Spatiotemporal Boundary Detection in the Sahara Desert Using Machine Learning Algorithms

• Activity Number: 431 - Contributed Poster Presentations: Section on Statistics and the Environment
• Type: Contributed
• Date/Time: Wednesday, August 10, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Stephen Matthew Tivenan* and Indranil Sahoo
• Companies: VCU and Virginia Commonwealth University
• Keywords: Spatio-Temporal ; Boundary Detection; Sahara Desert; Machine Learning Algorithms; Koppen-Trewartha

Abstract

The Koppen-Trewartha classification for dry climates is based on annual precipitation, annual winter precipitation and annual temperature. Koppen-Trewartha's simple classification allows for the ability to categorize a location as an arid or semi-arid location from year to year. Although the discreteness of the classification allows for easy visualization and quick computations, we transformed Koppen-Trewartha's classification onto a continuous spectrum to outline spatio-temporal trends of the Sahara Desert. The goal of this project is to do a boundary detection analysis of the Sahara Desert and examine the expansion of the arid and semi-arid regions over multiple consecutive years. Growth in the desert’s boundary or an increase of values from the spectrum would indicate desertification (that is, growth of the desert or increase in its intensity). Machine learning techniques such as k-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), and artificial neural networks (ANN) were used to detect boundaries across our region of study and the results were compared to the original Koppen-Trewartha’s classification. The region was examined annually from 1960-1989.