PR Assessing class reduction coefficient
From Wikicap - European Commission
Go up to the main pro rata page
- The systematic method relies on assessment of the eligible part of the grassland by deducting the ineligible parts. For this, the pro rata grassland land cover polygon is delineated and its full area is proportionally reduced based by a coefficient. To allow such approach, the coefficient must be unbiased and precise and this, in turn, strongly depends on how well the above criteria apply to the land cover class.
- Precision is only possible if the land cover class definition itself is precise in terms of the scattered ineligible features. Their identifiable nature and predictable scattering are a prerequisite for the extrapolation of a reduction coefficient over larger territories.
- Absence of bias can be best achieved by determining the abundance of the grass at the optimal timing. This optimal timing depends on the agricultural activity and/or the method to assess the reduction coefficient. This is not necessarily the same time of the source imagery which should allow for the optimal delineation of the land cover polygon borders.
- A RC assessment has to be performed on a representative (i.e. random) sample of the targeted PG land cover classes; a sample of 30 field occurrences should be sufficient to start. Verify the representativeness after producing the first assessment results.
- Strictly adhering to the delineation instruction of the pro rata grassland class, a single polygon of each occurrence has to be delineated, 30 representative polygons in total. Such land cover polygon boundaries not necessarily follow reference parcel perimeters. Although the following steps may require more detailed data (e.g. imagery of better GSD, additional field surveys), ensure that the delineation corresponds to what is normally created for LPIS purposes even if done on the detailed data.
- The grassland abundance of each polygon must then be established by a method appropriate for the class concerned. Different classes may well require different methods and a different timing. The LPIS community has been using several methods, any of which may be applied after demonstrating the result is realistic (i.e. compared to the reality on the ground) and relevant (in IACS terms): Some, not all, of the methods already demonstrated by LPIS custodians are:
- applying pre-set scorecards by both photo-interpreters and field observers,
- perform multi-temporal image analysis and classification (multiannual and seasonal vegetation monitoring) with a variety of sensors,
- automatic classification of lidar and radar imagery, and
- detailed photointerpretation or field survey.
- A statistical analysis of the sample should confirm its representativeness for a single and specific class, i.e. results adhere to a normal distribution and the standard deviation is less than 5%. If either one of these conditions is false, refining or redefining the classes of the eligibility profile should be considered. An iteration of the assessment should follow.
- The sample retained as representative for a specific pro rata PG land cover class shall be used for further processing of that land cover class:
- When visual photointerpretation or field observation are applied, the relevant data can be used to develop and document an interpretation key.
- For automatic classification, the sample (stored as multispectral and multi-temporal imagery) can be used training in supervised classification, or for post-classification of unsupervised classification.
Go up to the main pro rata page