Journal Pre-proof

Modelling land use and land cover dynamics of Dedza district of Malawi using hybrid
Cellular Automata and Markov model

M.G. Munthali, S. Mustak, A. Adeola, J. Botai, S.K. Singh, N. Davis

PII: S2352-9385(19)30278-2
DOI: https://doi.org/10.1016/j.rsase.2019.100276
Reference: RSASE 100276

To appear in: Remote Sensing Applications: Society and Environment

Received Date: 20 August 2019

Revised Date: 24 September 2019
Accepted Date: 31 October 2019

Please cite this article as: Munthali, M.G., Mustak, S., Adeola, A., Botai, J., Singh, S.K., Davis, N.,
Modelling land use and land cover dynamics of Dedza district of Malawi using hybrid Cellular Automata
and Markov model, Remote Sensing Applications: Society and Environment (2019), doi: https://
doi.org/10.1016/j.rsase.2019.100276.

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier B.V.

 Modeling land use and land cover dynamics of Dedza district of Malawi
using hybrid Cellular Automata and Markov model
M. G. Munthali1*, S. Mustak2, A. Adeola3,4, J. Botai1,3,5, S.K. Singh6, N. Davis1,
1
Department of Geography, Geoinformatics and Meteorology, University of Pretoria, Private Bag X20,
Pretoria 0028, South Africa
2
Ecoinformatics lab, Ashoka Trust for Research in Ecology and Environment (ATREE), Srirampura,
Jakkur, 560064, Bangalore, Karnataka
3
South African Weather Service (SAWS), Private Bag X097, Pretoria 001, South Africa;
4
School for Health Systems and Public Health, University of Pretoria, Pretoria 0002, South Africa
5
School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, Westville
Campus, Private Bag X54001, Durban 4000, South Africa.
6
K. Banerjee Centre of Atmospheric and Ocean Studies, IIDS, Nehru Science Centre, University of
Allahabad, Allahabad 211002, India
* Correspondence: [email protected]

Abstract
The spatiotemporal variation of any landscape patterns is as a result of complex
interactions of social, economic, demographic, technological, political, biophysical and cultural
factors. Modelling land use and land cover (LULC) changes is essential for natural resource
scientists, decision-makers and planners in developing comprehensive medium and long-term
plans for tackling environmental or other related sustainable development issues. The current
study used an integrated approach that combines remote sensing and GIS to simulate and predict
plausible LULC changes for Dedza district in Malawi for the years 2025 and 2035 based on
Cellular Automata (CA)-Markov Chain model embedded in IDRISI Software. The model was
validated using a simulated and actual LULC of 2015. The overall agreement between the two
maps was 0.98 (98%) with a simulation error of 0.03 (3.0%). The more detailed analysis of
validation results based on the kappa variations showed a satisfactory level of accuracy with a
Kno, Kstandard and Klocation of 0.97, 0.95 and 0.97, respectively. The future projections indicate that
water bodies, barren land and built-up areas will increase while agricultural land, wetlands and
forest land will substantially decrease by 2025 and 2035 respectively. According to the transition
probability matrix, almost 94.8%, 97.6% and 95.7% of water bodies, agricultural land and barren
land will more likely remain stable by 2025. In contrast, forest land exhibits the highest
probability of change of 64.8% and 85.9% by 2025 and 2035 respectively. Results also indicate
that the majority of the forest areas will be converted to barren land with a probability of 60.8%
and 79.6% by 2025 and 2035, respectively. These findings serve as an important benchmark for
planners, natural resource managers and policy-makers in the studied landscape to consider in
pursuit of holistic sustainable development policies/strategies/ guidelines sustainable natural
resource management.

Keywords: LULC; CA-Markov; Malawi; Multi-criteria evaluation; Modelling

1. Introduction

1
 The spatiotemporal variation of any landscape patterns is as a result of complex
interactions of various factors. Land use and land cover (LULC) change has become one of the
fundamental concerns in natural resource management, sustainable development and
environmental change in the local, national, regional and global landscape (Foley et al. 2005;
Yirsaw et al. 2017). As such documentation of these LULC dynamics provides vital information
for better understanding of historical land use and management practices, current land use
patterns and future LULC change trajectories. Several LULC change studies on environmental
changes have been widely investigated using multi-temporal remote sensed imagery approaches
(Basommi et al. 2016). These studies consistently demonstrate how human activities in
conjunction with natural causes are key drivers of LULC dynamics at all spatial and temporal
scales (Lamichhane 2008; Mishra et al. 2014; Singh et al. 2015; Basommi et al. 2016; Singh et
al. 2018; Varga et al. 2019). Across the globe, authors have identified agricultural expansion,
population growth, poverty, urban growth, charcoal production, firewood collection, just to
mention a few, as the drivers responsible for LULC dynamics at local, national, regional and
global scale (Serneels and Lambin 2001, Chomitz et al. 2007; DeFries et al.2010; Kindu et al.
2015; Pullanikkatil et al. 2016; Mannan et al. 2018; Solomon et al. 2018). Berihun et al. (2019)
identified changes in farming practices as one of the major drivers of LULC changes that took
place between 1982 and 2017 in three watersheds of drought-prone areas from different agro-
ecological zones of Upper Nile basin in Ethipia. . In another study by Yesuph and Dagnew
(2019), fuel wood and timber extraction, drought, expansion of farmlands and settlements, land
tenure insecurity, population pressure, terrain features of the area and population growth were
the major drivers behind LULC changes taking place in Beshillo catchent of the Blue Nile Basin
of North Eastern Highlands of Ethiopia. Like other countries, Malawi has experienced LULC
changes over the past decades. Some studies reported in the literature focused on documenting
the nature and extent of these historical changes as well the drivers behind these changes
(Palamuleni et al. 2010; Chavula et al. 2011; Munthali and Murayama 2011; Haack et al. 2014;
Munthali et al. 2014; Jagger and Perez-Heydrich 2016; Pullanikkatil et al. 2016). However, only
Munthali and Murayama (2014) modelled Dzalanyama Forest Reserve using the Multi-agent
simulation approach. Despite the focus on documenting historic changes in Malawi, very few
studies, however, have been conducted to simulate the future LULC in the country which formed
the purpose of this research.

2
 Pressure on different land uses are increasing all over the world and understanding the
implications of land use change patterns is deemed crucial in the context of future natural
resource management and planning. This fact highlights the need for innovative global scientific
research in the field of LULC modelling (Kamusoko et al. 2011; Qiang and Lam, 2015).
According to Paegelow and Olmedo et al. (2010), LULC modelling is defined as the
interpolation or extrapolation when the simulation exceeds the known specified period of time.
Pressure on different LULC types is increasing all over the world and understanding the future
land use patterns is very crucial and hence calls for global scientific research (Kamusoko et al.
2011; Qiang and Lam, 2015). Future LULC patterns and changes need to be understood as far as
natural resource management of available resources is concerned. Thus, spatiotemporal LULC
change modelling is imperative in order to predict the future LULC distribution for effective land
use management and planning (Regmi et al. 2014; Bhattacharjee and Ghosh 2015; Dezhkam et
al. 2017).Accordingly, modelling LULC dynamics also provides vital information to planners
and decision makers about current natural resource management policies and strategies and how
these actions may affect the future LULC patterns (Sun et al. 2012; Omar et al. 2014; Regmi et
al. 2014; Bhattacharjee and Ghosh 2015;Martinuzzi et al. 2015; Dezhkam et al. 2017). Accurate
LULC trajectories are thus pivotal for natural resource scientists, decision-makers and planners
in developing comprehensive medium and long-term plans earmarked to prevent the
consequences arising from the undesirable LULC changes in a landscape (Theobald and Hobbs
2002; Loomis 2002; Maestas et al. 2003). Better still, knowing the possible outcomes of the
predicted LULC changes can be helpful when making and implementing difficult policy
decisions (Sun et al. 2012). Additionally, the ecosystem service values (ESV) can also be
estimated in response to LULC dynamics on a landscape (Hu et al. 2008; Dallimer et al. 2015;
Kindu et al. 2016). Kelly et al. (2018) used CA-Markov to combat desertification by identifying
and predicting the areas that were susceptible/predisposed to desertification in Montes Claros,
Brazil. Recently, modelling LULC change has been considered as one of the most valuable
tools in ensuring that the present natural resource base guarantees a future and continuous supply
of natural resources. Over the last few decades, researchers have developed and used several
LULC change models/approaches for modelling LULC dynamics. Some of the most widely used
LULC approaches include evolutionary models (neural networks), mathematical models (linear
and static), multi-agent-based models, cellular models (Cellular Automata, CA), expert system

