University of Twente Student Theses


Lodging detection in wheat: a multi-sensor approach using Sentinel-1 and Sentinel-2

Ashenafi, Shawl Mengistu (2021) Lodging detection in wheat: a multi-sensor approach using Sentinel-1 and Sentinel-2.

Link to full-text:
(only accessible for UT students and staff)
Abstract:Lodging is the permanent dislocation of the crop stem and root from its original position, and it affects grain yield and quality. Lodging monitoring is essential for reducing yield loss, avoiding knock-on effects, and maintaining grain quality. Remote sensing (RS) based lodging monitoring would help to acquire precise and continuous spatiotemporal data; it could help farmers increase their productivity and ensure sustainable agricultural production. In this study, the capability of data from Sentinel-1, Sentinel-2 and their combination were explored for wheat lodging detection and classification using a Random Forest classifier. Backscatter (σ°) and spectral reflectance of individual bands were extracted from the Sentinel-1 and Sentinel-2 time series data, respectively. Then the temporal backsccatter and spectral behaviour of different wheat lodging score (LS) groups (healthy (He), moderately lodged (ML), and severely lodged (SL)) were explored at different wheat growth stages via statistical analysis. A Random Forest classifier was used to classify the Sentinel-1, Sentinel-2, and their combination data into different LS classes. The statistical analysis of the Sentinel-1 backscatter and Sentinel-2 spectral data identified relationships with the field based-LS. The combination of Sentinel-1 and Sentinel-2 data based Random Forest model provided a better (83%) classification accuracy than Sentinel-1 (79%) and Sentinel-2 (80%) alone. The result implies that both Sentinel-1 and Sentinel-2 datasets provide complementary information for the model. Cross-polarised backscatter (σVH°) was the most important variable in the Random Forest classification of Sentinel-1 data. The red-edge-1 (RE-1) spectral band was the most important variable in the Random Forest classification using Sentinel-2 data. Furthermore, σVH°, short wave infrared-1 (SWIR-1), and RE-1 were the top three most important variables in the Random Forest classification when Sentinel-1 and Sentinel-2 were combined. Although the backscatter and spectral features of Sentinel-1 and Sentinel-2 could distinguish wheat lodging effectively, the combination of the two datasets would help to improve the classification accuracy. Therefore, applying the combinations of high spatiotemporal resolution SAR and optical remote sensing data in lodging monitoring can help reduce crop production loss and achieve higher crop yield quantity and quality.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page