TY - JOUR
T1 - Soil salinity prediction using Machine Learning and Sentinel – 2 Remote Sensing Data in Hyper – Arid areas
AU - Kaplan, Gordana
AU - Gašparović, Mateo
AU - Alqasemi, Abduldaem S.
AU - Aldhaheri, Alya
AU - Abuelgasim, Abdelgadir
AU - Ibrahim, Majed
N1 - Funding Information:
This article is based on COST Action CA19123—Protection, Resilience, Rehabilitation of damaged environment, supported by COST ( European Cooperation in Science and Technology ) https://www.cost-phoenix.eu/ (accessed on December 1, 2022). The researchers thank the College of Graduate Studies at United Arab Emirates University for funding this research.
Funding Information:
In previous research, the authors have dealt with mapping soil salinity in the surrounding of rivers, lakes, and other water bodies (Wang et al., 2019; Hoa et al., 2019) and semi-arid areas (Abuelgasim and Ammad 2019; Akça et al., 2020). Thus, Elhag (2016), investigated soil salinity in an arid region with agricultural fields (Wadi Al Dawasir region), using 150 soil samples, where several salinity indices were compared against Electrical Conductivity (EC) data. A more minor investigation was done in an arid-area oasis in China, with 41 soil samples (Jiang, H., & Shu, H. 2019). The modeling of soil salinity was generally done using vegetation indices and linear regression. An investigation of modeling salinity in a hyper-arid area (desert) was done in the Unated Arab Emirates (UAE) using 80 soil samples (Alqasemi et al., 2021). However, the investigation was done using vegetation indices rather than salinity indices. Wang et al. (2021) used ML algorithms such as Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) for soil salinity mapping in arid areas. For this, they used 160 soil samples across the highway in the study area due to accessibility. The soil salinity modeling was also investigated in an agricultural area, a sugar cane irrigation farm (Asfaw et al., 2018). Yu et al. (2018) used 155 soil samples collected near the accessible roads in the study area for salinity modeling using Landsat – 8. Using vegetation and salinity indices derived from Sentinel – 2 data, Wang et al. (2019) used 64 soil samples for salinity modeling with ML algorithms in arid areas. The samples were collected along the accessible roads in the study area. Peng et al. (2019) used the same sampling technique in arid area with 225 soil samples. Using nonlinear regression models, Sentinel – 2 imagery has also been used for soil salinity estimation over dried lake bed (Farahmand and Sadeghi, 2020). Even though a high accuracy has been achieved, it should be noted that the numbers of samples were very limited.This article is based on COST Action CA19123—Protection, Resilience, Rehabilitation of damaged environment, supported by COST (European Cooperation in Science and Technology) https://www.cost-phoenix.eu/(accessed on December 1, 2022). The researchers thank the College of Graduate Studies at United Arab Emirates University for funding this research.
Publisher Copyright:
© 2023
PY - 2023/6
Y1 - 2023/6
N2 - We are experiencing a considerable increase in soil salinity as a result of the influence of climate change or environmental contamination produced by excessive industry and agriculture. To be able to cope with this issue, reliable and up-to-date soil salinity measurements are required. The use of remote sensing data allows for faster and more efficient soil salinity mapping. This paper investigates several Machine Learning approaches and modeling methodologies for predicting soil salinity in hyper-arid environments using Sentinel-2 satellite imagery. Thus, 393 soil samples collected and used for modeling and testing in the study area, United Arab Emirates. Also, the paper benefits from open-source data and programs, such as Google Earth Engine and Weka. Different modeling strategies have been applied over the data. The results of the modeling show a strong correlation (0.84) with the test results. This study also shows interesting findings that will be examined further in future studies at other sites. As machine learning methods are evolving on a daily basis, new approaches needs to be considered in future studies for the demands of more precise modeling and mapping of soil salinity.
AB - We are experiencing a considerable increase in soil salinity as a result of the influence of climate change or environmental contamination produced by excessive industry and agriculture. To be able to cope with this issue, reliable and up-to-date soil salinity measurements are required. The use of remote sensing data allows for faster and more efficient soil salinity mapping. This paper investigates several Machine Learning approaches and modeling methodologies for predicting soil salinity in hyper-arid environments using Sentinel-2 satellite imagery. Thus, 393 soil samples collected and used for modeling and testing in the study area, United Arab Emirates. Also, the paper benefits from open-source data and programs, such as Google Earth Engine and Weka. Different modeling strategies have been applied over the data. The results of the modeling show a strong correlation (0.84) with the test results. This study also shows interesting findings that will be examined further in future studies at other sites. As machine learning methods are evolving on a daily basis, new approaches needs to be considered in future studies for the demands of more precise modeling and mapping of soil salinity.
KW - Google earth engine
KW - Machine learning
KW - Modeling
KW - Remote sensing
KW - Sentinel-2
KW - Soil salinity
UR - http://www.scopus.com/inward/record.url?scp=85152133008&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85152133008&partnerID=8YFLogxK
U2 - 10.1016/j.pce.2023.103400
DO - 10.1016/j.pce.2023.103400
M3 - Article
AN - SCOPUS:85152133008
SN - 1474-7065
VL - 130
JO - Physics and Chemistry of the Earth
JF - Physics and Chemistry of the Earth
M1 - 103400
ER -