3
models, Markov chains and hybrid models (Stefanov et al. 2001; Verburg et al. 2002; Parker et
al. 2003; Xie et al. 2007; Araya and Cabral 2010; Guan et al.2011; Ralha et al. 2013; Subedi et
al. 2013; Han et al. 2015). According to Verburg et al. (2004), LULC models are very powerful
tools for examining the spatial pattern as well as the rate and drivers/causes of LULC dynamics.
Additionally, Verburg et al. (2004) highlights the value of LULC models in terms of evaluating
land use policies and predicting future land use demand. No single model is capable of
considering all the processes of LULC changes at various scales (Verburg et al. 2008).
Alternatively, integrated land-use models that combine one or two models are preferred.
Presently, the most widely used models in LULC change monitoring and prediction are cellular,
Markov chain and agent-based models or the mixed model based on these three types of models
(Stevens and Dragićević 2007; Zhao et al. 2012; Myint and Wang 2013; Sol and Claggett 2013).
One such mixed model is the Cellular Automata–Markov Chain (CA-Markov) model. The CA-
Markov model is one of the most reliable, robust and effective LULC change approaches for
predicting the long-term or decadal spatial and temporal variations of LULC in a complex
system along with geographic information systems(Thomas and Laurence 2006; Wu et al. 2006;
Kamusoko et al. 2009; Yu 2009; Sang et al. 2011; Steeb 2011; Arsanjani 2013; He et al. 2014;
Singh et al. 2015). According to Wang and Zhang (2001), the model combines both biophysical
and socio-economic data to simulate accurate LULC change in a plausible future. The CA-
Markov is used in management, planning, modelling and simulation of the spatial processes (Wu
and Webster 2000; O’Sullivan 2001; Wu 2002; Irwin et al. 2009; Araya and Cabral 2010; Singh
et al. 2015; ).
Based on the background above, the present study uses an integrated approach that
combines remote sensing and GIS to simulate and predict LULC changes for Dedza district for
the years 2025 and 2035 based on CA-Markov using IDRISI Software. Thus, the main objective
of this study is to simulate and predict the future spatio-temporal patterns of LULC dynamics of
Dedza district of using hybrid CA-Markov model. It is anticipated that the findings of this study
could guide the natural resource scientists, planners and decision makers to understand the future
effects of LULC dynamics in the study area and that will enable them to develop proper
interventions, land use planning and management policies for socio-economic development of
the study area and districts with similar settings. The CA-Markov approach has been adopted in
this study because many researchers have applied this approach in different landscapes to model,

4
monitor, predict and simulate LULC changes in their study areas and the method obtained very
accurate and reliable results (Subedi et al. 2013; Rendana et al. 2015; Singh et al. 2018; Liping et
al. 2018). Equally important, Liu et al. (2007), Qiuand Chen (2008) and Yang et al. (2012) stated
that the reliability of LULC change modelling approach can be tremendously improved by
coupling two or more modelling techniques to integrate the advantages of each simulation
model..
2. Materials and Methods

2.1 Study area

Dedza district is located in the Central Region of Malawi, approximately 81km from the
Capital city of Malawi, Lilongwe (Figure 1). It covers an area of 3,624 km2. It is characterized
by three topographic zones namely Lilongwe plain (altitude 1100-1300m), the Dedza highlands
(1200-2200m) and the Dedza escarpments (1000-1500m). Soils are ferruginous: generally deep
and brown to reddish in colour (GoM 1999). Clay and sandy loam soils are predominant in the
study area (GoM 2012; 2013). The district receives mean annual rainfall varying from 800mm to
1200 mm. The average annual temperature is 15.5°C. The district is endowed with rivers which
include Linthipe River originating from Dzalanyama Ranges (Dedza) and runs through Dedza,
Lilongwe and Salima districts. Linthipe River joins the Diampwe II River which drains the area
to the west and then turns north-eastwards towards Lake Malawi. Both Linthipe and Diampwe
are perennial.

Based on the latest Malawi population and housing census report, the district has a
population of 830,512 (Government of Malawi 2019). The annual population growth rate is
approximately 2.8 % while the population density is 221 people /km2 which is above the
average national density of 186 people/ km2. Approximately, 52.3% of the population are
female and the district has a sex ratio of 91.1. Proportionately, 2018 census reveals that about
96% of the total population lives in rural areas. The majority of these rural people are
categorized as extremely poor. According to GoM (2019), literacy levels of the people in the
district are low (57% of the population is literate) as compared to Likoma district (85% of the
population is literate).

According to GoM (2010; 2012), the main economy and source of livelihoods of the
majority of the communities of the study area are primarily based on natural resources especially

5
land, forests and water. Rainfed and irrigation agriculture, casual labour and small-scale
businesses are mainly praactised as means of diversifying income (Gom 2012). Thus, the main
economic activities of the local communities are farming especially in the plains, forestry in the
highlands and fishing along the lake. In terms of land use, agriculture remains the major land use
in Dedza district. Other notable LULC types include settlements, forests, wetlands, water bodies
and barren land. The district has two government-owned plantations namely Chongoni and
Dedza Mountain Plantations. Other forests include Mua-livulezi, Chongoni, Dzalanyama, Dedza
Mountain, Mua-tsanya and Dedza-Salima Escarpment Forest reserves

Insert Figure 1 here

2.2 Data used

The data used in this study were acquired from different sources. Landsat satellite imageries
from 1991, 2001 and 2015 were used to derived LUL classified maps for the studied landscape.
The satellite data was selected based on the availability and absence of cloud cover. Shuttle
Radar Topography Mission (STRM) DEM was also used. The detailed satellite data used in this
study are depicted in Table 1.

Insert Table 1 here

2.3 Image preprocessing, classification and accuracy assessment

The satellite data were all preprocessed wherein the images were atmospheric,
radiometric and geometric corrected. A hybrid supervised/unsupervised classification approach
was employed. Image classification was done using hybrid classification approach. Firstly,
unsupervised image classification was employed using Iterative Self-Organizing Data Analysis
clustering (ISODATA) to determine the number of spectral classes. Secondly supervised
classification was performed on unsupervised classified image for the final land use/cover
classification for the images of 1991, 2001 and 2015. Six (6) classes were identified based on
physiographical knowledge of the study area, supportive ancillary data, the researcher’s prior
local knowledge and visual interpretation using the historical function of Google earth (Table 2).

Insert Table 2 here

A stratified random sampling method was employed to collect 221 ground control points
and Google Earth images were used to extract reference data. The accuracy assessment was only

6
performed on 2015 images . Accuracy assessment was not performed on 1991 and 2001 images
due to the unavailability of ground validation data in the form of aerial photographs and archived
Google earth images. . In this case, the used signatures for the 2015 images were superimposed on older
images. Accuracy assessment was determined using the Kappa coefficient, overall accuracy,
producer’s and user’s accuracy which were derived from the error (confusion) matrix as
discussed by Liu et al. (2007) and Congalton and Green (2009).The accuracy assessment based
on the error (confusion matrices) showed an overall accuracy of 91.86% with a kappa coefficient
of 0.866 (Munthali et al. 2019a). A detailed description of all the classification and accuracy
assessment methods used to derive the LULC (Figure 2) is described in previous publications
(See Munthali et al. 2019 a & b).

Insert Figure 2 here

2.4 Modelling of future LULC dynamics

2.4.1 Markov chain model

The Markov chain model (MCM) is a stochastic model that describes the transition probability of
LULC type shifting from one mutually exclusive state (
 ) to another state (
 ) over a
specified period of time (Thomas and Lawrence 2006; Mishra et al. 2011; Liping et al. 2018).
The predicted future LULC changes usually depends on the subsequently transition probabilities
generated from the past or current LULC changes (Guan et al., 2008; Wu et al. 2010). However,
MCM does not give the right spatial distribution (allocation) of occurrences of LULC changes
rather estimate and predict the magnitude/quantity of these changes (Behera et al. 2012; Yang et
al. 2012). Mathematically, MCM for predicting LULC changes can be presented using the
conditional probability equation described by e.g., Yousheng et al. (2011); Ma et al. (2012);
Subedi et al. (2013); Singh et al. (2015) and Mondal et al. (2016);

() =  ×
() (1)

and

 ⋯ 
 = ⋮ ⋮ ⋮  (0 ≤  ≤ 1 and ∑  = 1, ,  = 1, 2, ⋯ , )
 ⋯ 
(2)

7
where  is the transition probability matrix,  and  are the LULC type at time of and +1
respectively,  is the no. of LULC types;
 and
 is LULC status at time of and + 1
respectively

2.4.2 Cellular automata model

According to Yang and Li (2007) and Guan et al. (2011), Cellular automata (CA) model is
a discrete model with a spatially extended dynamic system based on defined transition rule that
relates the new state to the previous state of LULC type and those of its neighbours.
Additionally, CA-based models have the ability to represent nonlinear and complex spatially
distributed processes thereby capable of providing insights into local, national, regional and
global LULC change patterns (He et al. 2006; ; Liping et al. 2018). However, the model has
notable components which need to be considered to get optimum simulation results and these
parameters are cells, transition rules, cell size, time and cell neighbourhoods (Wang et al. 2012;
Liping et al. 2018). Thus, the spatial and temporal state of the neighbours heavily depends on the
state of each cell (Kumar et al. 2014). According to Sang et al. (2011), Subedi et al. (2013),
Singh et al. (2015), Mondal et al. (2016) and Liping et al. (2018), the CA model can be
expressed as:

( , + 1) = " ((
 ), #) (3)

Where
is the set of states of the finite cells, # is the number of neighbourhood cells, and
+ 1 are different times and " is the transformation rule of local space

2.4.3 CA-Markov model

The CA-Markov approach is a model that effectively combines the advantages of Markov
Chain to predict long-term LULC changes and Cellular Automata models to accurately simulate
and predict future spatiotemporal LULC changes (Behera et al. 2012; Wang et al. 2012; He et al.
2014; Yang et al. 2014; Singh et al. 2015; Parsa et al. 2016). It is worth mentioning that the use
of integrated CA-Markov model in LULC studies is advantageous due to its dynamic explicit
simulation capability, simple calibration, high efficiency with data and ability to simulate
multiple LULC types and complex patterns (Mermarian et al. 2012; Yang et al. 2012; Hyandye
and Martz 2017; Singh et al. 2015; Singh et al. 2018).

