Carta ex Machina: Testing object-based machine learning and unsupervised classification in land use change detection mapping in the semi-arid governorate of Sidi Bouzid, Tunisia
(2021) In Student thesis series INES NGEK01 20211Dept of Physical Geography and Ecosystem Science
- Abstract
- Sidi Bouzid, Tunisia is an inland governorate in Tunisia that has undergone a rapid agricultural and urban development since the Tunisian independence in 1952 from being a
rural and largely nomadic region into a hub of irrigated agriculture. In 2010 Mohamed
Bouazizi sparked the Tunisian revolution by lighting himself on fire int he city of Sidi
Bouzid, with some blaming the inequality and water scarcity created by this rapid expansion in the irrigation farming as an important cause (Bayat, 2017; Malka, 2018). Sweden
has aided in this development from 1972-1992, and as part of an evaluation project by
Mårtensson et al (2019) a change detection analysis of the area around the Jabal al Kbar
mountain was requested.
In this thesis two... (More) - Sidi Bouzid, Tunisia is an inland governorate in Tunisia that has undergone a rapid agricultural and urban development since the Tunisian independence in 1952 from being a
rural and largely nomadic region into a hub of irrigated agriculture. In 2010 Mohamed
Bouazizi sparked the Tunisian revolution by lighting himself on fire int he city of Sidi
Bouzid, with some blaming the inequality and water scarcity created by this rapid expansion in the irrigation farming as an important cause (Bayat, 2017; Malka, 2018). Sweden
has aided in this development from 1972-1992, and as part of an evaluation project by
Mårtensson et al (2019) a change detection analysis of the area around the Jabal al Kbar
mountain was requested.
In this thesis two different object based classification methods (supervised random forest
on a segmented image and unsupervised ISODATA on a segmented image) were tested to
map and detect land use changes in a semi-arid environment in 4 areas around the Jabal
al Kbar over the years 1972 to 2021 based on 30m and 60m LANDSAT imagery in 10-year
intervals. Random Forest proved to have the highest accuracy of the two methods tested,
with a total accuracy of 72% and Cohen’s κ of 0.57 in a 60m resolution scene and 73% and
κ of 0.54 for a 30m resolution scene. ISODATA was not deemed adequate as it produced
too few classes with a standard number of iterations to be useful. A map produced with
a low number of iterations had an accuracy of only 51% and κ of 0.13. The resolution
was a limiting factor as it made identification of features difficult in the sampling process.
Nevertheless, the historic development of the study area from the literature in terms of
irrigation development and urbanization could be detected using Random Forest. Most
notably a rapid development in irrigation farming from 2004 to today in the area Hichria
was detected as well as decrease in the same period of irrigated area in the Sidi Bouzid
depression. for all 4 areas, starting trends observed from 2004 to to 2011 continued into
2021, thus not showing any noticeable effect of the revolution in the development. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9060595
- author
- Havnsgaard Paludan, Kristian Emil LU
- supervisor
- organization
- course
- NGEK01 20211
- year
- 2021
- type
- M2 - Bachelor Degree
- subject
- keywords
- Change detection, Land use mapping, LANDSAT MSS, LANDSAT TM, GEOBIA, Random Forest classification, ISODATA cluster classification, Object-based classification, Semi-arid agriculture, Irrigation mapping
- publication/series
- Student thesis series INES
- report number
- 545
- language
- English
- additional info
- Report no: 545 B
- id
- 9060595
- date added to LUP
- 2021-07-06 12:56:46
- date last changed
- 2023-03-08 13:25:11
@misc{9060595, abstract = {{Sidi Bouzid, Tunisia is an inland governorate in Tunisia that has undergone a rapid agricultural and urban development since the Tunisian independence in 1952 from being a rural and largely nomadic region into a hub of irrigated agriculture. In 2010 Mohamed Bouazizi sparked the Tunisian revolution by lighting himself on fire int he city of Sidi Bouzid, with some blaming the inequality and water scarcity created by this rapid expansion in the irrigation farming as an important cause (Bayat, 2017; Malka, 2018). Sweden has aided in this development from 1972-1992, and as part of an evaluation project by Mårtensson et al (2019) a change detection analysis of the area around the Jabal al Kbar mountain was requested. In this thesis two different object based classification methods (supervised random forest on a segmented image and unsupervised ISODATA on a segmented image) were tested to map and detect land use changes in a semi-arid environment in 4 areas around the Jabal al Kbar over the years 1972 to 2021 based on 30m and 60m LANDSAT imagery in 10-year intervals. Random Forest proved to have the highest accuracy of the two methods tested, with a total accuracy of 72% and Cohen’s κ of 0.57 in a 60m resolution scene and 73% and κ of 0.54 for a 30m resolution scene. ISODATA was not deemed adequate as it produced too few classes with a standard number of iterations to be useful. A map produced with a low number of iterations had an accuracy of only 51% and κ of 0.13. The resolution was a limiting factor as it made identification of features difficult in the sampling process. Nevertheless, the historic development of the study area from the literature in terms of irrigation development and urbanization could be detected using Random Forest. Most notably a rapid development in irrigation farming from 2004 to today in the area Hichria was detected as well as decrease in the same period of irrigated area in the Sidi Bouzid depression. for all 4 areas, starting trends observed from 2004 to to 2011 continued into 2021, thus not showing any noticeable effect of the revolution in the development.}}, author = {{Havnsgaard Paludan, Kristian Emil}}, language = {{eng}}, note = {{Student Paper}}, series = {{Student thesis series INES}}, title = {{Carta ex Machina: Testing object-based machine learning and unsupervised classification in land use change detection mapping in the semi-arid governorate of Sidi Bouzid, Tunisia}}, year = {{2021}}, }