Type of the Paper (Articleeajournal.unilak.ac.rw/Vol 8 Special Issue 2/PAPER ENAN... · The 5th...
Transcript of Type of the Paper (Articleeajournal.unilak.ac.rw/Vol 8 Special Issue 2/PAPER ENAN... · The 5th...
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. P81-102
81
Prediction of Land Use/Cover Change in the Congo Nile Ridge Region of Rwanda
Author: Enan Nyesheja Muhire 1,2,3,4, *
Affiliation:
1State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and
Geography, Chinese Academy of Sciences
2 University of Chinese Academy of Sciences, Beijing 100049, China
3 Faculty of Environmental Studies, University of Lay Adventists of Kigali (UNILAK),
P.O. Box 6392 Kigali, Rwanda
Corresponding author: [email protected] (E.N.M.), Mobile :+250788306534
Abstract: Simulation and mapping the changes of LULC became the major apprehension
to the environmental planners. In this research, we explored LULC dynamics in the Congo
Nile Ridge Region of Rwanda, from 1990 to 2010 and simulate future LULC in 2020 and
2030 with Land Change Modeler (LCM), MLP Neural network, Markov Chain (MC), and
GIS approaches. To map LULC changes and assess spatial trend, Landsat (TM and ETM+)
imageries of 1990, 2000, and 2010 were classified using supervised with likelihood
classification techniques in six classes. Kappa Statistics and overall accuracy for all LULC
(1990, 2000 and 2010) classification maps were over 94% and 93% respectively. Prediction
revealed that in 2020 and 2030, future LULC would expand for built-up 80 ha and 8,950 ha
for cropland, whereas grassland and forestland will decrease approximately to 2,550 ha and
6,480 ha, respectively; water bodies and wetland will remain constant. The findings from
this research can inform decision-making process concerning environmental protection.
Keywords: land use/cover; prediction; remote sensing; Rwanda
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. P81-102
82
1. Introduction
Land use refers to the purpose for which and
with the way in which, human beings
utilized land and its resources(REMA
(Rwandan Environmental Management
Authority), 2015). Land cover refers to the
habitat or vegetation types, such as forest,
water surfaces. Monitoring of Land User
and Land Cover (LULC) Changes can,
therefore, be valuable in ecologically
sensitive zones for protection of
biodiversity, improvement of land use
management and planning policies (Di
Gregorio, 2005; Lambin et al., 2003).
Moreover, the misuse of land cover
accelerates degradation of land,
eutrophication, desertification, variation
and change of climate, siltation, and
biodiversity loss (Akinyemi, 2017; Berakhi,
2013; Gillet et al., 2016; Shrestha and
Dwivedi, 2017). According to Wu et al.
(2012), from 2000 to 2050, the agriculture
land is estimated to be decreased in the
future in Central Africa, South Asia and in
the Eastern United States. It is therefore
believed that anthropogenic activities,
climatological change effects, inappropriate
land use management and natural processes
are the foremost causes of the LULC
dynamics (Arsanjani et al., 2013; Koranteng
and Zawila-Niedzwiecki, 2015; Lambin et
al., 2003).
As previously confirmed, globally, Land
Use and Land Cover Changes (LULCC) is
one of the main risks to aquatic living as
well as terrestrial habitats, in a special way
for forestland in Central Africa (Basnet and
Vodacek, 2015; Douglas, 2017; UNEP,
2016; Zubair et al., 2017). The research
conducted in Central Africa for tracking
LULC change in cloud prone area, the forest
cover loss in the epochs 1988 to 2001 and
2001 to 2011 were estimated of about 216.4
and 130.5 thousand hectares, respectively.
The triggering factors for the loss were
associated with deforestation during the
civil war in the Republic of Congo (Basnet
and Vodacek, 2015). Also, land
degradation has been the main problem in
Rwanda and particularly in the Congo Nile
Ridge Region of Rwanda which recorded
the peak land degradation level with
plentiful tropical precipitation (Goudie,
2013; Roose and Ndayizigiye, 1997). The
fourth population and housing census in
2012, indicated that Rwanda population was
10,515,973, with 52% female and 48%
male. Since 2002 till the 2012 census, the
population of Rwanda increased 2.4 million,
representing an average rate of growth
equaling to 2.6% annually, thus, a
population density of 415 persons/km2 in
2012 (NISR-Ministry of Finance and
Economic Planning, 2012). The high
population pressure to land also results in
land changes by increasingly disturbing the
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
83
biophysical and biogeochemical processes
of the earth’s surface (Basnet and Vodacek,
2015).
According to Nyandwi and Mukashema
(2011), in Congo Nile Ridge Region of
Rwanda, the two national park forests of
Nyungwe and Gishwati were decreased
approximately 74 % from 1988 to 2007, due
to settlement and agriculture provided by
the government of Rwanda to refugees
returning from exile after the 1994 genocide
perpetrated against Tutsi in Rwanda
(Lemarchand and Tash, 2015).
Furthermore, land conversion to cropland in
Africa is very predominant (Brink and Eva,
2011), and the Congo Nile Ridge Region
of Rwanda, as well as some other East
African countries, are dominated by land
dynamics in forest of tropical areas (Rattan
and Rattan, 2006), owing to the
transformation of land to cropland and built-
up (Berakhi, 2013). Additionally, LULC
dynamics monitoring and future prediction
necessitate a very particular attention by
land managers at all levels (local and
regional) as well as policymakers. This
helps to make knowledgeable decisions that
successfully stabilize the positive features
of development and its negative effects for
preserving environmental resources and
increasing socio-economic welfare (Di
Gregorio, 2005; Kamusoko et al., 2009).
LULC dynamics detection is becoming a
very important concern to conservationists,
environmentalists, policymakers, and land
use planners due to its significant impact on
natural ecosystems (Verburg et al., 2000).