8
 This study adopted an existing modelling technique, CA-Markov model with
Multicriteria Evaluation (MCE) module embedded in the IDRISI software version 17. Figure 3
depicts the methodology deployed for predicting LULC dynamics in the study area using the
CA-Markov model. According to Eastman (2000), the model changes various LULC classes of
cells by Markov transition matrix, a suitability map and a neighborhood filter.

Insert Figure 3 here

2.4.3.1 Preparation of suitability maps

Preparing transition or suitability maps for various LULC classes is a prerequisite for
modelling LULC is reported to be a difficult step as it depends on availability of data and
information (Keshtkar and Voigt 2016; Hyandye and Martz 2017). The suitability maps were
developed by applying the MCE module which includes two parts: the constraints (criteria that
limits the expansion of classes) and factors (give the degree of suitability for an area) to
determine the land to be considered for further development. The constraints were standardized
into the form of Boolean maps where 0 represented unsuitable land and 1 was a set value for
suitable land while the factors were standardized to a continuous scale of suitability from 0 (least
suitable) to 255 (most suitable). The constraints included existing water bodies, existing built-up
areas and protected areas especially forest reserves. Driving factors included the distance from
roads, distance from water bodies and distance from built-up areas (Table 3). These constraints
and factors were chosen based on their use in previous studies (Araya and Cabral 2010; Keshtkar
and Voigt 2016; Hyandye and Martz 2017; Rimal et al. 2017; Singh et al. 2018). The Fuzzy
function combined with Weighted Linear Combination (WLC) was used for standardization of
factors. During standardization, various control points were used and fuzzy functions applied in
this study included Sigmoidal and linear functions with monotonically increasing/decreasing or
symmetric. The weights of the factors were derived using the Analytic Hierarchy Process (AHP).
The final standardized suitability maps produced are depicted in Figure 4.

Insert Table 3 here

Insert Figure 4 here

2.4.3.2 LULC change prediction

9
 The LULC dynamics for the studied landscape were simulated by the Markovian model
in IDRISI software. The Markov module generated the transition probability matrices from 1991
to 2001 and from 2001 and 2015. The outputs from the Markov module were combined with the
suitability maps using the CA-Markov module to predict LULC changes for 2015 for model
validation. Classified LULC maps for 1991 and 2001 were used to produce simulated 2015
maps in order to validate the actual 2015 classified maps. The contingency filter of 5x5 pixels
was applied on suitability maps to define the neighbourhood of each cell of the LULC class as
depicted in Equation 1. Finally, the LULC map of 2015 was used as a base map to determine
LULC change predictions for 2025 and 2035 using the CA-Markov module integrated into
IDRISI software. Figure 4 shows the methodology deployed in this study.

0 0 1 0 0
00 1 1 1 03
/ 2
$% &'() 5 + 5 ", '- = /1 1 1 1 12 (4)
/0 1 1 1 02
.0 0 1 0 01

2.4.3.3 Validation of LULC prediction model

Model calibration and validation is an important step in predicting future decadal changes
where no datasets are available for accuracy of predicted data (Singh et al. 2015; Srivastava et al.
2013). In this study, the actual 2015 LULC map was used as a reference map to compare with
the results of the 2015 simulated LULC map based on Kappa variations. Kappa variations have
been strongly recommended and widely used to validate LULC change predictions (Pontius
2000; Subedi et al. 2013; Singh et al. 2015; Parsa et al. 2016). The Kappa variations used to
validate the CA-Markov model for LULC change predictions in this study were generated in
VALIDATE module included: traditional Kappa(Kstandard) or Kappa for no information/ability
(Kno), Kappa for location (Klocation) and Kappa for quantity (Kquantity) as expressed by Equations 5
to 8 according to procedure described by Omar et al. (2014). According to (Pontius 2000), Kno
indicates the proportion classified correctly relative to the expected proportion expected
classified correctly by simulation with no ability to specify quantity or location accurately. On
the other hand, Klocation and Kquantity are measures of validation between the actual maps and
simulated maps based on specified location and quantity respectively (Pontius and Schneider
2001; Pontius and Malanson2005; Sayemuzzaman and Jha 2014). The level of agreement of the

10
three agreement is considered to be perfect if the values equal to 1 and unsatisfactory or
imperfect is equals to 0 (Pontius and Schneider 2001; Nadoushan et al. 2015; Singh et al. 2015;
Singh et al. 2018). Therefore, a value of 0.80 and above is considered strong and it is reasonable
to make plausible future projections.

Kappa for no information = <= = (>(?)#( ))/(() − #( )) (5)

Kappa for location = <D=EF = = (>(?)#(?))/((?) − #(?)) (6)

Kappa for quantity = <GHF I = (> (?)J (?))/(< (?) − J (?)) (7)

Kappa standard = <LFMFNM = (>(?)#( ))/(() − #(?)) (8)

Where no information is defined by #(), medium straight information level by J (?), medium
grid cell level information by (> (?), perfect grid cell-level information given imperfect
stratum-level information by (< (?) mean and perfect grid cell-level information across the
landscape by (().

3. Results

3.1 LULC classification and accuracy assessment

Based on the first phase of the research analysis as captured in Munthali et al. (2019a),
the overall accuracy of the 2015 classified map was found to be 91. 86%. Thus, indicating the
suitability of the derived classified maps for effective and reliable LULC change analysis and
modelling. Post-classification analysis of the spatial metrics and their variations indicated that
agricultural land, forest land, wetlands and water bodies drastically decreased between 1991 and
2015 in the studied landscape (Table 4). On the other hand, barren land and built-up areas
substantially increased during the same period.

Insert Table 4 here

3.2 Validating LULC prediction model

The 2015 simulated LULC map was compared with the actual 2015 LULC map in order to
validate the LULC prediction model given by the CA-Markov model. Results of the comparison
analysis of the two maps are shown in Figure 5 and Table 5. Visual comparison of the two maps

11
clearly indicates that wetlands, forest land, agriculture land and barren areas had a strong
agreement (Table 5). However, the simulated map shows that wetlands and barren lands were
underestimated by 6.54% and 0.7%, respectively while forest land and agricultural land were
overestimated by 20.46% and 2.25%, respectively. On the other hand, results from water bodies
and built-up areas showed a weak agreement and they were both underestimated by 60.34% and
71.74%, respectively. Thus, intrinsic discrepancies were observed with the simulated 2015
LULC map especially for water bodies and built-up areas.

The ability of the model to simulate accurately the LULC maps for 2025 ad 2035 was
validated using the observed and simulated LULC maps of 2015 (Figure 5). From Table 6, the
results of the model evaluation show that the overall agreement between the observed LULC
map of 2015 and simulated LULC map of 2015 is 0.98 (97.5%) while the overall simulation
error is 0.03 (2.5%) which may be attributed to errors due to quantity disagreement (0.01) and
allocation disagreements (0.02. The more detailed validation results based on Kappa variations
are depicted in Table 7. The values of Kno, Klocation, KLocationStrata and Kstandard were 97.06%,
96.62%, 99.62% and 95.31%, respectively (Table 7) showing a satisfactory level of accuracy.

Insert Figure 5 here

Insert Table 5 here
Insert Table 6 here
Insert Table 7 here

3.3 Simulated land use and land cover changes

Results on the projected LULC changes for 2025 and 2035 are depicted in Figure 6 and
Table 8. About 2.13% of the total landscape area in the district was built-up areas in 2015 and is
predicted to expand up to 3.16% and 4.06% in 2025 and 2035 respectively (Table 8).
Agricultural land will continue to be the dominant LULC class by 2035. Modelling results
further indicate that forest land, agricultural land and wetlands will greatly decrease from 6,
237.63 ha (1.66%) to 5, 631.03 ha (1.50%), 260, 879.31 ha (69.41%) to 253, 394.01 ha (67.42%)
and 2, 680.29 ha (0.71%) to 2,471.49 ha (0.66%) by 2035 respectively. In contrast, there shall be
an expansion of water bodies, barren land and built-up areas by 0.28%, 26.09% and 4.06%
respectively.

12
 Tables 9 and 10 are the LULC change transition area and probability matrices showing
how each projected LULC class is projected to change between 2015 and 2035. According to the
transition probability matrix, almost 94.8%, 97.6% and 95.7% of water bodies, agricultural land
and barren land will more likely remain stable by 2025 (Table 9). On contrast, forest land
exhibits the highest probability of change of 64.8% and 85.9% by 2025 and 2035, respectively.
The majority of the forest areas will be converted to barren land with a probability of 60.8%and
79.6% by 2025 and 2035, respectively. Further, portion of forest land will be pressured by
agricultural land (3.2% in 2025 and 5% in 2035).

