are at- the UER is attributed to the gradual degradation of wooded savannah
tributable to increased use of agrochemicals and bad farming practices, areas (Gessner et al., 2015; Dimobe et al., 2017; Larbi et al., 2019).
which exacerbate land degradation, while increases in water bodies are
attributable to past and present Government irrigation programs. Shrub 3.4. Projected LULCC patterns and trends in Bongo and KNW districts
and tree savannah declined for the periods 1990 to 2000 and 2000 to between 2019 and 2030
2010 by -1% (-416.7 Ha) and -4% (-1680.9 Ha), respectively, in Bongo
District. Along the same period in Bongo District, the wooded savan- In Bongo District, the CA–Markov model based on business-as-usual
nah area also declined considerably by -8% (-3466.5 Ha) and -3.5% scenarios (in which poverty levels will keep rising, population growth
(-1467.8 Ha), respectively. From 2010 to 2019, while the wooded sa- will continue unabated, and inhabitants will continue to collect fire-
vannah area decreased considerably by -14.9% (6208.5 Ha), shrub and wood and produce charcoal) forecast that between 2019 and 2030, set-
tree savannah increased by 7.6% (3148.9 Ha). In KNW District however, tlement area will increase by 7.30 km2 at the expense of other land
apart from 1990 to 2000, in which shrub and tree savannah decreased use and cover forms, which represents a change of 24.48% of the total
by -3.9%, shrub and tree savannah areas significantly increased by 15% area (Table 4). Overall, the net change in settlement area is 3.26 km2 .
and 10.1% from 2000 to 2010 and 2010 to 2019, respectively, owing In KNW District, settlement and bare lands will witness net gains of

7
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Table 4
Projected LULCC in Bongo and KNW Districts between 2019 and 2030.

LULCC Bongo District KNW District

class 2 2 2
loss km gain km net change km % change loss km2 gain km2 net change km2 % change

Settlement –4.04 7.30 3.26 24.48 –2.04 10.23 8.19 38.25

Agriculture –26.67 44.88 18.21 14.00 –54.32 7.01 –47.31 –45.35
Bare land –7.03 2.59 –4.43 –46.24 –4.20 8.66 4.47 28.64
Water Body –2.05 1.18 –4.43 –46.24 –0.95 0.76 –0.20 –1.73
Shrub and tree Savannah –2.05 1.18 –0.87 –7.67 –12.25 162.85 150.60 24.97
Wooded Savannah –29.32 13.37 –15.95 –33.90 –117.06 1.31 –115.75 –148.74

Fig. 6. Projected LULCC maps for Bongo District between 2019 and 2050.

8.19 km2 and 4.47 km2 , accounting for 38.25% and 28.64% of the total 22 km2 , respectively, accounting for 29.01% and 14.47% of the total
area, respectively. area, respectively. Bare land, water bodies, shrub and tree savannah
In Bongo District, a net change of 18.21 km2 is expected to occur and wooded savannah areas in Bongo District are expected to decline
in agriculture areas, signalling a 14% change in the total area. The re- considerably by -7.34%, -1.88%, -1.63%, and -73,26% of total area, re-
maining land use and cover classes such as bare land, water Body, shrub spectively (Table 5). The decline in shrub and tree Savannah between
and tree savannah and wooded savannah are projected to decline by net 2030 and 2040 in Bongo District will constitute the highest proportion
losses of -4.43 km2 , -0.87 km2, -0.21 km2 , and -15.95 km2 , respectively (21.82 km2) due to expansion in agriculture and settlement area. The
between 2019 and 2030 in Bongo District. These net losses in bare land, net change of -19.90 km2 in wooded savannah areas will be the great-
water Body, shrub and tree savannah and wooded savannah represent est proportion of net change in a particular land use and cover type
declines in total area by -46.24%, -46.24%, -7%, and -33.90%, respec- between 2030 and 2040. Although the highest proportion of loss is in
tively (Table 4). shrub and tree savannah area, the net change in shrub and tree savan-
In KNW District, agriculture and wooded savannah areas are pre- nah area (-3.15 km2 ) is considerably low compared to the net change
dicted to witness significant net decreases of -47.31 km2 and - in wooded savannah areas (-19.90 km2 ) because the shrub and tree sa-
115.75km2, accounting for -45.35% and -148.74% of total area change, vannah area will gain a significant amount (18.67 km2 ) of the total area
respectively. The highest proportion of net gain in land area be- at the expense of the wooded savannah area. Bare lands are projected
tween 2019 and 2030 is shrub and tree savannah, with a net gain of to decrease by a net change of -4.18 km2 due to the implementation of
150.60 km2 and this net gain accounts for 24.97% of total land area. sustainable land-use practices, reclamation, and regeneration efforts in
Bongo District.
3.5. Projected LULCC in Bongo and KNW Districts between 2030 and In Kassena Nankana West District however, between 2030 and 2040,
2040 agriculture, bare land, water bodies, and wooded savannah areas are
From 2030 to 2040 in Bongo District, settlement area and agri- expected to witness a net change of -6.17 km2 , -0.84 km2 , -0.24 km2 ,
culture lands are projected to increase by net gains of 5.44 km2 and and -3.84 km2 , respectively, which translate to -6.14%, -5.71%, -2.15%

