Spatial relationships between satellite-derived Copernicus HR-VPP crop seasonal productivity maps and ground yield maps in wheat and maize in support of precision management strategies

Abstract

The estimation of the spatio-temporal variability of crop yield within a field is an essential part of the information required to improve the agronomic management in precision agriculture. For example, the identification, from the time series of yield maps, of stable high or low yielding areas, as well as temporally instable areas, supports the delineation of uniform management zones, for the creation of prescription maps. In the absence, for most Italian farms, of long time series of yield maps obtained from combine harvesters equipped with yield monitors, spatial and temporal patterns of crop productivity can be inferred from earth observation satellite data, such as Sentinel-2. In this context, the High-Resolution Vegetation Phenology and Productivity (HR-VPP) suite of products, delivered by the Copernicus Land Monitoring Service at 10 m resolution from January 2017 onwards, and freely available in the WekEO platform (www.wekeo.eu) are particularly interesting. The suite of products includes total productivity (growing season integral; TPROD) and seasonal productivity (growing season integral above the start and the end days; SPROD). TPROD and SPROD are calculated as the growing season integral of the Plant Phenology Index (PPI) computed as the sum of all daily values between the start and the end of the growing season, and as the sum of all daily values minus their base level value, respectively. They have been conceived as indicators of crop productivity, thus could be potentially used for supporting field zoning for precision agriculture purposes. This study aimed at assessing the spatial relationships between Sentinel-2 derived TPROD and SPROD products, and ground-based crop yield spatio-temporal patterns. For this purpose, data from a range of agricultural fields with herbaceous crops such as wheat and maize, located in Maccarese (Central Italy) and Jolanda di Savoia (Northern Italy) were used. For each field, yield data were recorded at harvesting by a combine harvester equipped with a GNSS yield mapping system. The raw data were pre-processed for outliers’ and errors removal with R and Vesper software was used for geostatistical interpolation using block kriging interpolation to a 10 m grid aligned with Sentinel-2 pixels. For each field, the TPROD and SPROD products were obtained from WEkEO, which is the EU’s Copernicus DIAS reference service for environmental data, virtual environments for data processing. The pixel-based spatial correlation between the TPROD or SPROD maps and the ground yield maps was assessed by obtaining a linear model of coregionalization and assessing the codispersion coefficient against the hull of perfect correlation, and computing the Dutilleul t-test for the spatial processes of the SpatialPack package implemented in R. This test allowed to evaluate the consistency of the maps predicted by the satellite using TPROD and SPROD, with the yield maps showing a good spatial correlation between the maps in most cases.