Insert Figure 6 here

Insert Table 8 here
Insert Table 9 here
Insert Table 10 here
4. Discussion
4.1 Model validation analysis

Understanding future spatial and temporal LULC patterns and changes of any landscape
is the pathway to scientific-evidence based sustainable management of natural resources. In this
study, the decadal LULC dynamics were simulated by employing an integration of GIS, remote
sensing and land change models. The modelling results showed that the observed and simulated
LULC maps for 2015 were reasonably similar despite some intrinsic discrepancies observed in
water bodies and built-up areas of the simulated map. This could be attributed to inadequate
suitability maps and the shape of contiguity filter, limited drivers and factors used for modelling
the results. Singh et al. (2015) and Hyandye and Martz (2017) also found similar results to this
study and concluded that the discrepancies between the real map and simulated LULC maps
were due to lack of suitable maps and choice of contiguity filter used in their study. Other
researchers emphasized that the quality of any simulated LULC map is based on not only the
visual interpretation of the two categorical maps but also the quality of the transition suitability
maps prepared, suitable transition matrix and the validation method employed (Verburg et al.
2006; Koomen and Stillwell 2007; Pena et al. 2007; Araya and Cabral 2010; Memarian et al.
2012).

13
 In order to justify the results above, a more detailed validation analysis was performed
using the kappa variations. Thus, the CA-Markov model was validated using the observed and
simulated LULC maps of 2015, where the Kno, Kstandard, Klocation and Kquantity were derived.
According to Landis and Koch (1977), a Kappa value of > 0.80 (80%) represents strong
agreement, and a value between 0.40 and 0.80 (40% - 80%) represents moderate agreement. The
validation results based on the kappa values showed that the overall agreement between the
observed and simulated LULC maps of 2015 was perfect. The main disagreement between the
two categorical maps in this study was due to allocation error rather than quantity errors.
Hyandye and Martz (2017) reported similar findings in Tanzania, however, their allocation error
(0.06) and quantity error (0.02) were higher than errors reported in this study. In addition, the
kappa variations Kno(0.97), Kstandard (0.95) and Klocation (0.97) found in this study showed a
satisfactory level of accuracy. Therefore, based on the kappa values obtained in this study, the
CA-Markov is suitable for accurate prediction of future spatio temporal LULC dynamics in the
studied landscape (Vierra and Garret 2005). Thus, LULC change prediction models with
accuracies ≥ 80% are typically considered as very strong predictive tools (Araya and Cabral
2010). The value of Kstandard in this study is a bit higher than those which have been reported in
other recent studies which employed the CA-Markov model in LULC change simulations, for
instance, 0.88 (Keshtkar and Voigt 2016), 0.68 (Hyandye and Martz 2017), 0.88 (Rimal et al.
2017), 0.59 and (Singh et al. 2018). Similarly, the value of Kno which gives the overall accuracy
of simulation of the simulated LULC maps was also higher than those reported by the same
studies. This validates the fitting of the current model in this study as the best fit.

4.2 Simulation of LULC dynamics for 2025 and 2035

The predicted results of spatiotemporal LULC dynamics reveal that built-up areas in the
studied landscape will continue to expand up to 3.16% and 4.06% by 2025 and 2035
respectively. The major decrease will be observed in forest land, agricultural land and wetlands
while barren land and water bodies will continue to increase by 0.28% and 26.09% respectively.
The transition probability matrix results predict that a large portion of forest land will be
transformed into barren land and agricultural land. The study has demonstrated that if the current
spatiotemporal LULC patterns and trends continue, only 14% of the total forest land will remain
forest by 2035. This will affect the continuous supply of timber for construction to Lilongwe

14
City and surrounding districts. The findings suggest that the continued increase in barren land is
an indication of the continued forest degradation and deforestation and this poses a great threat to
sustainable forest management and biodiversity conservation in the district. The decline in forest
cover also confirms that management decisions for protecting and conserving forest resources in
the studied landscape were not properly taken or implemented by natural resource managers and
planners. The observed results in the declined forest cover, wetlands and agricultural land are
expected based on the historical patterns and trends of LULC changes that has taken place
between 1991 and 2015 in the studied landscape (Munthali et al. 2019a). According to Munthali
et al. (2019b), the LULC changes were largely as a result of poverty, population growth,
firewood collection and charcoal production. It is worth noting these drivers will continue to
accelerate the undesired LULC changes to happen between 2015 and 2035. Based on the 2018
national population census, the population of Dedza has increased from 625,828 in 2008 and
830,512 in 2018 (Government of Malawi 2019). Therefore, these results imply that as the
population increases, more land is expected to be subjugated to cater for the growing population
for settlements and food production. Pandey and Khare (2017) and Hyandye and Martz (2017)
similarly reported that the continuous higher population flux in Usangu sub-Catchment of the
upper part of Rufiji Basin in Tanzania and the upper Narmada basin of Narmada river in India
respectively contributed to increased deforestation and expansion of undesired settlements.
Berihun et al. (2019) also identified population growth as one of the main driving forces of
LULC changes taking place in Upper Nile basin of Ethiopia.

Based on the authors’ local and field visit knowledge, it has been observed that newer
settlements have been predominantly expanding into the Dedza township. These new settlements
also tended to develop in close proximity to main roads (M1, M5 and M10 roads), along the
lakeshore of Lake Malawi (Mtakataka and Golomoti) and Lobi trading centres. Scattered
settlements and rural growth centres were also observed in the rural areas of the district. The
Mtakataka and Golomoti settlements situated along the lakeshore are tourist attraction sites of
the districts. The increase in tourism activities attracts people to centralize along also tends to
attract settlements along the lakeshore areas for job and business opportunities. It is worth noting
that Dedza district do not have a land-use plan and this may contribute to the township growth
and decline in agricultural land for crop production. The haphazard/unplanned nature of
township growth/development and scattered settlements in the studied landscape is therefore

15
seen as an outcome of ineffective and disorganized land-use planning in the area. Moreover, the
haphazard nature of settlements growth is also an indication of the absence of long-term
strategies to provide guidance for sustainable land use planning. This may contribute to the
expansion of uncontrolled settlements as observed through the modelled LULC changes in the
study area. Rimal et al. (2017) similarly reported that scattered settlements and random
urbanization in Jhapa district of southern Nepal were also a result of ineffective urban land use
planning. In order to reduce these undesired scattered settlements and random urbanization, their
study recommended the promotion of consolidated and compact settlements. Further, they
advised the decision-makers and land use planners to fully implement the land-use acts, policies,
laws and regulations the government of Nepal has introduced. Borrowing from the
recommendations from these studies, we similarly recommend that the planners in Dedza district
should also develop their land use policies, laws and regulations to address the undesired future
LULC changes for the achievement of sustainable natural resource management and
development.

The modelled results have also shown that agricultural land is likely to continue being as a
dominant, important and influential LULC class in the future. However, it is projected to decline.
The projected decline in agricultural land in Dedza district is a great concern which will affect
the crop production which will eventually contribute to food insecurity in the future. This may
also indirectly affect the sustainable management, use and conservation of land and other natural
resources in the studied landscape. Additionally, it is expected that the expanding residential
areas and increasing population will continue to highly contribute to the declining agricultural
land. This will make other LULC classes such as wetlands and forests the most vulnerable LULC
classes to spatial changes. The results are consistent with recent findings of declined cultivated
land on remaining forest resources in SSA, for instance, in Zimbabwe (Baudron et al. 2012),
Tanzania (Estes et al. 2012), Ghana (Appiah et al. 2015) and Ethiopia (Kindu et al. 2016).
Hence, an integrated land-use management approach is crucial to improve the multi-functionality
of Dedza landscape for natural resource management and conservation, food security and
livelihood enhancement.

5. Conclusion

16
 Understanding LULC spatial pattern, magnitude and trends of any landscape is imperative
for effective natural resource management, planning and use and secure sustainable
development. This study has demonstrated how the spatial-temporal LULC pattern and trends of
Dedza district for 2025 and 2035 can be predicted using an integrated CA-Markov model (a
hybrid of the Cellular automata and Markov models integrated into IDRISI software). The study
prediction of future LULC changes in Dedza district and Malawi at large; using CA-Markov
model is the first one of its kind as far as literature documented is concerned. In order to achieve
a perfect and better future LULC results, the model was validated using the kappa variations
namely; Kno, Klocation and Kquantity The model validation results show that the overall agreement
between the observed and simulated LULC maps of 2015 was perfect, hence, the CA-Markov
has proven to be a good and useful tool for accurate prediction of future spatial-temporal LULC
dynamics in the studied landscape. The model is therefore important to land use policy design
and planning especially involve with LULC development which requires a framework for
achieving goals and objectives of sustainable land use development.

Dedza has undergone tremendous LULC changes and these will continue as projected in
this study. The predicted LULC changes for 2025 and 2035 show the continuation of the same
trend and pattern of the recent past except for water bodies. The future projections indicate that
water bodies, barren land and built-up areas will increase while agricultural land, wetlands and
forest land will substantially decrease. The forest is being converted to agricultural land and
barren land. This is not good news for forest managers in Dedza district and these forest
conversions need to be controlled and a harmonized land-use plan needs to be developed that
promote forest resource rehabilitation and conservation. The undesired predicted LULC changes
are an early warning signal to natural resource managers, planners, policy makers and local
communities in the studied landscape to prepare better strategies and land-use policies to ensure
the expected unmannered expansion of barren land and built-up areas do not cause adverse
environmental impacts.