8
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Table 5
Projected LULCC in Bongo and KNW Districts between 2030 and 2040.

LULCC Bongo District KNW District

class 2 2 2
loss km gain km net change km % change loss km2 gain km2 net change km2 % change

Settlement –0.62 6.06 5.44 29.01 –2.94 3.69 0.76 3.41

Agriculture –3.39 25.38 22.00 14.47 –12.68 6.51 –6.17 –6.14
Bareland –4.21 0.03 –4.18 –7.34 3.72 2.88 –0.84 –5.71
Water Body –0.21 0.00 –0.21 –1.88 –0.51 0.27 –0.24 –2.15
Shrub and tree Savannah –21.82 18.67 –3.15 –1.63 –7.03 17.36 10.34 1.69
Wooded Savannah –20.85 0.95 –19.90 –73.26 –8.12 4.28 –3.84 –4.76

Fig. 7. Projected LULCC maps for KNW Districts between 2019 and 2050.

and -4.76% of total area change, respectively (Table 5). The only areas 3.7. Projected LULCC Maps for Bongo and KNW Districts between 2019
expected to witness significant expansion in KNW District are shrub and and 2050
tree savannah and settlement, with the net change of 10.34 km2 and
0.76 km2 , respectively, accounting for 1.69% and 3.41% of the total Fig. 6 presents LULCC maps for Bongo District between 2019 and
land area change. 2050. It can be observed that the highest proportion of net loss in the
area is wooded savannah, with a net decrease of -11.26 km2 (-70.83% of
3.6. Projected LULCC in Bongo and KNW Districts between 2040 and the total area) between 2019 and 2050. The second highest proportion
2050 of net decrease in area is shrub and tree savannah, which accounts for
a net decrease of -10.39 km2 (-5.68%) of total area between 2019 and
Similar dynamics observed over the period 2019 to 2040 in Bongo 2050.
District are expected to continue through to the 2050 period. Settle- Bare land will witness a net decrease of -2.51 km2 between 2019
ment area and agricultural land is projected to increase at the expense and 2050. Settlement and Agriculture lands will witness net increases
of other land use and cover forms at 23.1% and 10.97% of total area, of 5.63 km2 (23.10%) and 18.73 km2 (10.97%) of the total area in Bongo
respectively. Bare land, water bodies, shrub and tree savannah and District between 2019 and 2050 (Fig. 6).
wooded savannah areas are equally projected to decline by -86.73%, In KNW District, Shrub and tree savannah area will witness a net gain
-1.89%, -5.68% and -70.83% of total area, respectively (Table 6). Con- of 156.02 km2 (25.64%) while settlement and bare land areas will in-
versely, in KNW District, between 2040 and 2050, agriculture, wooded crease by a net gain of 8.60 km2 and 3.77 km2 of the total area change
savannah and bare land are expected to expand with a net change of (Fig. 7). Wooded savannah will decrease at a net loss of -111.49 km2
2.19 km2 (2.20%), 3.12 km2 (3.73%) and 0.14 km2 (0.94%), respec- (-135.83%) of total area change. Likewise, the agriculture area will de-
tively (Table 6). Shrub and tree savannah will decrease by a net change crease by a net loss of -56.25 km2 (-58.99%) of total area change at
of -4.92km2 (-0.18%). the expense of settlement area between 2019 and 2050. This finding is

