Page 717 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

Modelling and Predicting Multiple Land Use/Land Cover Changes Using Logical
Regression Method and Markov Chain with Choice Approach Among Existing Options

3
2*
1
Fatemeh Ghaffarpour , Parham Pahlavani , Behnaz Bigdeli

1 GIS MSc Student at School of Surveying and Geospatial Engineering, College of Engineering, University of
Tehran, Tehran, Iran
2 Assistant Prof. at School of Surveying and Geospatial Engineering, College of Engineering, University of
Tehran, Tehran, Iran
2 Assistant Prof. at School of Civil Engineering, Shahrood University of Technology, Shahrood, Iran

* pahlavani@ut.ac.ir

Abstract: The study of changes and the rate of destruction in previous years, as well as the possibility of predicting these
changes in recent years, plays an important role in optimal planning and management and limiting unusual changes in
the future. The study was conducted to identify land changes and land cover (1996-2006) and to predict future changes
(2026) using satellite imagery in the Massachusetts region of the United States using a logical regression method. To
increase the efficiency and optimization of the model, the efficient features have been selected using Cramer's V test. The
study area is classified into 18 classes, which as a result of the analysis showed a decreasing trend of 3.84% by grasslands,
while bare lands have increased by 9.78% and unconsolidated shore by 9.5%; as well as other land uses have changed
slightly between 1996 and 2006. Accordingly, Land-use/Land-cover changes have been predicted using the logical
regression method and Markov chain model by selecting from the available options using the Cramer's V test for 2026.
The accuracy of this method using the 2016 data had an overall accuracy of 84.21% and a Kappa index of 0.7150. The
results of this study showed rapid changes in land use for future years. The conversion of forest land into other lands,
especially grassland and pasture, was the most important change in land cover in the future. Therefore, planning for the
protection and restoration of the forest should be a key plan for decision-makers in the study area.
Keywords: Land-use Change Prediction, Logistic Regression, Cramer’s V, Markov Chain

1. Introduction
Changes in land use and subsequent changes in land cover have been ongoing for a long time and will continue in the
future [1], [2]. Monitoring the process of land-use change and land cover is very important because it provides a level of
knowledge about the local potential of a particular place as well as issues related to the global environment [3], [4]. Understanding
land use change over time is one of the most important parameters in land planning and management [4]. In addition to predictive
modelling, change detection studies play an important role in the effective and efficient management of natural resources and
the environment [5]. These changes are influenced by the interaction of human activities such as population growth, political
and economic conditions, and environmental factors such as climate, soil, and topography [4], [6].
Over the years, various methods and models have been proposed to model and predict land use and land use changes.
Such models provide flexible and innovative environments that simplify a complex network of land use and land use
developments, and ultimately play a role in informed decision-making and effective land management [5], [7], [8]. One of these
methods is logistic regression [9] and Markov chain [10], used to predict the future and simulate changes in land cover. In recent
years, significant researches have been conducted on land use/land cover changes in urban and non-urban areas. In these studies,
several independent variables have been considered to improve the efficiency of the models, such as topographic maps, height,
slope, distance from the road, distance from agricultural fields, etc. [5], [11]–[13]. In some cases, the number of these variables
has increased, but this increase in the number of variables ensures greater model performance. Therefore, for the model to be

1