With the effective use of CA-Markov model, the study findings have provided preliminary
guidelines and baseline information that can contribute towards the sustainable management of
natural resources and the reduction of forest degradation and deforestation. Findings from the
study also highlights the need to implement sustainable land-use plans and holistic sustainable

17
development policies/strategies/guidelines. These findings serve as an important benchmark to
planners, natural resource managers and policy-makers in the studied landscape for planning and
management of natural resources and rehabilitating the degraded forest areas through
afforestation and reforestation. Finally, the study proposes comparative studies to be undertaken
across different landscapes of Malawi for the CA-Markov model to be adapted to other districts
of similar settings and Malawi at large.

Acknowledgment: The research study was supported by the Organization for Women in Science for the Developing
World (OWSD), University of Pretoria, the National Research Foundation (NRF) of South Africa, and the
Schlumberger Foundation through its Faculty for the Future Program. We would also like to thank the Malawi
Department of Forestry and the NSO for the ancillary data used in this study. Harold Chisale for his support in
editing and reviewing the paper before submission for publication

Conflict of interest: The authors declare no conflict of interest

References

1. Appiah, D.O., Schröder, D., Forkuo, E.K. and Bugri, J.T. 2015. Application of geo-
information techniques in land use and land cover change analysis in a peri-urban district
of Ghana. ISPRS Int. J. Geo-Inf., 4, 1265–1289.
2. Araya, Y.H. and Cabral, P. 2010. Analysis and modeling of urban land cover change in
Setúbal and Sesimbra, Portugal. Remote Sens., 2, 1549–1563.
3. Arsanjani, J.J, Helbich, M., Kainz, W., and Boloorani, A.D. 2013). Integration of logistic
regression, Markov chain and cellular automata models to simulate urban expansion. Int J
Appl Earth ObsGeoinf 21:265–275
4. Basommi, L.P., Qing-feng G., Dan-dan, C. and Singh, S.K. 2016. Dynamics of land use
change in a mining area: a case study of Nadowli District. J Mt Sci. 13(4):633–642.
5. Baudron,F., Jens, A., Corbeels, M. and Ken E. Giller, K.E .2012.. Failing to Yield?
Ploughs, Conservation Agriculture and the Problem of Agricultural Intensification: An
Example from the Zambezi Valley, Zimbabwe, Journal of Development Studies, 48:3,
393-412
6. Behera, M.D., Borate S.N., Panda, S.N., Behera, P.R. and Roy, P.S. 2012. Modelling and
analyzing the watershed dynamics using Cellular Automata (CA)-Markov model—a geo-
information-based approach. J Earth SystSci 121:1011–1024
7. Berihun, M.L., Tsunekawa, A., Haregeweyn, N., Meshesha, D.T., Adgo, E., Tsubo,M.,
Masunaga,T., Fenta, A.A., Sultan, D., and Yibeltal, M.2019. Exploring land use/land
cover changes, drivers and their implications in contrasting agro-ecological
environments of Ethiopia. Land use policy, Volume 87, 2019, 104052
8. Bhattacharjee, S. and Ghosh, S.K. 2015. Spatio-temporal change modeling of LULC: a
semantic Kriging Approach ISPRS Ann. Photogramm., Remote Sens. Spat. Inf. Sci., II-
4/W2, pp. 177-184
9. Chavula, G., Brezonik, P. and Bauer, M. 2011. Land Use and Land Cover Change
(LULC) in the Lake Malawi Drainage Basin, 1982-2005. International Journal of
Geosciences, 2(2, pp. 172-178

18
10. Chomitz, K.M., Buys, P., De Luca, G., Thomas, T.S. and Wertz-Kanounnikoff, S. 2007
At loggerheads? Agricultural expansion, poverty reduction, and environment in tropical
forests. World Bank Policy Research Report. The World Bank, Washington, DC.
11. Congalton, R.G. and Green, K.2009. Assessing the Accuracy of Remotely Sensed Data:
Principles and Practices. The Photogrammetric Record, CRC Press, London
12. Dallimer, M., Davies, Z.G., Diaz-Porras, D.F., Irvine, K.N., Maltby, L., Warren, P.H.,
Armsworth, P.R. and Gaston, K.J. 2015. Historical influences on the current provision of
multiple ecosystem services. Global Environ. 31, 307–317.
13. Defries R.S., Rudel T., Uriarte M. and Hansen M.. 2010. Deforestation driven by urban
population growth and agricultural trade in the twenty-first century. Nature Geoscience 3:
178–181
14. Dezhkam, S., Amiri, B.J., Darvishsefat, A.A. and Sakieh, Y. 2017. Performance
evaluation of land change simulation models using landscape metrics. Geocarto Int., 32,
655–677
15. Estes, A.B., Kuemmerle, T., Kushnir, H., Radeloff, V.C. and Shugart, H.H. 2002.
Land-cover change and human population trends in the greater Serengeti ecosystem from
1984–2003. Biol. Conserv., 147, pp. 255-263
16. Foley, J.A., DeFries, R., Asner, G.P., Barford, C., Bonan, G., Carpenter, S.R., Chapin,
F.S., Coe, M.T., Daily, G.C., Gibbs, H.K., et al. 2005. Global consequences of land use.
Science , 309, 570–574.
17. Government of Malawi.1999. Dedza Social—Economic Profile; A Report by Dedza
District Council: Blantyre, Malawi, 1999.
18. Government of Malawi. 2010. Dedza district state of environment and outlook. A report
by Dedza District Council. Dedza, Malawi
19. Government of Malawi. 2012. Dedza town assemble urban social economic profile.
Ministry of Local Government, Capital Hill, Lilongwe 3
20. Government of Malawi. 2013. Dedza District Socio Economic Profile: 2013–2018; A
Report by Dedza District Council: Dedza, Malawi, 2013.
21. Government of Malawi. 2019. 2018 Population and Housing Census Main Report.
National Statistical Office, Zomba, Malawi
22. Guan, D.J., Li, H.F., Inohae, T., Su, W.C., Nagaie, T. and Hokao, K. 2011. Modeling
urban land use change by the integration of cellular automaton and Markov model. Ecol.
Model. 222, 3761–3772.
23. Guan, D.J., Gao, W.J., Watari, K. and Fukahori, H. 2008. Land use change of Kitakyushu
based on landscape ecology and Markov model. Journal of Geographical Sciences 18,
455–468.
24. Haack, B., Mahabir, R. and Kerkering, J.2014. Remote sensing‐derived national land
cover land use maps: A comparison for Malawi. Geocarto Int. , 30, 270–292.
25. Han, H., Yang, C. and Song, J. 2015. Scenario simulation and the prediction of land use
and land cover change in Beijing, China. Sustainability, 7, 4260–4279
26. He, C., Okada, N., Zhang, Q., Shi, P. and. Zhang, J. 2006. Modeling urban expansion
scenarios by coupling cellular automata model and system dynamic model in Beijing,
China. Applied Geography,.26, pp. 323-345.
27. He, D., Zhou, J., Gao, W., Guo, H., Yu, S. and Liu, Y. 2014. An integrated CA-markov
model for dynamic simulation of land use change in Lake Dianchi watershed. Beijing

