Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Turquie → High-resolution soil salinity mapping using machine learning based regression and classification methods

İstanbul Teknik Üniversitesi (2021)

High-resolution soil salinity mapping using machine learning based regression and classification methods

YILDIRIM Aylin

Titre : High-resolution soil salinity mapping using machine learning based regression and classification methods

Makine öğrenimi tabanlı regresyon ve sınıflandırma yöntemlerini kullanarak yüksek çözünürlüklü toprak tuzluluğu haritalaması

Auteur : YILDIRIM Aylin

Université de soutenance : İstanbul Teknik Üniversitesi

Grade : M.Sc. THESIS 2021

Résumé
In the 21st century, humankind faces global environmental problems, which lead to the concept of the sustainable usage of earth resources. In light of this information, soil salinization is one of the main environmental issues that pose a high threat to soil fertility, especially in arid and semi-arid regions of the world. In these regions, lack of precipitation and extreme evaporation rate increases salt accumulation on the surface soil. Rapid population growth raises demands for producing more agricultural products and increasing the number of livestock industries, and due to this fact, lands containing healthy soil are required. Soil is an essential natural resource that provides several benefits to the ecosystem, and it supports the regulation of the biogeochemical and hydrological cycles of the Earth. Within the last three decades, scientists considered the importance of detecting and monitoring soil salinity in order to enhance soil conservation and sustainable development of agricultural lands. The development of remote sensing techniques, machine learning algorithms, and modeling techniques supports researchers for temporal and relevant monitoring of salt-affected lands. Enhancement in quality and accessibility of satellite data assists in rapid and accurate soil salinity mapping in large-scale areas. Iran’s Lake Urmia, which is the largest lake in the Middle East can be examined as one of those salt-affected places. Hence, this lake is selected to evaluate the spatial distribution of soil salinity levels. Soil samples were collected during the field survey in the dry season in 2016. Since the study area consists of the natural global reserves and agricultural fields, tracking the damaging factors for these types of regions has become a crucial issue. This study aims to compare machine learning-based regression and classification methods for mapping the high-resolution soil salinity using Classification and Regression Trees (CART), Random Forest (RF), Support Vector Machine (SVM) algorithms. For this purpose, high-resolution multispectral satellite images SPOT-7 (Airbus), PlanetScope (Planet), and Sentinel 2A (ESA) were selected with the specific date and time stamps that will serve to minimize the temporal difference with the ground observation date and times. Not only the satellite imagery but also the ground samples were processed in Google’s cloud-based geospatial analysis platform, called Google Earth Engine (GEE).

Présentation

Version intégrale

Page publiée le 6 janvier 2023