TRANCO: Thermo radiometric normalization of crop observations

Abstract

Crop type maps are essential for a wide range of applications such as crop monitoring, and yield estimation. In addition, Earth Observation (EO) systems allow robust and timely mapping of the earth’s surface, usually based on time-series. Yet, existing crop type maps are either global at coarse spatial resolution, or have a local or regional scope. The reasons for this gap can be linked to the scarcity of global crop type datasets at field level to train the models, their bias towards the Northern Hemisphere, as well as the limited transferability of existing crop type models across different regions. One of the main limitations on the transferability is driven by the phenological shift of the crops’ radiometric time series detected by Earth Observation (EO) systems, which is mainly induced by the different climates across regions. In this study, we explore the normalization of EO-based wheat time series with the accumulation of Growing Degree Days (GDD): the Thermo-RAdiometric Normalization of Crop Observations (TRANCO) system. The TRANCO system is based on the assumption that crop phenology evolution is mainly driven by temperature accumulation, represented by the accumulated GDD from Start of Season (SOS) to End of Season (EOS) dates, derived from a crop calendar. We tested the proposed method to normalize wheat on a database of globally distributed samples, whose results show a great improvement of the GDD F1 score (0.90) compared to a simpler normalization approach based on Time windows defined from SOS calendars (0.87) and a baseline without normalization (0.83).

Keywords

• GDD
• Crop type classification
• Remote Sensing
• MSMD
• Spatial cross-validation