19
 DaxueXuebao (ZiranKexuBan)/ActaScientiarumNaturaliumUniversitatisPekinensis.
2014; 50(6):1095-1105.
28. Hu, H., Liu, W. and Cao, M. 2008. Impact of land use and land cover changes on
ecosystem services in Menglun, Xishuangbanna, Southwest China. Environ. Monit.
Assess. 146, 147–156.
29. Hyandye, C. and Martz, L.W. 2017. A Markovian and cellular automata land-use change
predictive model of the Usangu Catchment. Int J Remote Sens. 38(1):64-81.
30. Irwin, E.G, Jayaprakash, C. and Munroe, D.K. 2009. Towards a comprehensive
framework for modeling urban spatial dynamics. Landscape Ecol. 24(9):1223–1236.
31. Jagger, P. and Perez-Heydrich, C. 2016. Land use and household energy dynamics in
Malawi Environ. Res. Lett., 11 (12), 125004
32. Kamusoko , C., Aniya, M., Adi, B. and Manjoro M. 2009. Rural sustainability under
threat in Zimbabwe–simulation of future land use/cover changes in the Bindura District
based on the markov-cellular automata model. Appl Geogr 29(3):435–44
33. Kamusoko, C., Oono, K., Nakazawa, A., Wada, Y., Nakada, R., Hosokawa, T.,
Tomimura, S., Furuya, T., Iwata, A., Moriike, H., Someya, T., Yamase, T., Nasu, M.,
Gomi, Y., Sano, T., Isobe, T. and Homsysavath, K., 2011. Spatial simulation modelling
of future forest cover change scenarios in Luangprabang Province, Lao PDR. Forests 2,
707–729.
34. Kelly, O., Carlos, A., Gustavo, E., Alexandre, S., Nero, L., Getulio, F., José, R. and
Alexandre, R. 2018. Markov chains and cellular automata to predict environments subject
to desertification. Journal of Environmental Management, 225, 160–167.
35. Keshtkar, H. and Voigt, W. 2016. A spatiotemporal analysis of landscape change using
integrated Markov chain cellular automata model. Model. Earth Syst. Environ., 2, 1–13.
36. Kindu, M., Schneider, T., Teketay, T. and Knoke, T. 2015. Drivers of land use/land cover
changes in Munessa Shashemene landscape of the south‐central highlands of Ethiopia.
Environ. Monit. Assess. 187, 452.
37. Kindu,M., Schneider, T., Teketay, D., Knoke, T., 2016. Changes of ecosystem service
values in response to land use/land cover dynamics in Munessa–Shashemene landscape
of the Ethiopian highlands. Sci. Total Environ. 547, 137–147.
38. Koomen, E., and J. Stillwell. 2007. Modelling Land-Use Change, 1–22. Netherlands:
Springer.
39. Kumar, S., Radhakrishnan, N. and Mathew, S. 2014. Land use change modelling using a
Markov model and remote sensing. Geomat. Nat. Hazards Risk., 5, 145–156
40. Lamichhane, B. B. 2008. Dynamics and Driving Forces of Land Use/Forest Cover
Change and Indication of Climate Change in a Mountain Sub-watershed of Gorkha. MSc.
thesis, Tribhuvan University, Institute of Forestery, Nepal.
41. Landis, R.J. and Koch, G. G. 1977. The measurement of observer agreement for
categorical data. Biometrics 33(1): 159-174.
42. Liu X, Li X, Yeh AG-O, He J and Tao, J. 2007. Discovery of transition rules for
geographical cellular automata by using ant colony optimization. Sci China SerDEarthSci
50:1578–1588.
43. Liping, C., Yujun, S. and Saeed, S. 2018. Monitoring and Predicting Land Use and Land
Cover Changes Using Remote Sensing and GIS Techniques—A Case Study of a Hilly
Area, Jiangle, China. PLoS ONE, 13, e0200493

20
44. Loomis, J. 2002. Integrated public lands management: principles and applications to
national forests, parks and wildlife refuges, and BLM lands. 2nd ed. New York:
Columbia University Press. US
45. Ma, C., Zhang, G.Y, Zhang, X.C, Zhao, Y.J and Li, H.Y. 2012. Application of Markov
model in wetland change dynamics in Tianjin Coastal Area, China. Procedia Environ.
Sci. 13, 252-262.
46. Maestas, J.D., Knight, R.L. and Gilgert, W.C. 2003. Biodiversity across a rural land-use
gradient. Conserv Biol. 17(5):1425–34.
47. Mannan, A., Zhongke, F., Ahmad, A., Liu, J., Saeed, S., Mukete, B., 2018. Carbon
dynamic shifts with land use change in Margallah hills national park, Islamabad
(pakistan) from 1990 to 2017. Appl. Ecol. Environ. Res. 16 (3), 3197-214.
48. Martinuzzi, S., Radeloff, V.C., Joppa, L.N., Hamilton, C.M., Helmers, D.P., Plantinga,
A.J. and Lewis, D.J. 2015. Scenarios of future land use change around United States'
protected areas. Biol. Conserv. 184, 446–455.
49. Memarian, H., Balasundram, S. K., Talib, J. B., Sung, C. T. B., Sood, A. M. and
Abbaspour, K. 2012. Validation of CA-Markov for Simulation of Land Use and Cover
Change in the Langat Basin, Malaysia. Journal of Geographic Information System 4 (6):
542–554.
50. Mishra, M., Mishra, K., Subudhi, A. and Phil, M. 2011. Urban sprawl mapping and land
use change analysis using remote sensing and GIS: Case study of Bhubaneswar city,
Orissa. Proceedings of the Geo-Spatial World Forum, January 18-21, Hyderabad, India.
51. Mishra, V. N., Rai, P. K. and Mohan, K. 2014. Prediction of land use changes based on
land change modeler (LCM) using remote sensing: A case study of Muzaffarpur (Bihar),
India. Journal of the Geographical Institute Jovan Cvijic, SASA, 64 (1), 111–127.
52. Mondal, M.S., Sharma, N., Garg, P.K and Kappas, M. 2016 Statistical independence test
and validation of CA Markov land use land cover (LULC) prediction results. Egypt J
Remote Sens Space Sci19(2):259–272
53. Munthali, M.G., Botai, J.O., Davis, N. and Adeola, A.M. (2019a). Multi‐temporal
analysis of land use and land cover changes detection for Dedza District of Malawi using
geospatial techniques. International Journal of Applied Engineering Research, 14(5),
1151 – 1162.
54. Munthali, M. G., Davis, N., Adeola, A M., Botai, J.O., Kamwi, J.M., Chisale H.L.W.
and Orimoogunje, O.O.I. (2019b). Local Perception of Drivers of Land-Use and Land-
Cover Change Dynamics across Dedza District, Central Malawi Region" Sustainability,
MDPI, , vol. 11(3), 1-25
55. Munthali, K.G. and Murayama, 2011. Y. Land use/cover change detection and analysis
for Dzalanyama forest reserve, Lilongwe, Malawi. Procedia-Soc. Behav. Sci. , 21, 203–
211.
56. Munthali, K.G. and Murayama, Y.2014. Modeling Deforestation in Dzalanyama Forest
Reserve, Lilongwe, Malawi:A Multi‐Agent Simulation Approach. GeoJournal , 80, 1–
15.
57. Myint, S.W. and Wang, L. 2006. Multicriteria decision approach for land use land cover
change using Markov chain analysis and a cellular automata approach. Can J Remote
Sens. 32(6):390±404

21
58. Nadoushan, M.A., Soffianian, A. and Alebrahim, A. 2015. Modeling land use/cover
changes by the combination of markov chain and cellular automata markov (CA-Markov)
models. J. Earth Environ. Health Sci. 1, 16–21.
59. Omar, N.Q., Ahamad, M.S.S., Hussin, W.M.A.W., Samat, N. and Ahmad, S.Z.B. 2014.
Markov CA, multi regression, and multiple decision making for modeling historical
changes in Kirkuk City, Iraq. J Indian Soc Remote Sens 42(1):165–178.
60. O’Sullivan, D. 2001. Graph-cellular automata: a generalised discrete urban and regional
model. Environ PlannB.28(5):687–705.
61. Paegelow M., and Camacho Olmedo M.T., 2010, Modelos de simulación espacio-
temporal y Teledetección:el método de la segmentación para la cartografía
cronológica de usos del suelo”. Serie Geográfica Universidad de Alcalá n° 16; p.19-34
62. Palamuleni, L.G., Annegarn, H.J. and Landmann, T. 2010.Land cover mapping in the
Upper Shire River catchment in Malawi using Landsat satellite data. Geocarto Int., 25,
503–523.
63. Pandey, B.K. and Khare D .2017. Analysing and modeling of a large river basin
dynamics applying integrated cellular automata and Markov model. Environ Earth Sci
76(22):779
64. Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M.J. and Deadman, P. 2003.
Multi agent systems for the simulation of land-use and land-cover change: a review”,
Annals of the Association of American Geographers, 93, 314-337.
65. Parsa, V.A.; Yavari, A. and Nejadi, A. 2016. Spatio-temporal analysis of land use/land
cover pattern changes in Arasbaran Biosphere Reserve: Iran. Model. Earth Syst. Environ.
2, 1–13.
66. Peña, J., A. Bonet, J. Bellot, J. R. Sánchez, D. Eisenhuth, S. Hallett, and A. Aledo. 2007.
“Driving Forces of Land-Use Change in a Cultural Landscape of Spain.” In Modelling
Land-Use Change, 97–116. The Netherlands: Springer.
67. Pontius Jr., R.G., 2000.Quantification error versus location error in comparison of
categorical maps. Photogrammetric Engineering & Remote Sensing, 66 (8), 1011-1016
68. Pontius, R.G. and Schneider, L.C.2001. Land-cover change model validation by an ROC
method for the Ipswich watershed, Massachusetts, USA. Agric. Ecosyst. Environ. 85,
239–248
69. Pontius, G.R. and Malanson, J. 2005. Comparison of the structure and accuracy of two
land change models. Int. J. Geogr. Inf. Sci. 19, 243–265
70. Pullanikkatil, D., Palamuleni, L.G. and Ruhiiga, T.M. 2016. Land use/land cover change
and implications for ecosystems services in the Likangala River Catchment, Malawi.
Phys. Chem. Earth , 93, 96–103.
71. Qiang, Y. and Lam, N.S.N. 2015. Modeling land use and land cover changes in a
vulnerable coastal region using artificial neural networks and cellular automata.
Environ.Monit. Assess. 187.
72. Qiu, B and Chen, C. 2008. Land use change simulation model based on MCDM and CA
and its application. ACTA Geogr Sin 63:165–174
73. Ralha, C.G., Abreu, C.G., Coelho, C.G.C., Zaghetto, A., Macchiavello, B. and Machado,
R.B. 2013. A multi-agent model system for land-use change simulation. Remote Sens
Environ. 42:30-46.