For mapping and monitoring LULC
changes, Remote Sensing (RS) technology
is cost-effective, and it has been used
globally, locally and regionally (Giri, 2012).
The user-friendly applications with
effective LULCC modeling approaches are
significantly required for monitoring,
analyzing, validation, simulation and
predicting LULC (Aguejdad et al., 2017;
Kamusoko et al., 2009). More researchers
investigated the changes of LULC to
improve natural resources conservation in
Rwanda (Kamusoko et al., 2009; Nyandwi
and Mukashema, 2011). Spatial studies
related to LULC using technologies as
Remote Sensing (RS), Geographic
Information systems (GIS), LULCC
approaches with spatial statistics, transition
matrices, and spatial metrics have the
capability of measuring, analyzing,
mapping the spatial pattern of LULCC,
predicting the future land use and land
cover, modelling the relationship of LULC
and its driving factors (Arsanjani, 2011;
Bhatta, 2010; Rwanga and Ndambuki,
2017).
Different kinds of modelling systems of
LULCC were implemented to simulate
LULC for future prediction, such as Markov
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
84
Chain (Kumar et al., 2014; Yirsaw et al.,
2017), CLUES (Han et al., 2015; Verburg et
al., 2002), Land Change Modeler (Mishra et
al., 2014), Cellular Automata (Al-shalabi et
al., 2013), agent-based (Iizuka et al., 2017).
Among all the models, LCM has been
applied for Modelling LULCC for urban
and non-urban areas at vast spatial scales,
and also its feasibility as well as its
applicability, has been verified and tested in
significant numbers of research papers
(Iizuka et al., 2017). However, the spatial
model used in this research Multi-Layer
Perception (MLP) Neural Network, which is
able to simulate multiple-land use classes
(Houet and Hubert-Moy, 2006), and
Markov Chain (MC) the subroutine model
of Land Change Modeler (LCM) IDRISI
platform which are embedded in TerrSet
Software (Eastman, 2012). In this
framework, it is imperative to forecast the
LULC changes over the time and predict the
LULCC future status. Therefore, the
objectives of this research are:(i)
highlighting the most LULC dynamics in
quantity and clarifying its changes; (ii)
recognizing spatial patterns of land use and
land cover changes and its expansion
proportion; (iii) simulating the future LULC
change map for 2020 and 2030. The
outcomes of this investigation help to reveal
categories of LULC significant conversions
and disclose the prognosis of future LULC
change for 2020 and 2030. For sustainable
environmental development and wellbeing
of households, the current investigation can
be momentous. In addition, the results from
this study can inform decision making
process concerning the protection of
environment and it can serve as a baseline
for further researchers aiming at monitoring
LULCC and prediction of changes in areas
with similar LULC dynamics.
2. Materials and Methodology
2.1. Site Description
Congo Nile Ridge Region of Rwanda is
located between latitude 28°55ʹ30ʹʹS and
29°26ʹ00ʹʹS, longitude 10°33ʹE and 20°33ʹE
as shown in Figure 1, occupying an entire
area of 7991.32 square kilometers. The
region stretches eastwards to the
surroundings of Lake Kivu. The area under
study is landlocked and comprises nine
administrative districts namely, Nyaruguru,
Nyamagabe, Karongi, Rutsiro, Rubavu,
Nyabihu, Ngororero, Rusizi, and
Nyamasheke. The topography is hilly and
mountainous and has an altitude ranging
from 1700 meters to 4507 meters. The
highest peak is 4507 meters on mountain
Karisimbi (Byers, 1992; Mikova et al.,
2015). Rwanda has mountains that are
volcanic in nature in the northern frontier
and the Congo Nile Ridge Region of
Rwanda which extend over an unstable
mountainous area, while the central plateau
is dominated by scattered undulating hills
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
85
(Carr and Halsey, 2000; Van Damme et al.,
2013).
Figure 1. Locating the area of the study of Congo Nile Ridge Region of Rwanda
(CNRRR).
In regions of a high altitude, the annual
average temperatures range between 15°C
and18°C in the western and northwestern
mountains. The volcanic region has much
lower temperatures that can go below 0°C in
some places. The study area experiences a
rainfall climate of the tropics, with the
lowest rainfall of 1000 to 1200 mm/year
averagely. The highest average rainfall
rising up to 1800 mm/year, occurs in the
western mountain ranges of Rwanda, where
most of the rainfall runoff is generated. The
year is divided into four seasons; two are
wet (Mar-May and Sept-Dec) and the other
2 seasons are dry (Jan-Feb and Jun-Aug)
(Basnet and Vodacek, 2015; Karamage et
al., 2017; Nsabimana and Habimana, 2017;
Ntwali et al., 2016). Winds are generally
around 1‐3 m/s (Mikova et al., 2015).
2.2. Data Acquisition
2.1.1. Land Use/Cover
The use of satellite imagery has shown
immense potential in the study and analysis
of LULC in regional as well as global scales
(Basnet and Vodacek, 2015; Chen et al.,
2013). This investigation used the free
accessible data of Landsat TM/ETM+ of the
Congo Nile Ridge Region of Rwanda from
the Landsat chronicle. The developing
season is, by and large, more helpful for
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
86
mapping vegetation than lethargic periods.
The essential planting season in the area of
investigation is amid the March– May and
Sept-December wet seasons. Along these
lines, the essential Landsat data were those
acquired in June– August as the yields
developed. At the point when scenes were
not available amid the essential growing
time frame, data with insignificant overcast
cover were acquired from the second dry
season. Therefore, six satellite
images/scenes were downloaded from
Landsat data through United States
Geological Survey (USGS) Global
Visualization Viewer web site
(http://glovis.usgs.gov/) for conducting this
research. The area of the study stretches in
one path and two different rows of Landsat
image tile (path 173 and rows 61 and 62).
All data flow and information processes for
the methodology used in LULC dynamics
analysis and future prediction in the study
area are summarized in the Figure 2.