9
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Table 6
Projected LULCC in Bongo and KNW Districts between 2040 and 2050.

LULCC Bongo District KNW District

class 2 2 2
loss km gain km net change km % change loss km2 gain km2 net change km2 % change

Settlement –0.23 5.86 5.63 23.10 –2.43 2.09 –0.35 –1.58

Agriculture –3.00 21.73 18.73 10.97 –2.85 5.05 2.19 2.20
Bare land –2.52 0.00 –2.51 –86.73 –2.17 2.31 0.14 0.94
Water Body –0.21 0.00 –0.21 –1.89 –0.39 0.19 –0.20 –1.80
Shrub and tree Savannah –20.68 10.29 –10.39 –5.68 –9.23 4.32 –4.92 –0.81
Wooded Savannah –11.77 0.51 –11.26 –70.83 –3.64 6.77 3.12 3.73

contrary to previous studies which project significant increases in agri- Antwi-Agyei, P., Kpenekuu, F., Hogarh, J.N., Obiri-Danso, K., Abaidoo, R.C., Jeppesen, E.,
culture area by the 2050 period in Ghana (Aduah et al., 2018). The Andersen, M.N., 2019. Land use and land cover changes in the owabi reservoir catch-
ment, Ghana: implications for livelihoods and management. Geosciences 9 (7), 286.
findings however confirm the forecast that wooded savannah area will Arsanjani, J.J., Kainz, W., Mousivand, A.J., 2011. Tracking dynamic land-use change using
decline considerably by the 2050 (Larbi et al., 2019) as well as likely spatially explicit Markov Chain based on cellular automata: the case of Tehran. Int.
increases in settlement area (Ackom et al., 2020). J. Image Data Fusion 2 (4), 329–345.
Basommi, P.L., Guan, Q., Cheng, D., 2015. Exploring land use and land cover change in
the mining areas of Wa East District, Ghana using Satellite Imagery. Open Geosciences
4. Conclusions 1, 618–626.
Boafo, Y.A., Saito, O., Jasaw, G.S., Yiran, G.A., Lam, R.D., Mohan, G., Kranjac- Berisavlje-
Analysing and predicting past and future land use dynamics and land vic, G., 2020. Perceived community resilience to floods and droughts induced by cli-
mate change in semi-arid Ghana. In: Sustainability Challenges in Sub- Saharan Africa
cover changes have become a prerequisite for sustainable landscape gov- I. Springer, Singapore, pp. 191–219.
ernance, particularly in the drylands. Geospatial data in combination Boateng, P., 2017. Land access, agricultural land use changes and narratives about Land
with land surface modellers (e.g. CA-Markov chain) are effective tools degradation in the Savannahs of Northeast Ghana during the pre-colonial and colonial
periods. Soc. Sci. 6 (1), 35.
in providing valuable information on the dynamics of LULCC over space Bounoua, L., Nigro, J., Zhang, P., Thome, K., Lachir, A., 2018. Mapping urbanization in
and time. Understanding the dynamics of LULCC offers guidance in pol- the United States from 2001 to 2011. Appl. Geogr. 90, 123–133.
icy formulation in sustainable landscape governance. Agriculture and Brandt, M., Hiernaux, P., Rasmussen, K., Mbow, C., Kergoat, L., Tagesson, T., Fensholt, R.,
2016. Assessing woody vegetation trends in Sahelian drylands using MODIS based
settlement areas in both study districts have expanded at the expense of
seasonal metrics. Remote Sens. Environ. 183, 215–225.
wooded savannah and shrub and tree savannah areas between 1990 and Breiman, L., 2001. Random forests. Mach. Learn. 45, 5–32.
2019. Future predictions point to a continuous trend (i.e. from 2019 to Butt, A., Shabbir, R., Ahmad, S.S., Aziz, N., 2015. Land use change mapping and analysis
using remote sensing and GIS: a case study of Simly watershed, Islamabad, Pakistan.
2050), in which substantial expansion of agriculture and the settlement
Egypt. J. Remote Sens. Space Sci. 18 (2), 251–259.
area will occur at the expense of wooded savannah areas. The paper rec- Congalton, R.G., 1991. A review of assessing the accuracy of classifications of remotely
ommends that the government should implement necessary steps to re- sensed data. Remote Sens. Environ. 37 (1), 35–46.
claim and protect vulnerable landscapes from further degradation. This Dimobe, K., Goetze, D., Ouédraogo, A., Forkuor, G., Wala, K., Porembski, S., Thiom-
biano, A., 2017. Spatio-temporal dynamics in land use and habitat fragmentation
could be achieved through collaboration amongst key stakeholders such within a protected area dedicated to tourism in a Sudanian savanna of West Africa. J.
as forestry commission, Environmental Protection Agency (EPA), For- Landsc. Ecol. 10 (1), 75–95.