22
74. Regmi, R.; Saha, S. and Balla, M. 2014. Geospatial analysis of land use land cover
change predictive modeling at PhewaLakeWatershed of Nepal. Int. J. Curr. Eng. Tech. 4,
2617–2627.
75. Rendana, M., Rahim, S.A., Mohd, R.I.W., Lihan, T. and Rahman, Z.A. CA-Markov for
predicting land use changes in tropical catchment area: a case study in Cameron
Highland, Malaysia. Journal of Applied Sciences. 2015; 15(4):689-95.
76. Rimal, B., Zhang, L., Keshtkar, H., Wang, N., and Lin, Y. 2017. Monitoring and
modelling of spatiotemporal urban expansion and land-use/land cover changes using
integrated Markov Chain Cellular Automata Model. International Journal of
Geoinformation, 6, 288.
77. Sang, L.; Zhang, C.; Yang, J.; Zhu, D. and Yun,W. 2011. Simulation of land use spatial
pattern of towns and villages based on CA-Markov model. Math. Comput. Model., 54,
938–943
78. Sayemuzzaman, M.and Jha, M. 2014.Modeling of future land cover land use change in
North Carolina using Markov chain and cellular automata model. Am. J. Eng. Appl. Sci.,
7, 295–306.
79. Serneels, S. and Lambin, E.F. 2001. Proximate causes of land‐use change in Narok
district, Kenya: A spatial statistical model. Agric. Ecosyst. Environ., 85, 65–81.
80. Singh, S.K., Mustak, S., Srivastava, P.K., Szabó, S. and Islam, T. 2015. Predicting spatial
and decadal LULC changes through cellular automata Markov chain models using earth
observation datasets and geo-information. Environ Process. 2:61–78
81. Singh, S.K., Basommi, B.P., Mustak, S.k., Srivastava, P.K. and Szabo, S. 2018.
Modelling of land use land cover change using earth observation data-sets of Tons River
Basin, Madhya Pradesh, India. Geocarto Int. 33 (11). 1202-1222.
82. Sohl, T.L. and Claggett, P.R. 2013. Clarity versus complexity: Land-use modeling as a
practical tool for decisionmakers. J Environ Manage. 129:235±43.
83. Solomon, N., Hishe, H., Annang, T., Pabi, O., Asante, I., & Birhane, E. (2018). Forest
cover change, key drivers and community perception in Wujig Mahgo Waren forest of
Northern Ethiopia. Land, 7(1), 32.
84. Srivastava, P.K., Han, D., Rico-Ramirez, M.A. and Islam, T. 2013.Sensitivity and
uncertainty analysis of mesoscale model downscaled hydro-meteorological variables for
discharge prediction. Hydrol.Process.28(15),4419–4432.
85. Steeb, W.H. 2011. The nonlinear workbook: Chaos, Fractals, Cellular Automata, Neural
Networks, Genetic Algorithms, Gene Expression Programming, Support Vector Machine,
Wavelets, Hidden Markov Models, Fuzzy Logic With C++, Java And Symbolicc++
Programs: World Scientific
86. Stefanov, W.L., Ramsey, M.S. and Christensen, P.R. 2001. Monitoring urban land cover
change: An expert system approach to land cover classification of semiarid to arid urban
centers. Remote Sens Environ. 77 (2):173±85.
87. Stevens, D. and Dragićević, S. A. 2007. GIS-Based Irregular Cellular Automata Model of
Land-Use Change. Environ.Plann B. 34(4):708±24
88. Subedi, P.; Subedi, K. and Thapa, B. 2013. Application of a hybrid cellular automaton–
Markov (CA-Markov) Model in land-use change prediction: A case study of saddle creek
drainage Basin, Florida. Appl. Ecol. Environ. Sci., 1, 126–132.

23
89. Sun, Q., Wu, Z., and Tan, J. 2012. The relationship between land surface temperature
and land use/land cover in Guangzhou, China. Environmental Earth Sciences, 65(6),
1687–1694.
90. Thomas, H. and Laurence, H.M. 2006 Modeling and projecting land-use and land-cover
changes with a cellular automaton in considering landscape trajectories: An improvement
for simulation of plausible future states; EARSeLeProc. 5 63–76.
91. Theobald, D. and Hobbs, N. 2002. A framework for evaluating land use planning
alternatives: protecting biodiversity on private land. ConservEcol., 6(1):5.
92. Varga, O.G., Pontious Jr, R.G., Singh, S.K. and Szabo, S. 2019. Intensity Analysis and
the Figure of Merit’s components for assessment of a Cellular Automata – Markov
simulation model. Ecological Indicators. 101: 933-942.
93. Verburg, P.H.; Soepboer, W.; Veldkamp, A.; Limpiada, R.; Espaldon, V. and Mastura,
S.S. 2002. Modeling the spatial dynamics of regional land use: The CLUE-S model.
Environ. Manag. 30,391–405.
94. Verburg, P.H., Schot, P.P., Dijst, M.J. and Veldkamp A. 2004. Land use change
modelling: current practice and research priorities.GeoJournal 61:309–324
95. Verburg, P. H., K. Kok, R. G. Pontius Jr, and A. Veldkamp. 2006. Modeling Land-Use
and Land-Cover Change. Land-Use and Land-Cover Change, 117–135. Berlin,
Heidelberg: Springer.
96. Verburg, P.H., Bakker, M., Overmars, K.P. and Staritsky, I. 2008. Landscapelevel
simulation of land use change. In: Helming K, Pe ́rez-SobaM, Tabbush P (eds)
Sustainability impact assessment of land usechanges. Springer, New york, pp 211–227
97. Viera, A. J. and Garrett, J. M. 2005. Understanding interobserver agreement: the kappa
statistic. – Fam Med. 37 (5): 360-363.
98. Wang, Y. and Zhang, X. 2001. A dynamic modeling approach to simulating
socioeconomic effects on landscape changes. Ecol Model. 140(1):141–162.
99. Wang, S.Q., Zheng, X.Q. and Zang, X.B. 2012. Accuracy assessments of land use change
simulation based on Markov-cellular automata model. Procedia Environ Sci 13:1238–
1245.
100. Wu, F. and Webster, C.J. 2000. Simulating artificial cities in a GIS environment:
urban growth under alternative regulation regimes. Int J Geograph Inform Sci.
14(7):625–648.
101. Wu, F. 2002 Calibration of stochastic cellular automata: the application to rural–
urban land conversions. Int J Geogr Inf Sci 16(8):795–818
102. Wu, L. Shi, P. and Gao, H. 2010. State estimation and sliding-mode control of
markovian jump singular systems IEEE Transactions on Automatic Control, 55 (5), pp.
1213-1219
103. Wu, Q.; Li, H.Q.; Wang, R.S.; Paulussen, J.; He, Y.; Wang, M. and Wang, Z. 2006.
Monitoring and predicting land use change in Beijing using remote sensing and GIS.
Landsc. Urban Plan. 78, 322–333.
104. Xie, Y.C.; Batty, M. and Zhao, K. 2007. Simulating emergent urban form using
agent-based modeling: Desakota in the suzhou-wuxian region in China. Ann. Assoc. Am.
Geogr. 97, 477–495
105. Yang, Q.S. and Li, X. 2007. Integration of multi-agent systems with cellular
automata for simulating urban land expansion. ScientiaGeographicaSinica 27, 542–548.

24
106. Yang, X., Zheng X.Q and Lv L. 2012. A spatiotemporal model of land use change
based on ant colony optimization, Markov chain and cellular automata. Ecol Model
233:11–19.
107. Yang, X., Zheng, X.C. and Chen, R. 2014. A land use change model: integrating
landscape pattern indexes and Markov-CA. Ecol Model 283:1–7
108. Yesuph, A.Y. and Dagnew, A.B. 2019. Land use/cover spatiotemporal dynamics,
driving forces and implications at the Beshillo catchment of the Blue Nile Basin, North
Eastern Highlands of Ethiopia. Environmenttal Systems Research, 8:21
109. Yirsaw, E., Wu, W., Shi, X., Temesgen H. and Bekele B. 2017. Land Use/Land
Cover Change Modeling and the Prediction of Subsequent Changes in Ecosystem
Service Values in a Coastal Area of China, the Su-Xi-Chang Region. Sustainability,
9(7), 1204
110. Yousheng, W., Xinxiao, Y., Kangning, H., Qingyun, L., Yousong, Z. and Siming S.
2011. Dynamic simulation of land use change in Jihe watershed based on CA-Markov
model. Transactions of the Chinese Society of Agricultural Engineering. 2011(12).
111. Yu, F. 2009. Study on forecast of land use change based on Markov-CA. Land Res.
Inf. (In Chinese), 4, 38–46.
112. Zhao, L and Peng, Z.R. 2012. LandSys: an agent-based Cellular Automata model of
land use change developed for transportation analysis. J TranspGeogr. 25:35±49

25
 Table 1. Detailed information about datasets used in this study
Dataset Spatial resolution (m) Date of Source
acquisition
Landsat 5 TM (Path/row: 30 16/09/1991 USGS
168/070)
Landsat 7 30 19/09/2001 USGS
ETM+(Path/row:
168/070)
Landsat 8 OLI (Path/row: 30 18/09/2015 USGS
168/070)
STRM* 30
*Data collected from http://dwtkns.com/srtm30m/