Figure 2. Methodology used of the study for LULC dynamics prediction.
To cover the whole study area, two satellite
scenes set for each year were downloaded,
and thus six cloud-free or minimum cloud
images (Landsat 5: TM 1990 and Landsat 7:
ETM + 2000 and 2010) in the dry season
(December – January) in order to minimize
seasonal influences (Table 1). The
coordinate system of the images is the
Universal Transverse Mercator (UTM),
Zone 35S, referenced to the World Geodetic
System (WGS) 1984 ellipsoid. Moreover, it
has a spatial resolution of 30m. The
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
87
characteristics of the selected images are
tabulated in Table 1. At times, it was
essential to wipe out clouds and shadows by
making masks and utilizing complementary
images.
Table 1. Specifications of images of Landsat data information used in Congo Nile Ridge
Region of Rwanda with a spatial resolution of 30 m.
Sn Satellite/ Sensor Acquisition Date Cloud
Cover
(%)
Path Row
1
2
Landsat 4/TM 1989/08/04
1989/08/04
4
2
173
173
61
62
1
2
Landsat 7/ETM+ 1999/12/06
2000/06/15
7
9
173
173
61
62
1
2
Landsat 5 / TM
Landsat 5/ TM
2010/08/22
2010/08/22
12
14
173
173
61
62
2.2.2. Driving Factors
Four driving factors, namely, distance from
main roads, distance from rivers, elevation,
aspect and slope were utilized for LULC
future change prognosis through Land
Change Modeler. Elevation, aspect and
slope were derived from Digital Elevation
Model (DEM) of 30m resolution acquired
from the Shuttle Radar Topography
Mission(SRTM) of the National
Aeronautics and Space Administration
(NASA) the radar mapping of Earth’s
topography (National Aeronautics and
Space Administration United States). The
distance from roads and distance from rivers
were acquired as shapefiles from DIVA-
GIS website (DIVA-GIS). The map of
distance from mains roads and rivers was
created using Euclidean distance embedded
in ArcMap-Spatial analysist tools through
the shapefile of the road network datasets
and shapefile of river respectively of the
area under investigation. The slope and
aspect were derived from STRM DEM
(Basnet and Vodacek, 2015;
Nikolakopoulos et al., 2006). All variables
were prepared in raster format with the same
projection and resolution (30*30 m of cell
size) as done for LULC classification.
2.3. Image Processing
Image preprocessing was achieved by
stacking imagery, band selection and
consortium. To fill the gaps and harmonize
the data, interpolation and extrapolation
techniques were applied (Vázquez-Quintero
et al., 2016). Landsat imageries of
1990,2000 and 2010 were radiometrically
adjusted using the top of atmosphere
process (Akinyemi, 2017). This process is
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
88
very important for imageries applied to
compare different periods of times in a year,
as in the case of LULC change with Land
Change Modeler (LCM) where earlier
LULC and later LULC are needed to
analyze the changes between the two
periods (1990-2000, 2000-2010 and 1990-
2010) (Iizuka et al., 2017; Vázquez-
Quintero et al., 2016).
Equation (1) and (2) were used for the
radiometric conversion of sensors TM of the
satellite Landsat (Iizuka et al., 2017).
Lλ = (( Lmaxλ − Lminλ )/(QCALmax (1)
−QCALmin)) ∗ (QCALmin − QCALmi ) + Lminλ
ρλ =π ∗ Lλ ∗ d2
cos θs ∗ USUNλ
(2)
in which, Lλ is the spectral radiance, ρλ
is top of the atmosphere, QCAL is the
calibrated and digital conversion number,
𝐿𝑚𝑖𝑛λ = to the spectral radiance scales to
QCALmin, 𝐿𝑚𝑎𝑥λ represents the
spectral scales radiance to QCALmax,
QCALmin is equivalent to the maximum
quantized calibrated pixel value, QCALmax
is the maximum quantized calibrated pixel
value, d stands for the distance from the sun
to earth, 𝑈𝑆𝑈𝑁λ indicates the minimum
solar exoatmospheric irradiance, also θs
represents the solar zenith angle. Lλ is in
units of watts per square meter per steradian
per micrometer (w/m2/sr/µm). (Xiong et al.,
2017). The false color composite was
presented in the red, green and blue,
respectively, was used to show differences
between diverse LULC categories (Chander
et al., 2007). Besides, after atmospherical
rectification, each image set containing 2
scenes of each was mosaicked using seam
line mosaicking to create a single scene of
each year. Finally, the single scene
generated for each year was clipped
according to the shapefile of the study area
as a raster in ArcMap.
2.4. Classification of Land Use/Cover
The modified Anderson Level I
classification Scheme(Churches et al., 2014;
Lui and Coomes, 2015) which is the
worldwide accepted and widely used LULC
classification scheme were used in this
research. This was done to achieve desired
LULC categories from satellite imageries of
1900, 2000 and 2010. In the context of
estimating LULCC over time and future
prediction of the area under investigation,
the first phase in Land Change Modeler of
IDRISI method is change analysis,
transition potential and change prediction.
In Land Change Modeler software, in order
to compare two maps of different periods of
time, it is mandatory for both imageries to
have the equal number of LULC categories,
same background, and similar projection
and coordinate system to meet the common
land type structure (Bhatta, 2010).