est Research Institute, traditional authorities and the Ministry of Lands Ellis, E.C., Kaplan, J.O., Fuller, D.Q., Vavrus, S., Goldewijk, K.K., Verburg, P.H., 2013.
Used planet: a global history. Proc. Natl Acad. Sci. 110 (20), 7978–7985.
and Natural Resources. The landscape governance actions should focus Gessner, U., Knauer, K., Kuenzer, C., Dech, S., 2015. Land surface phenology in a west
on protecting forest areas and promoting sustainable land management african savanna: impact of land use, land cover and fire. In: Remote Sensing Time
practices in the savannah ecological zone of Ghana. The classification Series. Springer, Cham, pp. 203–223.
Ghana Statistical Service (GSS), (2021). 2021 population and housing census. Regional
accuracy results were not perfect due to presence of cloud cover which Analytical report. Upper East Region. Accra, Ghana: Ghana Statistical Service.
affected the image pixel value. Also, it was cumbersome calibrating the Greiner, C., 2016. Land-use change, territorial restructuring, and economies of anticipa-
model due to the high intensity of changes in the LULCC that occurred tion in dryland Kenya. J. Eastern Afr. Stud. 10 (3), 530–547.
Guan, D., Li, H., Inohae, T., Su, W., Nagaie, T., Hokao, K., 2011. Modeling urban land
in the period under analysis owing to expansion in agriculture activities.
use change by the integration of cellular automaton and Markov model. Ecol. Modell.
This notwithstanding, the findings of the study are important and can 222, 3761–3772 20-22.
be applied to other areas in the savannah zone of Ghana with similar Halmy, M.W.A., Gessler, P.E., Hicke, J.A., Salem, B.B., 2015. Land use/land cover
change detection and prediction in the north-western coastal desert of Egypt using
socio-economic and ecological features.
Markov-CA. Appl. Geogr. 63, 101–112.
Hayes, M.M., Miller, S.N., Murphy, M.A., 2014. High-resolution landcover classification
Declaration of Competing Interest using random forest. Remote Sens. Lett. 5 (2), 112–121.
Karki, S., Thandar, A.M., Uddin, K., Tun, S., Aye, W.M., Aryal, K., Chettri, N., 2018. Im-
pact of land use land cover change on ecosystem services: a comparative analysis on
The authors declare that they have no known competing financial
observed data and people’s perception in Inle Lake, Myanmar. Environ. Syst. Res. 7
interests or personal relationships that could have appeared to influence (1), 1–15.
the work reported in this paper. Khwarahm, N.R., Qader, S., Ararat, K., Al-Quraishi, A.M.F., 2021. Predicting and mapping
land cover/land use changes in Erbil/Iraq using CA-Markov synergy model. Earth Sci.
Data availability Inf. 14 (1), 393–406.
Kleemann, J., Baysal, G., Bulley, H.N., Fürst, C., 2017. Assessing driving forces of land use
and land cover change by a mixed-method approach in north-eastern Ghana, West
Data will be made available on request. Africa. J. Environ. Manag. 196, 411–442.
Klutse, N.A.B., Owusu, K., Boafo, Y.A., 2020. Projected temperature increases over north-
References ern Ghana. SN Appl. Sci. 2 (8), 1–14.
Koranteng, A., Zawila-Niedzwiecki, T., Adu-Poku, I., 2016. Remote sensing study of land
use/cover change in West Africa. J. Environ. Prot. Sustain. Dev. 2, 17–31.
Abdullah, A.Y.M., Masrur, A., Adnan, M.S.G., Baky, M.A.A., Hassan, Q.K., Dewan, A.,
Lambin, E.F., Meyfroidt, P., 2010. Land use transitions: socio-ecological feedback versus
2019. Spatio-temporal patterns of land use/land cover change in the heterogeneous
socio-economic change. Land Use Policy 27 (2), 108–118.
coastal region of Bangladesh between 1990 and 2017. Remote Sens. 11 (7), 790.
Larbi, I., Forkuor, G., Hountondji, F.C., Agyare, W.A., Mama, D., 2019. Predictive land
Ackom, E.K., Adjei, K.A., Odai, S.N., 2020. Monitoring land-use and land-cover changes
use change under business-as-usual and afforestation scenarios in the vea catchment,
due to extensive urbanization in the Odaw River Basin of Accra, Ghana, 1991–2030.
West Africa. Int. J. Adv. Remote Sens. GIS 8, 3011–3029.
Model. Earth Syst. Environ. 6 (2), 1131–1143.
Liping, C., Yujun, S., Saeed, S., 2018. Monitoring and predicting land use and land cover
Aduah, M.S., Toucher, M.L., Jewitt, G.P., 2018. Estimating potential future (2030 and
changes using remote sensing and GIS techniques-a case study of a hilly area, Jiangle,
2040) land use in the Bonsa catchment, Ghana, West Africa. S. Afr. J. Geomat. 7 (3),
China. PLoS One 13 (7), e0200493.
279–291.

