S1S2Crops

Starting point / motivation

Imagery provided by the Sentinel satellites has found rapid uptake by the agricultural sector as a powerful means for providing timely information about crop type and condition. However, some important crops are still notoriously difficult to separate, most importantly wheat and barley.

Worldwide, wheat and barley are the second- and fourth most produced cereals worldwide. As they belong to a similar botanical taxonomy, they are challenging to separate from space. The best hope to distinguish crops is using 10-50m spatial resolution satellite image time series such as provided by Sentinel-1 and Sentinel-2 satellites.

The methodology relies on observing the micro- and macro-scaled temporal changes in crop phenology and morphology. These changes in the timing of the development stages of crops are related to differences in field management practices (planting dates, etc.), soil properties and meteorological conditions.

Optical imagery is in principle well suited for depicting changes in the phenology of crops, yet, cloud cover often hinders a correct identification of important crop development stages. Microwave remote sensing can observe earth at night and through cloud conditions, but here the structure of vegetation can have a substantial effect on the scattering mechanisms.

Contents and goals

The objective of this Austrian-Swiss project is to explore the feasibility of using the Sentinel-1 cross-polarisation ratio (ratio between VH and VV polarisations, CR), Sentinel-2 Level 2a product reflectance, and climate data to distinguish major crops including wheat and barley through the identification of critical growth stages and phenology.

Methods

The analysis will be carried out on the Sentinel-1 backscatter data cube available at the EODC Earth Observation Data Centre, which is a worldwide collection of pre-processed Sentinel-1A and 1B Interferometric Wide Swath images.

Sentinel-2 multispectral data cube available at the Google Earth Engine platform is also a worldwide collection of pre-processed Sentinel-2A and Sentinel-2B top of surface reflectance images. Different machine learning methods, such as random Forest, Support Vector Regression and Convolutional Neural Network will be tested to identify the best method for crop classification. Furthermore, stakeholders will be consulted to develop a concept and strategy for future funding applications and research and development.

Expected results

Furthermore, a first step in analyzing workflows that would allow integrating the EODC S1 data cube and S2 data cube into existing commercial services of AgriCircle will be taken, meeting the requirements of the envisioned stakeholders.

Coordinator

Vienna University of Technology - Department of Geodesy and Geoinformation

Project partner

AgriCircle AG

Vienna University of Technology
Department of Geodesy and Geoinformation
Univ.Prof. Wolfang Wagner
Wiedner Hauptstraße 8-10/E120.1
A-1040 Vienna