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
89
Referring to Table 2, the LULC maps of
Congo Nile Ridge Region of Rwanda for
1990, 2000, and 2010 were generated using
Anderson’s classification system level one
through supervised training classification
method (Anderson, 1976). Six LULC
categories were therefore identified,
including forestland, grassland, cropland,
settlement, wetland and water bodies. For
the current study, a supervised classification
approach and the maximum likelihood
classification algorithm (Mwangi et al.,
2017) were used as indicated in the Equation
numbered four and five. To generate the
desired categories from the Landsat images
we applied the maximum likelihood images
classification as it is indicated by equations
numbered three and four. Hence, the
classification based on maximum likelihood
is considering together with the covariances
and variances of the category signatures
when assigning each cell to one of the
categories representing in the signature file
(Friehat et al., 2015). As a rule, the
maximum likelihood assigns pixel to the
categories based on probability (Srivastava
et al., 2012). The probability density
function for a category (Wi) is derived by:
P(X/Wi)=1
((2𝜋)12)σi
exp (−1
2∗
(𝑋−µi)2
σi2 ) (3)
in which, P is density function probability,
Wi is the LULC category, X represents the
values of brightness, µi is the average of the
estimation of all values in LULC category
training category.
For n-dimensional multivariate ordinary
thickness equation is written as:
P(X/Wi)=𝟏
((𝟐𝝅)𝟏𝟐)|𝑽i|
𝟏𝟐
exp (−𝟏
𝟐∗ (𝑿 − 𝐌)𝑻 𝑽i
−𝟏 (𝑿 − 𝐌i)) (𝟒)
in which, Vi is covariance matrix
determinant, Vi-1 is the matrix of covariance
inverse, (X-Mi) T is the transpose of the
vector X-Mi, Mi is the mean vector, V is the
covariance of the matrix. Figure 3 is the
outcome of classification procedure,
depicting maps of LULC of the area of
investigation for the different periods under
consideration in this research
.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
90
Table 2. Classification schemes of LULC in Congo Nile Ridge Region of Rwanda.
LULC
Category
Descriptions
Forestland This is a developed plantation approximately with the elevation
greater than five meters, and it is greater or equal to 60 % ground
surface covered by trees and canopy is even with green foliage.
Forestland category contains natural montane forests and plantation
forests such as Pine as well as stands of trees remaining with
croplands.
Grassland This category of land contains primarily of shrubs, grasses, bushes and
scrub. It is mainly used as pasture grassland suitable for pasturage.
Cropland Cropland fields land area, cultivated lands used for mixed crops.
This land grows a variety of food crops and fiber with greater or equal
to 80 % of land cover in crop-producing fields and wetlands.
Built-up This type of land is made up of about 80 to 100 percent of
buildings of residential, non-residential and other infrastructures such
as roads, play-grounds without or with vegetations, and car parking.
These are raised by cooked bricks, uncooked bricks, concrete material
or mud. The buildings are surrounded by a small open space with
flowers, grasses and trees.
Wetland Mix of water with herbaceous and other vegetation; swamps
Water bodies Flowing water, Perennial water body, rivers, lakes, ponds, dam
and other water reservoirs with greater or equal 95 percent of water
cover.
In this classification scheme, six classes are
identified, namely, Forestland, Grassland,
Cropland, settlement, Wetland and Water
bodies.
2.5. Classification Accuracy Assessment
For accuracy assessment evaluation, the
overall accuracy and Kappa statistics were
calculated (Basnet and Vodacek, 2015; Yeh
and Qi, 2015). In total, 360 points samples
were utilized as a part of the precision
evaluation of the classified images of 1990,
2000 and 2010. Each of the six LULC
classes were utilized as stratum to create the
arbitrary sampling points. In all, sixty
arbitrary points were utilized for each land
category (forestland, grassland, cropland,
settlement, wetland and water bodies). A
Kappa statistic for the stratified random
sampling was also used using the equation
(5)(Foody, 2004; Mubea and Menz, 2014).
K =N ∑ (Xii) r
i−1 −∑ (Xi+)( X+i) r
i=1
N2−∑ (Xi+)( X+i) r
i=1
(5)
in which, r stands for the quantity of rows in
the matrix, Xii represents the quantity of
observations in row-i and column-i (the
diagonal elements), X+i and Xi+ represent
the marginal totals of row-r and column-i,
respectively, and N stands for observations
quantity. The accuracy assessment
summary of produced accuracy, user
accuracy, overall accuracy and the kappa
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
91
statistics of both imageries classified are
depicted in Table 3.
2.6. Land Use/Cover Prediction
The simulation can be done for the future
date; the MLP neural network was applied
to learn the status of transition with driving
potential variables of the LULCC. The
potential transition map resulting from MLP
neural network was used in combination
with Markov Chain (MC) for generating the
future date projection land use/cover. The
MC is a stochastic procedure model that
defines in what way likely one state is to
transform to another state, which stands to
transition probability matrix (Arsanjani et
al., 2013). In Markov Chain, two different
LULC maps at earlier and later land
user/cover should exist, and it is possibly to
compute the transition probability between
earlier and later lands (Zhang et al., 2011).
The prediction of the future LULC in LCM
is done in two very important phases, the
first phase is the transition potential sub-
model and the second one is the prediction
of change for a specific future date
(Aguejdad et al., 2017). The potential
change modeling is developed by applying
transition of sub-model grouped into
important variables affecting land use and
land cover change categories, in our study,
the MLP neural network sub-routine is used
(Aguejdad et al., 2017).
Future change prediction is therefore
performed by Markov Chain, after
determination of the prediction date
(Arsanjani et al., 2013; Zubair et al., 2017).
The validation is used for assessing the
quality of predicted output map. Markov
Chain transition probability matrix is
computed by the Equation (7) and
(8)(Stewart, 2009).
∑ Aij
𝑛
𝑖=1
= 1𝑖, 2,3, … … … 𝑛 (7)
A = (Aij) |
𝐴11 𝐴12 … 𝐴1𝑛
𝐴21 𝐴22 𝐴2𝑛
𝐴𝑛1 𝐴𝑛2 𝐴𝑛𝑛
| (8)
(8)
in which, Aij is the transition probability
from one land use and land cover to another
land use and land cover, n is the category of
land use and land cover of the area under
investigation, Aij is the value between 0 and
1. The accuracy assessment is accepted
when it is achieved within the vicinity
equaling to 80% (Eastman, 2009; Eastman,
2012).
