The Netherlands is the second largest exporter of food and agricultural products in the world, exporting 65 billion euro of agricultural produce annually. These exports represent 17.7% of total Dutch exports [1
] and the sector is expected to grow further in the next 15 years [2
]. The provision of timely, reliable, accurate and high resolution satellite remote sensing data are essential to facilitate a transition from parcel-level decision-making towards precision agriculture [3
]. This transition towards precision agriculture is expected to yield increased productivity, lower environmental impact, transparent production and smarter production methods. Furthermore, the commercialization of remote sensing data processing and added-value product generation has the potential to become a valuable export commodity.
The current abundance of high-resolution optical data offers unprecedented opportunities for agricultural applications. Satellite remote sensing provides valuable information for many users from individual farmers to food producers, as well as national and international governmental agencies. Optical imagery can be used to predict yields, delineate management zones, support variable rate application and to monitor inter-field, intra-field and interannual variability [5
The European Union aims to reform the Common Agricultural Policy (CAP) in 2020. The Integrated Administration and Control System (IACS) implements area-based, direct and rural development payments for farmers using high, and very high resolution satellite imagery. Given the abundance of open and free satellite data available through the European Commission’s Earth Observation programme (Copernicus), the EU is encouraging member states to change the IACS system to use Sentinel imagery for 100% monitoring of agricultural parcels. Therefore, many EU member states are actively investigating the feasibility of exploiting Sentinel data for monitoring crops [9
The Netherlands Space Office (NSO) makes satellite imagery freely available for The Netherlands with a view to stimulating the development of innovative applications in agriculture and other domains [10
]. In addition to freely available imagery from MODIS, Sentinel-2, etc., NSO acquires satellite imagery from commercial providers and makes it freely available for The Netherlands. However, the reliability of optical imagery in The Netherlands is severely undermined by cloud cover. Van der Wal et al. [11
] used 20 years of data from a Royal Netherlands Meteorological Institute (KNMI). weather station data at Eelde (the Netherlands) to highlight the influence of cloud cover on the availability of optical imagery in The Netherlands. They showed that there is about a 20% chance of obtaining a clear (<2 Oktas) satellite acquisition during the growing season. While UAVs have been embraced by some growers, large-scale monitoring still depends on satellite remote sensing.
Low frequency (1–10 GHz) radar penetrates cloud cover and does not depend on solar illumination, so its potential to provide the timely and reliable observations essential for agricultural applications has been recognised for many decades [12
]. More importantly, radar observations in this frequency range are sensitive to the moisture content, size, shape and orientation and roughness of the vegetation constituents as well as the moisture content, texture and roughness of the underlying soil [12
]. In other words, low frequency radar is well suited to sensing soil moisture as well as water content and structural changes in agricultural crops. Several decades of ground-based experiments, and campaigns based on airborne and spaceborne sensors have demonstrated the suitability of low frequency radar for agricultural applications, particularly soil moisture monitoring, crop classification and crop monitoring [17
]. The European Space Agency’s (ESA) Sentinel-1 Mission offers unique opportunities in terms of the operational use of radar observations for agricultural monitoring for two reasons: (1) its revisit time is unprecedented and (2) the imagery is freely distributed. The two satellites—the C-band Sentinel-1 Mission (Sentinel-1A and 1B) was launched in 2014 and 2015, respectively. They are in the same orbital plane, providing an average revisit time of two days above
N/S, and achieve global exact repeat coverage every two weeks. The default acquisition mode over land in Europe is the Interferometric Wide-swath (IW) mode that provides both VV and VH data [18
]. Temporal coverage varies across the globe, but combining ascending and descending tracks from both satellites yields observations every 1–2 days in The Netherlands.
The high temporal revisit of Sentinel-1 provides insight into the temporal variability of vegetation, which has been exploited in many recent classification studies [19
]. The current study, however, is motivated by recent research highlighting the potential of Sentinel-1 for crop monitoring during the growing season. Veloso et al. [24
] compared Sentinel-1 data to Normalized Difference Vegetation Index (NDVI) estimates from optical data and ground observations of precipitation, temperature, green area index and fresh biomass. They demonstrated that Sentinel-1 data, particularly the VH/VV ratio, could yield useful information on crop development. In particular, they highlighted the potential to distinguish between crops based on the temporal variation of backscatter. They also showed that, for barley and maize, the NDVI and VH/VV agreed well with Green Area Index (GAI) and fresh biomass. More recently, Vreugdenhil et al. [25
] analyzed time series of Sentinel-1 backscatter and polarization ratio (VH/VV), and explored their relationship to vegetation water content (VWC), height, biomass and leaf area index (LAI). They showed that Random Forest modelling could be used to estimate VWC from Sentinel-1 imagery. The relationship between Sentinel-1 observables and VWC varied during the growing season, however, with variations in backscatter in some periods being dominated by structural changes. The studies, conducted in France and Austria, illustrate the potential to use Sentinel-1 imagery for crop monitoring.
The current study is focused on one of the most productive agricultural areas in The Netherlands to explore the value of Sentinel-1 in monitoring regionally important crops, namely sugar beet, potato, maize, winter wheat and English rye grass. By comparing Sentinel-1 imagery to hydrometeorological data and ground measurements of phenological stage, and vegetation height, it is shown that the time series of Sentinel-1 backscatter data reflects moisture and structural changes associated with phenological development of crops during the growing season in this region. It is shown that key dates of interest (emergence and closure dates) can be mapped using Sentinel-1 backscatter data. Finally, it is shown that, in addition to backscatter data, the coherence between consecutive Sentinel-1 images is influenced by the structural changes associated with (potato) haulming and harvest.
Results presented here illustrate the potential of Sentinel-1 SAR backscatter and interferometric coherence for crop monitoring and the detection of key dates for important agricultural crops in The Netherlands. Furthermore, the prevalence of rainy (and cloudy) conditions during this growing season underscore the need to use Sentinel-1 data by itself, or combined with optical data to provide timely and reliable information on crop monitoring in The Netherlands.
Time series analyses showed that, for each of the crops considered, structural and biomass changes associated with crop development influenced the backscatter throughout the season. Similar to the results of Vreugdenhil (2018) and Veloso (2017), the VH/VV ratio proved to be particularly useful as it reduces the influence of soil moisture. It is particularly sensitive to the increase in fresh biomass during the vegetative stages, and decreases during senescence as the vegetation water content decreases.
Key dates such as emergence and closure dates were estimated by fitting polynomials to the time series of backscatter, and were validated using field photos. Harvest detection proved to be more difficult in all crops apart from wheat. For sugar beet and potato, coherence data were needed to detect the harvest date.
This study used data from a single beam mode resulting in a temporal resolution of six days, replicating the more common Sentinel-1 acquisition strategy. Combining all beam modes would yield almost daily data in The Netherlands, improving the temporal resolution of emergence, closure and harvest date estimates. The sensitivity of interferometric coherence to structural and fresh biomass change is under-used at present, particularly in crop monitoring. The temporal density of Sentinel-1 data, particularly in Europe, offers an opportunity to capitalize on the synergy between backscatter and coherence data for agricultural monitoring.