10
P. Aniah, S. Bawakyillenuo, S.N.A. Codjoe et al. Environmental Challenges 10 (2023) 100664

Lu, Y., Wu, P., Ma, X., Li, X., 2019. Detection and prediction of land use/land cover change Platform on Biodiversity and Ecosystem Services on pollinators, pollination, and food
using spatiotemporal data fusion and the cellular automata–Markov model. Environ. production. Bonn, Germany, Secretariat of the Intergovernmental Science-Policy Plat-
Monit. Assess. 191 (2), 68. form on Biodiversity and Ecosystem Services, 556pp.
Millennium Ecosystem Assessment (MA), 2003. Ecosystem and Human Well-Being: A Ruelland, D., Levavasseur, F., Tribotté, A., 2010. Patterns and dynamics of land-cover
Framework For Assessment. World Resources Institute, Washington, DC. changes since the 1960s over three experimental areas in Mali. Int. J. Appl. Earth
Millennium Ecosystem Assessment (MA), 2005. Ecosystems and Human Well-Being: Wet- Obs. Geoinf. 12, S11–S17.
lands and Water. World Resources Institute. Santos, D., Cardoso-Fernandes, J., Lima, A., Müller, A., Brönner, M., Teodoro, A.C., 2022.
Meadows, D., Randers, J., 2012. The Limits to Growth: The 30-Year Update. Routledge. Spectral analysis to improve inputs to random forest and other boosted ensemble
Meyfroidt, P., Lambin, E.F., Erb, K.H., Hertel, T.W., 2013. Globalization of land use: distant tree-based algorithms for detecting NYF pegmatites in Tysfjord, Norway. Remote Sens.
drivers of land change and geographic displacement of land use. Curr. Opin. Environ. 14 (15), 3532.
Sustain. 5 (5), 438–444. Song, X.P., Hansen, M.C., Stehman, S.V., Potapov, P.V., Tyukavina, A., Vermote, E.F.,
Mishra, V.N., Rai, P.K., 2016. A remote sensing aided multi-layer perceptron-Markov chain Townshend, J.R., 2018. Global land change from 1982 to 2016. Nature 560 (7720),
analysis for land use and land cover change prediction in Patna district (Bihar), India. 639–643.
Arabian J. Geosci. 9 (4), 249. Tavares, P.A., Beltrão, N.E.S., Guimarães, U.S., Teodoro, A.C., 2019. Integration of sen-
Mitsova, D., Shuster, W., Wang, X., 2011. A cellular automata model of land cover change tinel-1 and sentinel-2 for classification and LULC mapping in the urban area of Belém,
to integrate urban growth with open space conservation. Landsc. Urban Plan. 99 (2), eastern Brazilian Amazon. Sensors 19 (5), 1140.
141–153. Timm Hoffman, M., Skowno, A., Bell, W., Mashele, S., 2018. Long-term changes in land