LCM in TerrSet was used for predicting
LULCC of the area under investigation for
2020 and 2030. Applying the historical
maps of change at the earlier time
(LULC1990) and another land use/cover at
later time (LULC2000), with the
combinations of layers of distance from
roads, distance from rivers, slope, aspect,
and elevation, were used to project LULC
map of the year 2010 (LULC2010).
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
92
Accordingly, the map predicted of land use
and land cover of 2010 was validated by
relating it to the actual map of 2010
classified. Thereafter, with the positive
validation, the land use and land cover map
of 2000 and 2010, were then applied to
predict LULC in 2020, while the LULC of
1990 and LULC of 2010 were utilized to
model the future land use and land cover in
2030.
3. Results and Discussion
3.1. LULC Classification 1990,2000 and
2010
The consecutive equal time periods of 10
years between imageries of Landsat during
1990–2010 (1990, 2000 and 2010) were
adopted to synchronize with LULC
dynamics analysis. Over the entire study
area of 7991.35 square kilometers, as it is
shown in Figure 3 and Tables 2 and 3, six
lands use and land cover classes (forestland,
grassland, cropland, built-up, wetland and
water bodies) were classified using
supervised maximum likelihood for both
three years (1990, 2000 and 2010). The
overall accuracy for 1990,2000 and 2010
maps was 95.28 %, 96.67% and 94.72%,
respectively, whilst the Kappa statistics was
94.33%, 96.00% and 93.67%, respectively
(Table 3). The Kappa statistics (>80 %) for
the three maps of three different time frames
indicate a high degree of accuracy with
minimal error propagation in classification
method. Producer’s and user’s exactitudes
of specific categories were consistently
high, extending from 83.67% to 100% and
86 to 100%, respectively (Table 3).
Therefore, forestland (59.31 %), cropland
(23.55 %), grassland (4.35 %) and water
bodies (12.69 %) were the foremost
represented LULC classes in Congo Nile
Ridge Region of Rwanda in 1990, whereas
built-up (0.07%) and wetland (0.03%) were
the slightest represented in the area under
investigation of 799,135 hectares.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. P81-102
93
Figure 3. LULC Mapping of 1990, 2000 and 2010.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
94
Table 3. Summary of Landsat classification accuracy evaluation (Overall accuracy,
user accuracy and Kappa statistics) for 1990, 2000 and 2010.
Year LULC
Category
Producer
Accuracy
(%)
User
Accuracy
(%)
Overall
Accuracy
(%)
Kappa
(%)
1990
Forestland
Grassland
Cropland
Settlement
Wetland
Water bodies
100.00
91.67
96.67
93.33
90.00
100.00
98.36
94.83
86.57
98.25
98.18
96.77
95.28 94.33
2000
Forestland
Grassland
Cropland
Settlement
Wetland
Water bodies
100.00
95.00
100.00
93.33
91.67
100.00
100.00
98.28
92.31
96.55
98.21
95.24
96.67 96.00
2010
Forestland
Grassland
Cropland
Settlement
Wetland
Water bodies
100.00
90.00
100.00
96.67
83.67
100.00
98.36
93.10
89.55
96.67
96.08
95.24
94.72 93.67
3.2. Land Use/Cover Forecasting for 2010,2020 and 2030
The predicted land use /land cover of 2010
(Figure 4) was generated using land use and
land cover of 1990 and 2000, and after that
it was validated based on the actual LULC
of 2010 (Figure 4), in order to verify the
accuracy before continuing with the
simulation future land use and land cover for
2020 and 2030. Conferring to Pontius and
Schneider (2001), validation is playing a
significant role in the processing of a
modelling system of prediction. Validation
stipulate how the simulated land use and
land cover map is correlated to actual map
or the map of reference (Koranteng and
Zawila-Niedzwiecki, 2015).
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. P81-102
95
Figure 4. The actual land use and a land cover map of 2010 and the predicted land use
and land cover of 2010 map applying the land use and land cover maps of 1990 and
2000, both used for validation of simulated LULC map of 2010.
Therefore, the model was achieved with a
Kappa overall of 83.57% and a percentage
of correctness of 89.35 %. The accuracy
assessment results achieved in this
prognosis was satisfactory and showed a
great model with magnificent expectation
(Koranteng and Zawila-Niedzwiecki,
2015). Furthermore, there are few
inconstancies in simulated LULC 2010 for
built-up category. The concentration of
built-up category to the southeast of the
forecasted map of land use and land cover
2010 was compared to the actual LULC map
of 2010. However, applying the LULC of
2000 as an earlier period and LULC of 2010
as a later period, the land use/cover of 2020
was then simulated (Figure 6). The
prediction of land use and land cover in
2030 (Figure 6) was executed using the
earlier land use/cover of 2000 and the later
land use/cover of 2010.
Figure 5 displays the prognosis of 2020 and
2030 maps of land use and land cover. In
2020, forestland and cropland continue to be
the peak in land use and land cover category
with approximate rates amount of 333,559
hectares and 317,976 hectares, respectively.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
96
Cropland is the uppermost gaining, while
the built-up is slightly increasing. Then,
water bodies and wetland remain
unchangeable. As shown in Figures 5 and 6
as well as in Table 4, the forecasted land use
and land cover of 2030, depicted that the
forestland and grassland lands reduced the
extent of about 6480 hectares and 2550
hectares, respectively, whereas built-up and
cropland increased approximatively 1.12 %
and 0.01 %, respectively, while water bodies
and wetland remained constant.
Table 4. Projected Land Use and Land Cover for 2020 and 2030.
