University of Twente Student Theses


Variability of wheat phenology from Sentinel-1 and -2 time series : a case study for Brandenburg, Germany

Hoque, A. (2022) Variability of wheat phenology from Sentinel-1 and -2 time series : a case study for Brandenburg, Germany.

[img] PDF
Abstract:Crop phenology plays a vital role in regulating agricultural practices and managing natural resources. Accurate estimation of different phenological stages of crops can help in timely irrigation, yield prediction and fertilization. Remote sensing has proven to be a viable technique to monitor phenology over traditional field measurements. High spatial and temporal resolution of two European satellites, Sentinel-1 and -2 provide a unique opportunity to monitor phenology of crops. However, accurate estimation of crop-specific phenology is lacking at a large scale. This study estimated wheat phenology from Sentinel-1 and -2 time series for Brandenburg state of Germany for 2017 to 2021 using parcel information from Land Parcel Identification System (LPIS) data and ground truth information from the German Meteorological Station (DWD). The spatial and temporal variability of wheat phenology were explained with temperature and precipitation data. Time series of three SAR parameters from Sentienl-1 (VV, VH and VH/VV (CR)) and Enhanced Vegetation Index (EVI) from Sentinel-2 were generated for wheat fields in Google Earth Engine. LOESS and Double Logistics (DL) methods were tested to fit the data to reduce noise. As LOESS performed better in terms of representation of the temporal trajectory of wheat, LOESS was retained and used for subsequent analysis. Both CR and EVI were proven to be sensitive to wheat growing stages. Breakpoints and threshold methods were tested on LOESS fitted CR and EVI to match the DWD reported in-situ phenology phases. Third breakpoint from CR had median values of 0 to 3 days (different depending on the year) with milk ripeness phase, while the fourth breakpoint had median values of 1 to 5 days with the harvest phase. Peak of season (POP) estimated from EVI was correlated with the ears emerge phase with a median difference of 0 to 10 days. In general, the estimated dates were earlier than the field dates. For most cases, median values were less than the revisit period of both the sensors signifying the agreement between the in-situ and remote sensing estimates. However, due to outlier profiles, the RMSE values were between eight to twelve days. Coefficient of determination (R²) between mean estimated phenology and DWD reported phenology were 0.24 for ears emerge, 0.18 for milk ripeness, and 0.24 for harvest. It is noted that the low R² values do not coincide with the achieved RMSE and median values. For most cases, a southnorth spatial pattern was noticed where the south had earlier retrievals of phases than the north. A similar gradient was noticed in temperature, where the southern part experienced higher temperature than the northern counterpart. Interannual variability was evident from the findings where warmer temperatures led to earlier retrievals (R² 0.24 for ears emerge, 0.02 for milk ripeness and 0.25 for harvest). Weaker positive correlations were found between precipitation and estimated stages of crops. This study demonstrates that Sentinel-1 and -2 have a high potential to monitor crop-specific phenology of wheat for a large area which can be used in crop productivity monitoring. The information can help in risk damage assessment in agricultural systems.
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