Agriculture is one of the most obvious industries with potential to benefit from EO data as identified by many market studies, such as a recent Copernicus Market Report. So far, the majority of EO services aimed at the agriculture industry focus on only two sectors: support for public policy (e.g. CAP) and precision farming services for individual farmers. However, EO data is as yet little used to support other elements of the AgroFood industry beyond the farmer, which contribute significant proportion of the added value in the entire “farm-to-fork” chain and its support web.

The transformation of EO data into derived information of value to various customers requires the development and advancement of numerous “component” services, such as refined crop type determination at local and regional scale, improved field parcel delineation, crop growth assessment along the season as well as crop yield forecasting and soil-moisture determination. This presentation is targeted to digital agriculture, a new scientific field that uses data intense approaches to drive agricultural productivity while minimizing its environmental impact.

We will show some lessons learned on the most relevant state of the art EO products derived using machine learning and AI methods for business-to-business applications with a section/focus on the importance of input data quality. For example, considering the potato industry in particular, there is potential for EO data to improve disease control, logistics, storage, and trade. But, also other business to business opportunities arise with the advent of the unprecedented observations collected by the Sentinel satellite family such as operational risk and impact assessments for the insurance sector or the support of global agricultural and rural development activities by Multilateral Development Banks and their national stakeholders.

A newly developed, scalable system will provide the means to integrate satellite data and other information streams directly in existing end user workflows in a business-friendly format, starting with the AgroFood sector. EO PLUG-IN is a new project supported by ESA’s “Investing in Industrial Innovation in Earth Observation” (InCubed) initiative, the first of its kind for the newly founded ESA Phi-lab. The project will provide a system of user enabled onboarding to ready-to-use satellite data-based information streams for the AgroFood industry to support, amongst others: business operations, production management, and operational efficiency.

A key component is to build a scalable processing infrastructure and an APIs, which turn satellite and ancillary data streams into valuable derived information solutions, delivered directly to the commercial end users who evoke a specific data stream. A generalised and flexible platform is being developed to provide a number of specific services initially requested by and designed in collaboration with AgroFood industry partners in the Netherlands and to be rolled-out globally. The setup of this platform is however in a way that it can be also expanded to further customers and topics, thereby growing the opportunities of the Earth observation industry in the agricultural sector.

Nitrogen is a biological fundament of crop cultivation – in conventional as well as ecological farming practices. Since the implementation of the EU Water Framework Directive though, nitrogen fertilization has been negatively brought into focus of the wider public as many member states failed to comply with the regulations with their common practices. The large aim now is to reduce the consumption of fertilizer and the pollution of natural resources (SDGs 6, 12 and 15) while maintaining crop yields at a constant level at the same time (SDG 2). To achieve this goal, the most expedient solution is to think about new ways to increase the efficiency of a limited nitrogen input.

The work presented here describes the development of such a new way in terms of optimizing nitrogen input by side-specific fertilizer application with the help of satellite data analysis (Sentinel-2). The idea is an integrated management system which is divided into two parts:

The first part is a complex long-term field analysis of crop specific biomass and yield formation of the last years to identify consistent growth patterns within the field. This analysis uses high-resolution optical satellite data as well as a sophisticated physically-based crop growth model (PROMET). The result is a site-specific yield potential map in tons per hectare which builds the basis for the calculation of the crop’s total nitrogen demand in kg N per hectare for every 10x10m point in the field – long before any actual biomass growth becomes visible on the field. This information enables varying the amount of fertilizer within the field already at the first nitrogen application of the year and can thus save resources wherever the potential is lower than the field average or apply them targeted in areas of high potential.

The second component which is especially relevant for the multiple-fertilizing strategy of winter wheat consists of an analysis of current biomass development. As soon as the crop can be monitored by the satellite, Sentinel-2 image analysis can be used to adjust the originally planned fertilization strategy to actual growth conditions. Spectral image analysis of Sentinel-2 data by inversion of the radiative transfer model SLC (Soil-Leaf-Canopy) results in maps of green leaf area [m²/m²] and chlorophyll content [μg/cm²] – two variables out of which the nitrogen uptake until then in kg N per hectare can be derived. Knowing the yield potential and thus the total N demand at each point in the field, the N uptake up to that time reveals the remaining demand.

As the current year may differ from average conditions represented in the yield potential, crop monitoring can be used now to adjust the original fertilization strategy to the actual conditions. Crop growth simulation under current meteorological conditions in combination with data assimilation of the analyzed satellite information leads to a more detailed estimation of the yield expectations for the present year at every point in the field. In case of worse than average conditions due to drought or low temperatures, for example, the remaining N demand can be downregulated according to the decreased yield expectations in the affected areas which finally reduces the risk of undesired nitrogen entry into the ground water.