2020 2030 Change
LULC
Category
Extent
(%) Extent(ha) Extent (%) Extent(ha) Extent(ha)
Forestland 41.74 333559 40.93 327079 -6480
Grassland 5.23 41795 4.91 39245 -2550
Cropland 39.79 317976 40.91 326926 8950
Built-up 0.3 2397 0.31 2477 80
Wetland 0.01 80 0.01 80 0
Water
bodies 12.93 103328
12.93 103328
0
TOTAL 100 799135 100.00 799135
The trend of LULC from 1990 to 2030 as
depicted from Figure 7, extends across a
period of 40 years. Since 1990 to 2030, the
figure shows a significant decreasing and
increasing changes for forestland and
cropland respectively. The forestland is
highly losing; whereby it changed from
about 473,969 hectares in 1990 to
approximately 327,079 hectares in 2030 as
per LULC simulation.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. P81-102
97
Figure 5. Simulated land use/cover maps:2020 (using LULC of 2000 and LULC of
2010) and
2030 (using LULC of 1990 and LULC of 2010).
Figure 6. Predicted LULC for 2020,30 and LULC since 1990- 2010.
0.00
20.00
40.00
60.00
80.00
100.00
120.00
1990 2000 2010 2020 2030
Ch
ange
in %
Forestland Grassland Cropland
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
98
Figure 7. The trend of LULC of Congo Nile Ridge Region of Rwanda from 1990 -2030.
The simulation of the changes elaborated
that the cropland continues to increase so
that it will probably reach the forestland
(which used to be dominant from 1990-
2010) in 2030. Given that the forestland
loses and cropland increases, it is obvious
that forestland remains dominant but with a
high likelihood for cropland to become
dominant after 2030, if no adequate
measures will be put in place. In that case,
the cropland will be estimated to 326,926
hectares whereas forestland will be 326,926
hectares. For the year 2030, forestland is
higher 153 hectares than cropland
comparing to 285,786 (35.76 %) in 1990.
Additionally, the results showed that from
2020 to 2030, grassland will decrease from
41,795 hectares to 39,245 hectares
equivalent to a loss of 2,550 hectares.
Inversely, the built-up will increase about 80
hectares, while wetland and water bodies
will remain unchanged (Figures 6,7).
4. Conclusions
The main goals of this research emphasized
on prediction of LULC in Congo Nile Ridge
Region of Rwanda for 2020 and 2030 years.
Additionally, this study was implemented
using LCM where the future LULC
prediction, Markov chain embedded in Land
Change Modeler was applied, and Multi-
Layer Perceptron (MLP) Neural Network
sub-routine of LCM was used for transition
potential map based on driving factors
(distance from roads, distance from rivers,
slope, aspect and elevation) and previous
change trend. The outcomes of this study
confirmed ample proofs of transition of land
cover changes.
y = -5.433x + 65.329R² = 0.8272
y = 0.2002x + 4.4156R² = 0.1641
y = 5.096x + 17.602R² = 0.8641
y = -0.006x + 0.036R² = 0.75
y = 0.072x + 12.62R² = 0.7289
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
1990 2000 2010 2020 2030
Ch
ange
in %
Forestland Grassland CroplandBuilt-up Wetland Water bodiesLinear (Forestland ) Linear (Forestland ) Linear (Grassland )Linear (Cropland) Linear (Wetland) Linear (Wetland)
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
99
LULC changes were assessed by
increases/decreases in different land
categories in the future land use and land
cover prediction for 2020 and 2030 except
for two lands (water bodies and wetlands)
remaining constant. Furthermore, the
prediction for 2020 and 2030 of land use and
land cover revealed that forestland will
continue to lose as well as grassland to the
extent of 6,480 ha and 2,550 ha
correspondingly, whilst cropland and built-
up will increase around 8,950 ha and 80 ha
respectively, and both wetland and water
bodies will be remained constant.
From the findings of the current study, it
would be recommended that forestland in
Congo Nile Ridge Region of Rwanda
should be dedicated to the mechanisms of
protection and conservation as well legal
means. In addition, other managed and
effective regulations should be established
to curb the deforestation and destruction of
forests as to safeguard the forests in the
study area. Concurrently, some mechanisms
need to be put in place like planned
harvesting of trees, wood cutting for
firewood, built-up, and adoption of new
alternative sources of energy rather than
firewood. This would therefore reduce the
continuing stress and excessive
consumption of forests. Notwithstanding,
thorough enforcement of existing policies,
rules and regulations related to construction
measures needs to be enhanced. To achieve
this, the government of Rwanda should
embark on new approaches aiming to reduce
the mismanagement of the limited land
resources. These include: grouped and
planned settlements, construction in vertical
ways instead of current horizontal which is
consuming a lot of fertile lands which
should be used for agriculture. Accordingly,
the population growth should be controlled,
because the population increase exacerbates
pressure on land use and land cover.
Introduction of well-planned farmland
activities accommodate more than 80% of
the overall population in the country.
5. Acknowledgments
This study has been bolstered by the China–
East Central Africa Joint Research Center
Project of Chinese Academy of Sciences.
The authors express plentiful gratitude to
USGS and NASA team for providing
information and data used in this research.
The authors would like also to thank
Professor Xi Chen from Xinjiang Institute
of Ecology and Geography of Chinese
Academy of Sciences, for supervising,
motivating, improving the quality of the
study and his unreserved scientific support
in order to accomplish this paper. Finally,
the special thanks are additionally addressed
to the joint cooperation between the
University of Lay Adventists of Kigali
(UNILAK and Xinjiang Institute of Ecology
and Geography of Chinese Academy of
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
100
Sciences (XIEG-UCAS), which offered a
financial support for achieving the
Doctorate Studies (PhD) of which this study
is a fruit.
6. References Aguejdad, R., Houet, T., Hubert-Moy, L.
2017- Spatial Validation of Land Use Change Models Using Multiple Assessment Techniques: A Case Study of Transition Potential Models. Environmental Modeling & Assessment, 22(6): 591-606.
Akinyemi, F.O. 2017- Land change in the central Albertine rift: Insights from analysis and mapping of land use-land cover change in north-western Rwanda. Applied Geography, 87: 127-138.