... & Mohajane, M., Essahlaoui, A., Oudija, F., Hafyani, M.E., Hmaidi, A.E., Ouali, A.E., use, land cover and vegetation in the Karoo drylands of South Africa: implications for
Teodoro, A.C., 2018. Land use/land cover (LULC) using landsat data series (MSS, TM, degradation monitoring. Afr. J. Range Forage Sci. 35 (3-4), 209–221.
ETM+ and OLI) in Azrou Forest, in the Central Middle Atlas of Morocco. Environments Tolessa, T., Senbeta, F., Kidane, M., 2017. The impact of land use/land cover change on
5 (12), 131. ecosystem services in the central highlands of Ethiopia. Ecosyst. Serv. 23, 47–54.
Omar, N.Q., Ahamad, M.S.S., Hussin, W.M.A.W., Samat, N., Ahmad, S.Z., 2014. Markov Tuffour-Mills, D., Antwi-Agyei, P., Addo-Fordjour, P., 2020. Trends and drivers of land
CA, multi regression, and multiple decision making for modeling historical changes cover changes in a tropical urban forest in Ghana. Trees Forests People 2, 100040.
in Kirkuk city. Iraq. J. Indian Soc. Remote Sens. 42, 165–178. Wang, S.W., Munkhnasan, L., Lee, W.K., 2021. Land use and land cover change detection
Osumanu, I.K., Akomgbangre, J.N., 2020. A growing city: patterns and ramifications of and prediction in Bhutan’s high altitude city of Thimphu, using cellular automata and
urban change in Wa, Ghana. Spat. Inf. Res. 28 (5), 523–536. Markov chain. Environ. Chall. 2, 100017.
Osumanu, I.K., Akongbangre, J.N., Gordon, N., Tuu, Y., Owusu-Sekyere, E., 2019. From Wardell, D.A., Reenberg, A., Tøttrup, C., 2003. Historical footprints in contemporary land
patches of villages to a municipality: time, space, and expansion of Wa, Ghana. In: use systems: forest cover changes in savannah woodlands in the Sudano-Sahelian
Urban Forum, 30. Springer, Netherlands, pp. 57–74. zone. Glob. Environ. Change 13 (4), 235–254.
Peng, S., Ciais, P., Maignan, F., Li, W., Chang, J., Wang, T., Yue, C., 2017. Sensitivity of Yiran, G.A.B., Kusimi, J.M., Kufogbe, S.K., 2012. A synthesis of remote sensing and local
land use change emission estimates to historical land use and land cover mapping. knowledge approaches in land degradation assessment in the Bawku East District,
Global Biogeochem. Cycles 31 (4), 626–643. Ghana. Int. J. Appl. Earth Obs. Geoinf. 14 (1), 204–213.
Potts, S.G., Ngo, H.T., Biesmeijer, J.C., Breeze, T.D., Dicks, L.V., Garibaldi, L.A., Van-
bergen, A. (2016). The assessment report of the Intergovernmental Science-Policy

11