Global Crop Area Monitoring at High Resolution Exploiting Complementary Use of Free and Open SAR and VSNIR/SWIR Sensor Data Sets

Abstract:

Earth Observation imaging sensors with spatial resolutions in the 10-30 m range allow for separation of the area and crop status contributions to the radiometric signatures, typically at parcel level for a wide range of arable crop production systems. These sensors complement current monitoring efforts that deploy low (100-1000 m) resolution VSNIR/SWIR sensors like MODIS, METOP or PROBA-V, which provide denser time series, but with aggregated and mixed radiometric information for cropped areas.

“Free and Open” access to US Landsat imagery has recently been complemented by the European Union’s Copernicus program with access to Sentinel-1A C-band SAR and Sentinel-2A visual, near and short-ware infrared (VSNIR/SWIR) sensor data in the 10-20 m resolution range. Sentinel-1A has already proven that consistent time series can be generated at its 12 day revisit frequency. The density of Sentinel-2 time series will greatly expand the availability of [partially cloud covered] VSNIR/SWIR imagery. The release of this large new data flow coincides with wider availability of “big data” processing capacity, the public release of ever more detailed ancillary data sets that support extraction of georeferenced and robust indicators on crop production and their spatial and temporal statistics and developments in crowd-sourced mobile data collection for data validation purposes.

We will illustrate the use of hybrid SAR and VSNIR/SWIR data sets from Sentinel-1 and Landsat-8 (and initially released Sentinel-2 imagery) for a number of selected examples. These include crop area delineation and classification in the Netherlands with the support of detailed parcel delineation sets for validation, detection of winter cereal cultivation in Ukraine, impact of the Syrian civil war on irrigated summer crop cultivation and recent examples in support to crop anomaly detection in food insecure areas (North Korea, Sub-Saharan Africa). We discuss method implementation, operational issues and outline the needs for further research in support to our crop “knowledge inference” framework. Most of our work is based on the use of Google Earth Engine (GEE) and the first batch of 12,000 geocoded Sentinel-1A images that was recently included.