Al-shalabi, M., Billa, L., Pradhan, B., Mansor, S., Al-Sharif, A.A. 2013- Modelling urban growth evolution and land-use changes using GIS based cellular automata and SLEUTH models: the case of Sana’a metropolitan city, Yemen. Environmental earth sciences, 70(1): 425-437.
Anderson, J.R. 1976- A land use and land cover classification system for use with remote sensor data, 964. US Government Printing Office.
Arsanjani, J.J. 2011- Dynamic land use/cover change modelling: Geosimulation and multiagent-based modelling. Springer Science & Business Media.
Arsanjani, J.J., Helbich, M., Kainz, W., Boloorani, A.D. 2013- Integration of logistic regression, Markov chain and cellular automata models to simulate urban expansion. International Journal of Applied Earth Observation and Geoinformation, 21: 265-275.
Basnet, B., Vodacek, A. 2015- Tracking land use/land cover dynamics in cloud prone areas using moderate resolution satellite data: A case study in Central Africa. Remote Sensing, 7(6): 6683-6709.
Berakhi, R.O. 2013- Implication of Human Activities on Land Use Land Cover Dynamics in Kagera Catchment, East Africa, Southern Illinois University Carbondale.
Bhatta, B. 2010- Analysis of urban growth and sprawl from remote sensing data. Springer Science & Business Media.
Brink, A.B., Eva, H.D. 2011- The potential use of high-resolution Landsat satellite data for detecting land-cover change in the Greater Horn of Africa. International journal of remote sensing, 32(21): 5981-5995.
Byers, A.C. 1992- Soil loss and sediment transport during the storms and landslides of May 1988 in
Ruhengeri prefecture, Rwanda. Natural hazards, 5(3): 279-292.
Carr, R.H., Halsey, A.H. 2000- Land of a thousand hills: my life in Rwanda. Penguin.
Chander, G., Markham, B.L., Barsi, J.A. 2007- Revised Landsat-5 thematic mapper radiometric calibration. IEEE Geoscience and remote sensing letters, 4(3): 490-494.
Chen, X. et al. 2013- Changes in land use/land cover and ecosystem services in Central Asia during 1990–2009. Current Opinion in Environmental Sustainability, 5(1): 116-127.
Churches, C.E., Wampler, P.J., Sun, W., Smith, A.J. 2014- Evaluation of forest cover estimates for Haiti using supervised classification of Landsat data. International Journal of Applied Earth Observation and Geoinformation, 30: 203-216.
Di Gregorio, A. 2005- Land cover classification system: classification concepts and user manual: LCCS. Food & Agriculture Org.
DIVA-GIS Computer Program for Mapping and Geographic Data Analyis (DIVA-GIS). Available online at : http: //www.diva-gis.org (accessed on 10 March 2017).
Douglas, I. 2017- Flooding in African cities, scales of causes, teleconnections, risks, vulnerability and impacts. International Journal of Disaster Risk Reduction, 26: 34-42.
Eastman, J. 2009- IDRISI taiga, guide to GIS and remote processing (pp. 234–256). Worcester: Clark University.
Eastman, R. 2012- IDRISI Selva ,Version 17.0 .Computer Software Clark Labs, Clark University, Worcester, MA . Available online at: http: //www. clarklabs.org ( accessed on 15 May 2017).
Foody, G.M. 2004- Thematic map comparison. Photogrammetric Engineering & Remote Sensing, 70(5): 627-633.
Friehat, T., Mulugeta, G., Gala, T. 2015- Modeling urban sprawls in Northeastern Illinois. J. Geosci. Geomat, 3(5): 133-141.
Gillet, P. et al. 2016- What are the impacts of deforestation on the harvest of non-timber forest products in Central Africa? Forests, 7(5): 106.
Giri, C.P. 2012- Remote sensing of land use and land cover: principles and applications. CRC Press.
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
101
Goudie, A.S. 2013- The human impact on the natural environment: past, present, and future. John Wiley & Sons.
Han, H., Yang, C., Song, J. 2015- Scenario simulation and the prediction of land use and land cover change in Beijing, China. Sustainability, 7(4): 4260-4279.
Houet, T., Hubert-Moy, L. 2006- Modeling and projecting land-use and land-cover changes with Cellular Automaton in considering landscape trajectories. EARSeL eProceedings, 5(1): 63-76.
Iizuka, K. et al. 2017- Modeling Future Urban Sprawl and Landscape Change in the Laguna de Bay Area, Philippines. Land, 6(2): 26.
Kamusoko, C., Aniya, M., Adi, B., 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. Applied Geography, 29(3): 435-447.
Karamage, F. et al. 2017- Modeling Rainfall-Runoff Response to Land Use and Land Cover Change in Rwanda (1990–2016). Water, 9(2): 147.
Koranteng, A., Zawila-Niedzwiecki, T. 2015- Modelling forest loss and other land use change dynamics in Ashanti Region of Ghana. Folia Forestalia Polonica, 57(2): 96-111.
Kumar, S., Radhakrishnan, N., Mathew, S. 2014- Land use change modelling using a Markov model and remote sensing. Geomatics, Natural Hazards and Risk, 5(2): 145-156.
Lambin, E.F. et al. 2003- The causes of land-use and land-cover change: moving beyond the myths. Global environmental change, 11(4).
Lemarchand, G.A., Tash, A. 2015- Mapping research and innovation in the Republic of Rwanda, 4. UNESCO Publishing.
Lui, G.V., Coomes, D.A. 2015- A comparison of novel optical remote sensing-based technologies for forest-cover/change monitoring. Remote Sensing, 7(3): 2781-2807.
Mikova, K., Makupa, E., Kayumba, J. 2015- Effect of Climate Change on Crop Production in Rwanda. Earth, 4(3): 120-128.
Mishra, V.N., Rai, P.K., 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.
