Exploring data uncertainty in landscape planning: Assessing and evaluating uncertainties in different land use/land cover data

Data uncertainty has been identified as the most prevailing source of uncertainty in environmental assessments. Potential uncertainties in the data will spread to the output, which in some cases leads to an outcome that can be rendered questionable. In Science, the scale of the data should ideally match the processes that are being studied. A condition that cannot and must not be achieved in most cases of landscape planning. Land cover maps, for example, can contain uncertainties, one being the shape and location of objects (geometric uncertainty) and another one being the values attributed to these objects (thematic uncertainty). There is a considerable amount of methods for the assessment of uncertainties, but most of them are of limited use for landscape planning practice as they require substantial resources to conduct. The assessment of uncertainties is therefore still a predominantly scientific issue.
The aim of this research is to bridge this gap in creating information on data uncertainty for the implementation into landscape planning and to explore ways to communicate this information to planners and decision makers.
We used confusion matrices and descriptive statistics to assess the geometric and thematic uncertainties of different land use/land cover (LU/LC) data, for example Corine Land Cover (CLC) in comparison with the gold standard for landscape planning on most scales, which are biotope type maps. The study was realized for the federal states of Lower Saxony and Saxony-Anhalt in Germany. The different land cover types were classified into certainty levels according to their accuracy to give an easy to comprehend information about underlying data uncertainty.
Our Results show that the use of coarse data like CLC in landscape planning goes hand in hand with a loss of valuable thematic information but that the overall geometric uncertainty is low, with accuracy values between 70% and 90% for different data sets. This overall low uncertainty however is misleading. Land cover classes that are of high importance for landscape planning, like for example arable land, grassland and forest have significantly higher uncertainty linked to specific parcel sizes, predominantly when the size is small. When looking at ATKIS data which is remote sensed LU/LC data used in Germany that has the actual parcels as basis, the accuracy within the smallest set category (0-1ha) varies between 0 and 100% with an overall accuracy of only 50%. Other than that it is difficult to identify trends showing again that uncertainty in LU/LC-data should be communicated.
The proposed methods could, for example be used by the geodesic services of every federal state to generate uncertainty information on LU/LC data for the whole area of Germany. This information could be communicated by using free map variables or through extra layers that accommodate the LU/LC-data giving landscape planers easy to use information on uncertainty without requiring additional resources.