The objective is to show that some policies and legislation affecting a territory might come from upper levels and that in any case, legislation and rules will be adapted to the local management system. That happens with the international RAMSAR convention paper, that derives in the habitat conservation legislation at European level. To promote the accomplishment of the legislation, the local administration subscribes to an established UE subvention. The subvention would define some requirements, including flooding practices. Farmers would fill in a form in which they would state which % of their fields will be flooded in a 4 months period.
The local administration should then evaluate the farmers' declarations. In order to guarantee the farmer has kept his word, they need a control system as objective as possible. This task is entrusted to a research centre that will develop the corresponding Expert system.
From the GEOSS datastore a compilation of RS scenes (once a month during the subvention period) and aerial images is extracted in order to obtain a multitemporal dataset. Part of this flow is repeated monthly. The scenes will undergo preprocessing corrections and then a classifier will be applied in order to obtain a categorical map with the following classes: flooded/not flooded. In fact, a high/low distinction is included for each category, so in the end we have 4 categories, which constitute ordinal categorical variables. The classification of the image is validated using the test areas which are revised in another field survey that must take place at most two days after the corresponding RS scene date. This is a determinant step related to the validation of dynamic processes. The confusion matrix estimation, in turn is vital for the classification accuracy quantification. Methods of quantification, classification and error modeling are compared. Accordingly, pixel level accuracy, object level indicators and global estimates are estimated. Completeness and thematic accuracy issues are thoroughly revised in this study.
The categorical map obtained from the classification step is analysed at the field scale, considering each farmer's parcels. This is the natural unit to be assessed: the real world is divided in parcels rather than pixels. Consequently, the cadaster is used as the vector layer delimiting the field contours in order to enable the disaggregation of the categorical surface into a categorical parcel surface. Some attributes related to the declaration form will be added to the attribute table in this layer. This might derive in consistency quality issues evaluation. In turn, the cadaster layer must be revised to avoid possible errors, which involves photointerpretation and ortophotographies revision. In the end, we have obtained a classification for each parcel. The uncertainty integration will define the uncertainty thresholds of the obtained product.
This process is done for each of the scenes: hence, a multitemporal dataset covering four months is obtained. Next, we can evaluate the compliance of the subvention criteria by defining some decision tree rules in several steps for each image. The results will be either that it is certain that the % of flooded surface stated by the farmer is flooded (and maybe even more surface than stated), that it is certain that the % of flooded surfaces is inferior to the surface stated by the farmer or else that we can't be certain of neither because there are missing data and/or the uncertainty threshold estimated is too wide for us to take that choice. In some cases, results can be refined, for instance checking special circumstances from external sources (e.g., maintenance works in the channels). Other issues take place when the parcel has a strange shape that results in that the number of pixels of the scene makes it difficult to calculate the flooded surface % in relation to the farmer's declared surface or when the parcel falls in shared pixels in the limit between two different categories. Maybe other factors need to be revised: e.g., some maintenance works in the channels have made impossible for the farmer to flood the field during some days including the date of the RS data recording. The uncertainty threshold that must be considered in the uncertain results comes from the uncertainty assessment system, and collects the diverse error sources and estimations.
The results are given to the local administration which has to take the final decision. Most probably, a farmer that receives a notification of approval and gets the money finishes the story here. Alternatively, there can be farmers that although sure of having accomplished with the subvention criteria have been denied the subvention and decide to start a legal process against the local administration. This litigation gives us the opportunity of analysing other issues related to quality both of processes and the general system in GEOSS: traceability and quality indicators , peer-review and even trustability, for instance. Note that some of these issues can be considered a two ways road (e.g. the so called backlinks and the other way round) and lead to metadata updating issues, user feedback, and reliability of producer data models.