Sentinel-2 satellite imagery for retrieving and mapping soil properties using machine learning
Paper ID : 1076-ICRSSSA-FULL (R1)
Authors
Mohamed E. S. Amin *1, Mahmoud Ali Abdelfattah2, Mohsen Nabil1, Elsayed Said Mohamed3, Abdelaziz Belal4, Ali Gaber Mahmoud5, Ihab Samir Mohamed Abdellatif1, Sayed A. Ahmed6
1​National Authority for Remote Sensing and Space Sciences (NARSS)
2Faculty of Agriculture-Fayoum University
3National Authority for Remote Sensing and Space Sciences (NARSS)
4​​National Authority for Remote Sensing and Space Sciences (NARSS)
5Faculty of Agriculture- Fayoum University
6National Authority for Remote Sensing and Space Sciences
Abstract
One of the most critical techniques to forecast agriculture production is to re-trieve soil properties, aiming to decrease risk in decision-making since these properties impact the product amount and quality. Using Sentinel 2 data and machine learning techniques, this work aimed to construct models for soil pa-rameter retrieval over irrigated pivot field in Southwest Ismailia, Egypt. To achieve that, two machine learning (ML) models, Random Forest (RF) and Linear Regression (LR) were developed and statistically verified for seven soil properties retrieval, namely clay, silt, bulk density (BD), calcium carbonates (CaCo3), available nitrogen (N), available potassium (K), and electrical conduc-tivity (ECe). Sentinel-2 bands with a spatial resolution of 10 m coupled with measured soil data were employed to build the models. The results showed that LR model is the best for K retrieval, followed by the RF with R² values of 0.63 and 0.60, and mean square error (MSE) of 2.33 and 3.21, respectively. While using the LR approach, BD and CaCO3 revealed improved results with R2 of 0.49 and 0.57 and MSE of 0.006 and 0.1, respectively. These findings demonstrate the capability of developed machine learning algorithms for re-trieving soil characteristics. They may be used as quick decision-making tools for retrieving soil attributes in other areas with similar conditions.
Keywords
Sentinel-2; Machine learning; Soil properties
Status: Accepted (Oral Presentation)