This integrated nitrogen management system (Fig. 1) has been developed and tested in cooperation with several farmers throughout Germany for the last two years and can be extended to other countries in Europe. The retrieval of green leaf area, chlorophyll content and N uptake of wheat has been validated with in situ data in Bavaria. Partly finished and partly still pending is the assessment of the nitrogen efficiency in terms of how much nitrogen [kg N per hectare] has been used for producing each ton per hectare of yield. This assessment is done with yield measurements and nitrogen application data of the participating farmers.

The long-term goal for the future is to use an explicit nitrogen module in PROMET for simulating the nitrogen cycle and plant growth with respect to nutrient supply. This complex submodule already exists and has been validated successfully for wheat. Now it is planned to carry out simulations with this model for the above-mentioned test farms and compare the results to the previous method described in this abstract.

In the actual context of global warming it is essential to reduce the chemical fertilisation to the strict crop needs avoiding over fertilisation and the consequent environmental impacts. This study lies on an extensive field dataset of 3 years to estimate the wheat nitrogen nutrition status at the regional scale in Belgium. Taking advantage of existing tools, the wheat nutrition status was evaluated by the combination of Sentinel-2 images covering the entire Belgium every 5 days and the nitrogen nutrition index (NNI) proposed as a diagnosis tool by Lemaire et al. (2008). This tool allows farmers to decide the timing and the rate of N fertiliser to apply in their fields. The NNI is based on critical dilution curves widely established in the literature for specific crops and agronomical regions. These curves are empirically built on the link between the crop aerial dry matter (DM) and the associated N concentration (NC) required to reach the maximum yield called the critical NC (CNC). The NNI is then computed as the ratio between the actual crops NC in the DM per hectare and the CNC. These two variables (DM and CNC) were estimated with Sentinel-2 images in 2016, 2017 and 2018 to compute the NNI. Then, based on the conditional probability that the NNI exceed one, a threshold was defined to detect fields for which a nitrogen supplementation would lead to over fertilisation.

This study is divided in two parts, the first part focuses on the accuracy of the estimated variables to compute the NNI and the second part is dedicated to analyse the NNI in an operational framework and its application at the regional scale. Firstly, the objective was to assess the seasonal robustness of several vegetation indexes and biophysical variables (BV) to estimate NC and DM in wheat using 3 years of data from farmer fields spread in Wallonia, Belgium. The analyses conclude on (i) the added value of the BV to estimate the three variables of interest and on (ii) the possibility to estimate these variables either from a dataset of other years and period of data acquisition or from a limited sampling scheme during the year of interest. The CNC was estimated with the BV Canopy Chlorophyll Content with a Mean Absolute Error (MAE) between 3.36 kg/ha (R²=0.90) and 29.33 kg/ha (R²=0.86) depending on the field dataset used and the data acquisition period. According to the same methodology, the DM was estimated with the BV Green Area Index (GAI) with a MAE between 0.64 t/ha (R²=0.93) and 1.77t/ha (R²=0.72) and the accuracy of the computed NNI varied between 0.06 (R²=0.87) and 0.12 (R²=0.13).

Secondly, the impact of the errors made on these two variables on the NNI accuracy was investigated and a probability table at two different confidence levels was built at the threshold of 1. Two confidence levels (95% and 75%) were defined to support farmers in their decision to apply nitrogen fertilisation. Three labels were consequently built per field (i) crops able to absorb the N supplementation to increase the yield, (ii) crops either unable to absorb more N or crop whose the N added would not result in an increase in yield, and finally (iii) in between, crops around the optimal N fertilisation, leaving the option to the farmer to apply fertiliser without the insurance of making any profit.

Nitrogen (N) is essential for crop productivity and is one of the main limitations to crops reaching yield potential. Traditional N fertilisation practices – involving uniform applications at a fixed rate – are economically and environmentally inefficient. Alternatively, within-field variable rate N applications offer farmers the opportunity to apply an optimum N rate that maximises productivity whilst mitigating the negative environmental impacts of excessing N use. In the United Kingdom, national recommendations for N fertilisations are based on three principle components: crop N demand, soil N supply and fertiliser N recovery. Determining these parameters at the field-scale is complicated by a typically high spatial and temporal variability in soil properties. However, the use of multispectral sensors, mounted on Unmanned Aerial Vehicles (UAVs) and satellite Earth observation (EO) platforms, can be used to provide spatial estimates of crop N content.