Mubea, K., Menz, G. 2014- Spatial Effects of Varying Model Coefficients in Urban Growth Modeling in Nairobi,
Kenya. Journal of Geographic Information System, 6(06): 636.
Mwangi, H.M. et al. 2017- Characterizing the Intensity and Dynamics of Land-Use Change in the Mara River Basin, East Africa. Forests, 9(1): 8.
National Aeronautics and Space Administration United States Advanced Spaceborne Thermal Emission and Reflexion Radiometer (ASTER) Global Digital Elevation Model. Available online at https: //search.earthdata.nasa.gov (accessed on 5 February 2017).
Nikolakopoulos, K.G., Kamaratakis, E.K., Chrysoulakis, N. 2006- SRTM vs ASTER elevation products. Comparison for two regions in Crete, Greece. International Journal of remote sensing, 27(21): 4819-4838.
NISR-Ministry of Finance and Economic Planning 2012- Fourth Population and Housing Census. Available online: http://www.lmis.gov.rw/scripts/publication/
reports/Fourth%20Rwanda%20Population%20and%20Housing%20Census_Housing.pdf (accessed on 15 November 2017).
Nsabimana, A., Habimana, O. 2017- Asymmetric effects of rainfall on food crop prices: evidence from Rwanda.
Ntwali, D., Ogwang, B.A., Ongoma, V. 2016- The impacts of topography on spatial and temporal rainfall distribution over rwanda based on wrf model. Atmospheric and Climate Sciences, 6(02): 145.
Nyandwi, E., Mukashema, A. 2011- Excessive deforestation of Gishwati Mountainous forest & biodiversity changes. Participatory Geographic Information Systems (P-GIS) for natural resource management and food security in Africa. Available on http://www. crdi. ca/Documents/ICT4D_article_forests_nyandwi_EN. pd f.
Pontius, R.G., Schneider, L.C. 2001- Land-cover change model validation by an ROC method for the Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems & Environment, 85(1): 239-248.
Rattan, L., Rattan, L. 2006- Encyclopedia of Soil Science Vol-1. CRC Press Taylor and Francis Group.
REMA (Rwandan Environmental Management Authority) 2015- Statement of the Environment and Outlook Report 2015, Green agriculture with Resource efficient, low carbon and climate resilent practices . Available online at : https ://www.nmbu.no/sites
The 5th ICEED, KIGALI-RWANDA-UNILAK AUGUST, 2018 EAJST Special Vol.8 Issue2, 2018 by Enan N. et al P81-102
102
/default/files/pdfattachments/state_of_environment_and_outlook _report_2015.pdf (accessed on 10 October 2017). Roose, E., Ndayizigiye, F. 1997-
Agroforestry, water and soil fertility management to fight erosion in tropical mountains of Rwanda. Soil Technology, 11(1): 109-119.
Rwanga, S.S., Ndambuki, J. 2017- Accuracy Assessment of Land Use/Land Cover Classification Using Remote Sensing and GIS. International Journal of Geosciences, 8(04): 611.
Shrestha, S., Dwivedi, P. 2017- Projecting land use changes by integrating site suitability analysis with historic land use change dynamics in the context of increasing demand for wood pellets in the Southern United States. Forests, 8(10): 381.
Srivastava, P.K., Han, D., Rico-Ramirez, M.A., Bray, M., Islam, T. 2012- Selection of classification techniques for land use/land cover change investigation. Advances in Space Research, 50(9): 1250-1265.
Stewart, W.J. 2009- Probability, Markov chains, queues, and simulation: the mathematical basis of performance modeling. Princeton University Press.
UNEP 2016- Rivers Partners , Applying Ecosystem-based Disaster Risk Reduction (ECO-DRR) in Integrated Water Resouce Management ( IWRM) in the Lukaya Basin, Democratic Republic of the Congo. Available online https ://web.unep.org/disastersandconflicts/river-partners-applying-ecosystem-based-disaster-risk-reduction-eco-drr-integrated-water-resource (accessed on 10 December 2017).
Van Damme, J., Ansoms, A., Baret, P.V. 2013- Agricultural innovation from above and from below: Confrontation and integration on Rwanda's Hills. African Affairs, 113(450): 108-127.
Vázquez-Quintero, G. et al. 2016- Detection and projection of forest changes by using the Markov Chain Model and
cellular automata. Sustainability, 8(3): 236.
Verburg, P.H., Chen, Y., Soepboer, W., Veldkamp, A. 2000- GIS-based modeling of human-environment interactions for natural resource management.
Verburg, P.H. et al. 2002- Modeling the spatial dynamics of regional land use: the CLUE-S model. Environmental management, 30(3): 391-405.
Wu, S., Mickley, L.J., Kaplan, J., Jacob, D.J. 2012- Impacts of changes in land use and land cover on atmospheric chemistry and air quality over the 21st century. Atmospheric Chemistry and Physics, 12(3): 1597-1609.
Xiong, J. et al. 2017- Nominal 30-m Cropland Extent Map of Continental Africa by Integrating Pixel-Based and Object-Based Algorithms Using Sentinel-2 and Landsat-8 Data on Google Earth Engine. Remote Sensing, 9(10): 1065.
Yeh, A.G.-O., Qi, Z. 2015- Short-interval monitoring of land use and land cover change using a time series of RADARSAT-2 polarimetric SAR images, Space-Time Integration in Geography and GIScience. Springer, pp. 353-371.
Yirsaw, E., Wu, W., Shi, X., Temesgen, H., 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.
Zhang, R. et al. 2011- Using Markov chains to analyze changes in wetland trends in arid Yinchuan Plain, China. Mathematical and Computer Modelling, 54(3): 924-930.
Zubair, O.A., Ji, W., Weilert, T.E. 2017- Modeling the Impact of Urban Landscape Change on Urban Wetlands Using Similarity Weighted Instance-Based Machine Learning and Markov Model. Sustainability, 9(12): 2223.