University of Twente Student Theses


Evaluation of temporal Sentinel-2 and Landsat-8 data for wheat lodging detection

Weldeyohannes, Andemichael Zerabruk (2021) Evaluation of temporal Sentinel-2 and Landsat-8 data for wheat lodging detection.

Link to full-text:
(only accessible for UT students and staff)
Abstract:Lodging stress limits wheat production and reduces grain quality. Timely and accurate detecting of lodging and its severity in wheat fields is needed to predict yields, estimate losses, and improve mechanical harvesting efficiency. The lodging score (LS) index is a commonly used quantitative indicator of crop lodging. Field measurement of LS is time-consuming and requires laborious field sampling. Remote sensing (RS) data provide timely and reliable information that can be used to study LS across large fields. Only a few studies have considered detecting and mapping wheat lodging using LS and free optical satellite data. The heterogeneous distribution of lodging, its temporal variability, and cloud presence make it challenging to detect LS using optical RS data. Sentinel-2 and Landsat-8 data with their high spatio-temporal resolutions may allow detection and monitoring of wheat lodging using LS. The study aimed to expand the current knowledge on crop lodging detection based on the LS using the SVM classification model and by utilising temporal Sentinel-2 and Landsat-8 data. Wheat fields were visited in Bonifiche Ferraresi farm, Italy, at multiple growth stages (stem elongation until ripening), and their crop biochemical and biophysical properties, including CAI and % of the lodged area (LA), were measured. LS was then calculated from CAI and LA and used to divide the field data into four classes: healthy (He), moderate lodged (ML), severely lodged (SL), and very severely lodged (VSL). Sentinel-2 and Landsat-8 images that corresponded to the dates of field visits were selected for further processing, and metrics, including reflectance of individual bands and vegetation indices (VIs), were derived from them. Statistical analysis was performed to examine whether there are significant differences among the different lodging classes. Next, support vector machine (SVM) classification models were used to classify lodging using the LS and spectral bands and VIs of Sentinel-2 and Landsat-8. SVM classification was performed in three scenarios using: (i) spectral bands, (ii)VIs and (iii) a combination of spectral bands and VIs of temporal Sentinel-2 and Landsat-8 data. The performances of the prediction models were evaluated using overall classification accuracy (OA) and Kappa statistics (Kappa). The result of the SVM model showed that the combination of ten spectral bands and 15 VIs from Sentinel-2 exhibited better classification results (OA=80.5% & Kappa =0.67) than the combined six spectral bands and ten VIs of Landsat-8 (OA=77.8% & Kappa =0.63). The best predictive Sentinel-2 bands were the Red edge band 2 (RE2) (740 nm), followed by Red edge band 1 (RE1) (705 nm) and NIR1 (865 nm) band with their highest SVM variable of importance in differentiating the lodging classes. At the same time, the NIR (850 - 880 nm) band of Landsat-8 had the highest SVM variable of importance, followed by the Green (530 – 590 nm) band. The best predictive Sentinel-2 VIs were modified normalised difference water index (MNDWI), Green ratio vegetation index (GRVI), chlorophyll index Green (CIg), ratio vegetation index (RVI), and chlorophyll index Red edge (CLre), with their highest SVM variable of importance in distinguishing the lodging classes. At the same time, MNDWI, GRVI, RVI, CIg and normalised difference vegetation index (NDVI) of Landsat-8 VIs had the highest SVM variable of importance. Finally, the lodging severity classes of the wheat fields in Bonifiche Ferraresi farm was mapped using the best SVM model (combination of spectral bands and VIs of Sentinel-2). Our results are important for wheat lodging management decisions, including lodging spatial distribution and operational crop lodging evaluation. These results are also essential to support breeders in selecting lodging resistant wheat varieties and increasing world food security.
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