This study evaluates the performance of UAV and EO-derived multispectral data, along with empirical calibrations, for spatially resolved recommendations on winter wheat crop N applications. We first estimate the key components for quantifying N requirements by fitting a model to data derived from destructive samples extracted from experimental field trails of varying N application. Second, using a calibrated empirical model that links crop N content to a vegetation index, we determine the validity of using UAV multispectral observations to calculate the N requirements. Third, using a UAV multispectral sensor array that matches the key wavelength intervals as the European Space Agency’s Sentinel-2 satellites, we hypothesise the uncertainty that would be introduced when using spatially up-scaled N content estimates within the N requirement calculations.

We demonstrate the agreement between wheat crop N estimates derived from the UAV multispectral camera to measured N values. We compare the corresponding differences in the N deficit calculation when using these measured and estimated N content values. In the context of precision agriculture, we further discuss the use of additional datasets and process-based modelling approaches for providing more robust and reliable estimates of crop N fertiliser requirements.

The recognition of spatial patterns within agricultural fields, presenting temporally stable similar yield potential areas, is very important for a range of applications, for example in precision management of sowing, fertilization and irrigation practices.
This can be currently performed at low-cost by using open access, high spatial resolution multi temporal imagery, such as that provided by Landsat and Sentinel-2 satellites. For this purpose, a number of clustering algorithms have been developed, mainly based on the analysis of vegetation indices, possibly integrated by yield maps and geophysical surveys, when available.
The suitability of soil conditions for crop growth, however, is likely to depend on the interaction of a crop species and variety, with the microclimatic and soil conditions, in respect to a combination of seasonal factors limiting crop growth, such as water and nitrogen availability, thus interfering with vigor related indices. As a consequence, we expect that a clustering algorithm that takes into account such variability, may benefit by a reduction in potential confounding factors, thus resulting in higher accuracy of species and variety-specific yield pattern recognition.
In order to assess novel clustering methods that take into account the seasonal dynamic patterns of crops, weather and soils, we assembled a dataset over five cropping seasons in agricultural plots in Central Italy. Meteorological data were obtained from nearby meteorological stations, and individual years and sites classified according to three rainfall regimes (dry, median and wet years). Crops were classified as irrigated and non-irrigated. Sentinel-2 and Landsat 8 images were downloaded for every year and field, and vegetated ones retained for clustering, based on vegetation indices thresholds. For every site, images were grouped based on climatic years and crop season, resulting in six groups of images. Morphological features (catchment, slope and aspect) internal to each plot were derived from a digital elevation model (DEM). Spatial correlation between reflectance in water absorption bands on bare soil images and the DEM derived features was used to test for the appropriateness of the latter to represent topsoil moisture conditions. A k-means clustering method based on vegetation indices, seasonality, climatic and morphological information was then built in order to assess the intra-plot spatial yield patterns. The data were validated using yield maps collected in the field in different years. The methodology provided interesting results when compared to a previous clustering algorithm which did not take into account the seasonal dynamics of crops and weather.

Different users in the agricultural sector can benefit from real time information on crop development. Farmers can optimize their use of resources by having information on growth stage and biomass, allowing them to increase their profits while minimizing their environmental impact. Food producers can use information on crop development for yield forecasting, and logistics planning.

Although time series of high-resolution optical data is a potentially useful source of information for crop monitoring, their reliability in the Netherlands is severely undermined by cloud cover. Radar offers a solution to the cloud cover challenge in the Netherlands. Until recently, monitoring crop dynamics was hindered by the lack of satellite data time series with high temporal and spatial resolution. The launch of the Sentinel-1 C-band SAR constellation developed by European Space Agency, provides an unprecedented, continuous all-weather dataset with global coverage. This opens a new opportunity for detailed monitoring of crop dynamics and phenology in the Netherlands.

The goal of the current study is to demonstrate the potential of Sentinel-1 data for monitoring the growth cycle and extracting reliable and meaningful information on crop development and key dates of five key crop types in the Netherlands. We performed ground validation during a full growing season to support Sentinel-1 interpretation and validation. Crop height and growth stage were monitored weekly in a total of 24 agricultural parcels of maize (5 parcels), potato (4 parcels), sugar beet (5 parcels), maize (5 parcels) and English rye grass (5 parcels) in the province of Flevoland, Netherlands. Hydrometeorological data were collected throughout the season. This study utilizes the dense Sentinel-1 Interferometric Wide time series over the Flevopolder. Combining data from Sentinel-1A and 1B, the Flevopolder is covered by 4-5 tracks which provides an average of 20-25 acquisitions per month.

The results of this research study indicate that Sentinel-1 data follow the phenological stages and reflect changes in wet biomass and structure as the plant develops. For each of the crops considered, the increase in biomass during the vegetative stages and phonological development is clear in backscatter time series. Furthermore, maps of key dates can be made using Sentinel-1 data.