Published on 9 February 2022
WorldCereal Validation Approach
Throughout this first phase of WorldCereal, several classification algorithms have been benchmarked for the different WorldCereal products (annual cropland, active cropland, irrigation, maize and wheat) taking into account the specific growing season. For a full description of the products, the seasons and the timing of the products, see VITO's earlier blog post.
After benchmarking and before moving to global demonstration of the products and system, the robustness of the prototypes needed to be validated on larger areas. Validation is an important part of any map production process as such it was already done over smaller areas during benchmarking, but this needed to be repeated over larger areas.
For large scale demonstration validation, we define "validation" as product comparison against independent reference data sets that were not used in the production. However, depending on a data sampling design and data quality, such a comparison could be split into statistical validation and verification against other products.
For each WorldCereal product, a different validation approach was followed:
- The annual cropland was subject to a formal statistical validation. We used a customized Geo-Wiki interface to develop a new validation data set (further the Geo-Wiki data set), satisfying all necessary requirements for statistical validation.
- The active cropland layers were compared against the Geo-Wiki data set, which also includes information about main growing season.
- The irrigation products were also compared the Geo-Wike data set, which includes information about irrigation signs but only those detectable via visual inspection of VHR imagery.
- The crop type products, a verification against available in-situ reference data was performed. A full statistical validation was not always possible due to the low availability and poor spatial distribution of the reference data.