Agriculture


World Cereal prototype global 10 metre crop products for wheat and maize. National pilot products are available for five countries.

Dataset Description

Due to the wide variety of landscape dynamics, crop types, growing seasons and agricultural management practices, mapping of cropland extent, crop types and irrigation practices at the global scale still remain very challenging tasks. Several attempts have already been made to come up with accurate global maps, but until now not one has succeeded in providing seasonal information at field level on a global scale. Current crop map layers either lack spatial detail or fail to provide regular updates. As such, the AFOLU community is still in need of a system that can provide seasonal global agricultural monitoring information at field level. For this reason the European Space Agency, in collaboration with stakeholders in global agriculture like GEOGLAM, FAO, AMIS, has initiated the WorldCereal project to demonstrate such a system based on open and free data.

A total of three WorldCereal products are defined in the project:

  • Dynamic global annual cropland extent map
  • Seasonal global distinction between irrigated and rainfed cropland
  • Seasonal global crop type maps for wheat and maize
The three products will not be generated independently, but rather in a hierarchical fashion, as illustrated in Figure 1. The primary product will be the global annual cropland extent, and only cropland areas according to this product will be further processed into irrigated/rainfed cropland and global maps of wheat and maize. 1. The global annual cropland extent product is a binary mask at 10m spatial resolution indicating whether or not a certain pixel contains annual cropland.

WorldCereal will follow JECAM’s 2018 definition for annual cropland from a remote sensing perspective: ‘The annual cropland from a remote sensing perspective is a piece of land of minimum 0.25 ha (min. width of 30 m) that is sowed/planted and harvestable at least once within the 12 months after the sowing/planting date. The annual cropland produces an herbaceous cover and is sometimes combined with some tree or woody vegetation* .’**

*the herbaceous vegetation expressed as fcover (fraction of soil background covered by the living vegetation) is expected to reach at least 30 % while the tree or woody (height >2m) cover should typically not exceed a fcover of 20%.

**There are 3 known exceptions to this definition. The first concerns the sugarcane plantation and cassava crop which are included in the cropland class although they have a longer vegetation cycle and are not yearly planted. Second, taken individually, small plots such as legumes do not meet the minimum size criteria of the cropland definition. However, when considered as a continuous heterogeneous field, they should be included in the cropland. The third case is the greenhouse crops that cannot be monitored by remote sensing and are thus excluded from the definition.’ When applying this definition, the issues with field size, agro forestry are solved.

The base cropland mask product will be generated once a year for which the timing depends on the seasonality and/or reporting periods in a particular agro-ecological zone (AEZ).

2. The second WorldCereal product is a global seasonal mask at 10m spatial resolution where the cropland extent is divided into irrigated and rainfed agriculture. The definition of irrigated agricultural area as defined by the World Bank is as follows:

Irrigated agricultural area refers to area equipped to provide water (via artificial means of irrigation such as by diverting streams, flooding, or spraying) to the crops (https://databank.worldbank.org/metadataglossary/world-development-indicators/series/AG.LND.IRIG.AG.ZS) Not all area equipped for irrigation is also actively irrigated. Irrigation may not occur due to for instance crop rotation, water shortages, and damage of infrastructure (Siebert et al., 2013) . Within the WorldCereal system only actively irrigated areas are identified as irrigated agriculture. Actively irrigated areas are identified when scheduled irrigation is needed for a crop to grow given the climatic conditions for that specific pixel. In other words, when rainfall alone is not enough to produce a crop. The above definition implies that only areas equipped for irrigation are included in the irrigated cropland mapping.

3. The third WorldCereal product is a global seasonal crop type map for wheat and maize.

Crop type maps indicate the presence of specific crop types in a certain time range for a specific portion of land. The WorldCereal crop type maps will be generated specifically for wheat and maize. This dynamic map shall follow the major wheat and maize growing seasons across the globe, using as a reference crop calendars from GEOGLAM Crop Monitor (Becker-Reshef et al., 2019) or FAO (Fischer et al., 2012), indicating which parts of the seasonal cropland extent product are covered by wheat or maize (Skakun et al., 2017; Becker-Reshef et al., 2018). Production of these maps will be scheduled according to the globally compiled maize and wheat crop calendars. Other crop types are not covered during the project, although it is noted that the technical capability of producing crop type maps for other crops will be supported, as also requested by the users. These other crop types are captured by the seasonal “active cropland” flags whenever the system performs a seasonal update.

For COP26 these products will be demonstrated for 1 year and for 5 countries (Argentina, Spain, France, Ukraine and Tanzania).

Usage

Users should keep in mind that the WorldCereal maps are datasets at global scale, generated with a single methodology applied over all regions. As such, the accuracy of the map may vary between locations. That said, the dataset can therefore be expected to be most useful for countries that do not have their own agricultural monitoring systems. In situ data on crop type, irrigation is of course very crucial for training the classification algorithms. This data is very fragmented, where in some areas a lot of information is available for several years and in other parts of the world no data is available. Although the methodology has gone through benchmarking and testing it is clear that areas with a lack of in situ data in the training database will have less accurate results than areas with sufficient in situ data.

Methodology

At the start of the WorldCereal project a thorough benchmarking exercise was set up to compare the performance of different classification algorithms in different zones of the world and depending on different input data.

It is clear that the success of WorldCereal will depend upon the robustness of the algorithms, taking into account the large differences of available input data, temporal behavior of the crop etc. to name but a few of the variables.

The seasonal nature of the WorldCereal products requires a careful assessment of global crop calendars for wheat and maize. A major task has been the creation of such global pixel-based crop calendars covering all possible wheat and maize seasons, and combining the results in grouped agro-ecological zones (see figures below), thereby leveraging as much as possible existing sources of information, such as GEOGLAM Crop Monitor, FAO crop calendars and JRC-ASAP. Based on this information, the WorldCereal system exactly knows when to process which area to generate end-of-season crop type maps for our crops of interest.

Global pixel-based crop calendars for winter and summer cereals (SOS = start of season; EOS = end of season)

Agro-ecological zone groups, where grouping is based among others on the similarity of wheat and maize crop calendars

Based on this zoning and the input data sets gathered the benchmarking of the different classifiers was done. At the time benchmarking was done on > 100,000 reference samples. For each sample, a spatio-temporal input datastack of 640 m X 640 m covering 1,5 years of input data was extracted and stored for Sentinel-1 backscatter data, Sentinel-2 L2A reflectance values and ancillary data such as AgERA5 meteo inputs and the Copernicus 30 m DEM. Of course, even with this amount of reference data (and the reference database is still growing, (in close collaboration with the in-situ data working group of GEOGLAM) there are a significant amount of AEZ without training data. The grouping of agro-ecological zones according to their crop calendars led to a significant reduction of these regions without any useful reference data. However, still many regions with limited to no data remained. This introduces a difficult challenge: training one globally applicable model would lead to too many compromises. Training many regional models on the other hand leaves many agro-ecological zones without a model due to lack of reference data and it also could lead to artefacts at the zone borders.

It was decided to go for the best of two worlds and setup a hierarchical model approach. Train a global base model on all available reference data and progressively finetune the model in regional zones if sufficient reference data is available. As such a base model can be applied anywhere, while having potentially locally finetuned models that inherit from the base model but are better adapted to local conditions. Another advantage of this approach is that the regional finetuning needs much less training data, as lower-level features have already been learned in the global base model which could benefit from all reference data.

Hierarchical model approach with one global base model and several locally finetuned AEZ models

Uncertainty and Accuracy

As mentioned in the data set section and methodology section, uncertainty and accuracy of the results will be depend upon the area of the World. Areas with large fields and where we training data will have higher accuracies than areas with small fields, complex seasonality and little to none training data. Accuracies will also be product depended, whereas cropland extent will normally have a higher accuracy than the irrigations product. The overall product accuracy for cropland extent should be 80%, for the croptype maps this should be 70%.

Dataset Sustainment

The WorldCereal system is built upon the open and free data stets of the Copernicus Programme (Sentinel 1 and Sentinel 2) and Landsat 8. These data sets will be sustained by the different space agencies in the future. Next to this, all algorithms and methods used in WorldCereal are going to be open-sourced! The reference database that is being built and a key asset of WorldCereal will be further supported by GEOGLAM, through the GEOGLAM in situ data working group.

Technical Characteristics

Spatial resolution: 10m

Geographical coverage: Global

Temporal coverage: 2022 onwards

Update frequency: Seasonal


Associated Guidance or User Manual

Manual / guidance: In progress – meantime refer to https://esa-worldcereal.org/en

Web address for dataset download: In progress – meantime refer to https://esa-worldcereal.org/en


Points of contact for queries

Sven Gilliams
WorldCereal Project Manager
Flemish Institute for Technological Research (VITO)
Mol, Belgium
Email: sven.gilliams@vito.be

Zoltan Szantoi
WorldCereal Technical Officer
European Space Agency
ESA/ESRIN
Frascati, Italy
Email: Zoltan.szantoi@esa.int