Table 2. Description of LULC classes

LULC class Description

Water bodies Rivers, permanent open water, lakes, ponds, reservoirs
Wetland Permanent and seasonal grasslands along the lake, river and streams, marshy
land and swamps
Agricultural All cultivated and uncultivated agricultural lands areas such as farmlands, crop
land fields including fallow lands/plots and Horticultural lands.
Forest Protected forests, plantations, deciduous forest, mixed forest lands and forest on
customary land.
Built-up area Residential, commercial and services, industrial, socio-economic infrastructure
and mixed urban and other urban, transportation, roads and airport.
Barren land Areas around and within forest protected areas with no or very little vegetation
cover including exposed soils, stock quarry, rocks, landfill sites, and areas of
active excavation.
 Table 3: Constraints and factors (Boolean approach) criteria development
Constraint/Factor Source Description
LULC Classified LULC map Agricultural land was considered the possible
LULC class available for development
Existing water Classified maps and All the existing rivers and water bodies were
bodies Department of Surveys, considered unsuitable.
Malawi
Protected areas Department of Forestry, All the areas under government forest were
(forest reserves) Malawi considered unsuitable
Distance from Department of Surveys, All areas above a protection buffer zone of 30m
roads Malawi from the road networks were considered suitable for
development as stated in Malawi legislation.
Distance from Classified LULC map Protection of buffer of 50m from the rivers and
water bodies and Department of other water bodies was created and within 50m of
Surveys, Malawi the water bodies were considered unsuitable
Distance from Classified LULC map Areas close to the built-up areas were more suitable
built-up areas for development than areas far from built-up areas

Table 4: Temporal distribution of LULC classes (ha) and percentages of change

Change in LULC structure

LULC class 1991 2001 2015 ∆ % (1991- ∆ % (2001- ∆ % (1991-
2001) 2015) 2015)
Water 1,380.60 793.26 899.55 -74.04 11.82 -53.48
Wetland 3,626.73 2,954.07 2,680.29 -22.77 -10.21 -35.31
Forest 9,939.15 8,354.70 6,237.63 -18.96 -33.94 -59.34
Agriculture 267,977.43 267,469.83 260,879.31 -0.19 -2.53 -2.72
Barren 92,185.38 94,731.66 97,174.62 2.69 2.51 5.13
Built-up 761.67 1,567.44 7,999.56 51.41 80.41 90.48

Table 5: Comparison of the actual and simulated LULC classes in 2015

Differences between actual and

Area (ha)
LULC class projected LULC
Actual 2015 LULC Predicted 2015 LULC ∆ (ha) ∆ (%)
Water 899.55 356.76 -542.79 -60.34
Wetland 2,680.29 2,504.97 -175.32 -6.54
Forest 6,237.63 7,514.10 1,276.47 20.46
Agriculture 260,879.31 266,744.43 5,865.12 2.25
Barren 97,174.62 96,490.26 -684.36 -0.70
Built-up 7,999.56 2,260.44 -5,739.12 -71.74
Table 6: Results of validation analysis on agreements/disagreement components

Information of quantity
Information of location No [n] None Medium[m] Perfect [p]
Perfect[P(x)] P(n) = 0.4030 P(m) = 0.9927 P(p) = 1.0000
PerfectStratum[K(x)] K(n) = 0.4030 K(m) = 0.9927 K(p) = 1.0000
MediumGrid[M(x)] M(n) = 0.3919 M(m) = 0.9948 M(p) = 0.9740
MediumStratum[M(x)] H(n) = 0.1429 H(m) = 0.4623 H(p) = 0.4609
No [N(x)] N(n) = 0.1429 N(m) = 0.4623 N(p) = 0.4609

Chance agreement 0.1429

Quantity agreement 0.3195
Allocation agreement 0.5124
Allocation disagreement 0.0179
Quantity disagreement 0.0073

Table 7: Summary of Kappa variations of the CA-Markov model

Statistic Index
Kappa no information/ability 0.9706
Kappa location 0.9662
Kappa locationStrata 0.9962
Kappa Standard 0.9531

Table 8: Distribution of LULC changes from 2015 to 2035 (ha)

2015 2025 2035

LULC class
Area % Area % Area %
Water bodies 899.55 0.24 970.65 0.26 1,033.65 0.28
Wetlands 2,680.29 0.71 2,555.19 0.68 2,471.49 0.66
Forest 6,237.63 1.66 5,779.53 1.54 5,631.03 1.50
Agricultural land 260,879.31 69.41 256,875.21 68.34 253,394.01 67.42
Barren land 97,174.62 25.85 97,803.72 26.02 98,076.42 26.09
Built-up areas 7,999.56 2.13 11,886.66 3.16 15,264.36 4.06
Total 375,870.96 100.00 375,870.96 100.00 375,870.96 100.00
 Table 9: Transition probability of areas (ha) and matrix from 2015 to 2025

Actual LULC map of 2015

2015/2025 Water Wetlands Forest Agriculture Barren Built-up
Water Area 860.58 2.43 2.88 33.39 0.27 0.00
P 0.9566 0.0027 0.0032 0.0371 0.0030 0.0000
Predicted LULC map of 2025

Wetlands Area 0.54 1818.00 36.18 13.05 812.43 0.00

P 0.0002 0.6783 0.0135 0.0049 0.3031 0.0000
Forest Area 1.35 41.76 2197.17 202.86 3791.70 2.79
P 0.0002 0.0067 0.3523 0.0325 0.6079 0.0004
Agriculture Area 106.92 3.33 311.76 254619.27 1063.26 4774.77
P 0.0004 0.0000 0.0012 0.9760 0.0041 0.0183
Barren Area 0.0000 689.85 3229.92 1089.63 92128.50 36.72
P 0.0000 0.0071 0.0332 0.0112 0.9481 0.0004
Area 0.0000 0.0000 1.62 917.19 8.01 7,072.74

Built-up P 0.0000 0.0000 0.0002 0.1147 0.0010 0.8841

Note: Bold numbers on the diagonal represent LULC proportions will likely remain constant/stable between 2015 to 2025,
while others are the areas are likely to be converted from one class to another; P means probability

Table 10: Transition probability of areas (ha) and matrix from 2015 to 2035

Actual LULC map of 2015

2015/2035 Water Wetlands Forest Agriculture Barren Built-up
Water Area 825.66 4.05 3.87 64.62 3.15 0.54
P 0.9151 0.0045 0.0043 0.0719 0.0035 0.0006
Predicted LULC map of 2035

Wetlands Area 0.99 1,231.47 64.71 31.68 1,351.08 0.27

P 0.0004 0.4595 0.0242 0.0118 0.5041 0.0001
Forest Area 1.89 70.38 881.01 313.20 4,962.69 8.55
P 0.0003 0.0113 0.1412 0.0502 0.7956 0.0014
Agriculture Area 206.82 15.12 452.07 249,065.55 2,240.82 8,899.02
P 0.0008 0.0001 0.0017 0.9547 0.0086 0.0341
Barren Area 0.54 1,150.02 4,226.85 2,209.50 89,499.60 88.11
P 0 0.0118 0.0435 0.0227 0.9210 0.0009
Built-up Area 0.18 0 3.33 1,709.73 19.26 6,266.97
P 0 0 0.0004 0.2137 0.0024 0.7434
Note: Bold numbers on the diagonal represent LULC proportions will likely remain constant/stable between 2015 to 2035,
while others areas are likely to be converted from one class to another; P means probability
 List of Figures

Figure 1: Location of the study area

Figure 2. LULC maps for 1991, 2001 and 2015 (Munthali et al. 2019a)
 Data sources

Land change driver

Landsat images
Ancillary data • Field surveys variables e.g.
• Distance to roads
1991, 2001 and 2015 • Google archived • Distance to water
(Road, River and
images bodies
Protected Forest) • Protected areas (forest
reserves)
Image preprocessing
3. Results
Standardization (Fuzzy)
and weighting (AHP)
Image classification
Accuracy
assessment Multi Criteria Evaluation
LULC maps (MCE)
1991, 2001 & 2015

Transition suitability
Markov model Conditional maps
probability map

Transition Transition
probability matrix area matrix
CA-Markov
model
Predicted LULC maps for
2025, 2035
Model validation
(Actual and
Calibrated CA- simulates LULC
Markov model map of 2015)

Figure 3. Flow chart for predicting LULC changes in the study area
Figure 4: Suitability maps for different LULC classes
Figure 5: Actual map and simulated map of LULC for 2015
Figure 6: Projected maps of LULC for 2025 and 2035
Ethical Statement for Remote Sensing Applications: Society
and Environment

I testify on behalf of all co-authors that our article submitted to Remote Sensing
Applications: Society and Environment:

Title: Modeling land use and land cover dynamics of Dedza district of
Malawi using hybrid Cellular Automata and Markov model

M. G. Munthali, S. Mustak, A. Adeola, J. Botai, S.K. Singh, N. Davis:

1) this material has not been published in whole or in part elsewhere;

2) the manuscript is not currently being considered for publication in another
journal;
3) all authors have been personally and actively involved in substantive work
leading to the manuscript, and will hold themselves jointly and individually
responsible for its content.

Date: 2019-08-20

Corresponding author’s signature: