Assessment and visualisation of uncertainty in remote sensing land cover classifications

摘要: The ability of space- and airborne instruments to measure the amount electromagnetic radiation reflected emitted by Earth’s surface has proved to be valuable for understanding our environment, as it provides an overwhelming flow data on appearance condition planet. data yielded remote sensing can be subjected various types computer-assisted manipulation, arrive at derived sets tailored different application. Computer-assisted classification remotely sensed into qualitative classes, for example, is useful extracting information that exploited cartographic purposes, such in generation thematic maps land cover types. For a proper cartographic application, fitness use set sensed data needs assessed. practicability their be established means an accuracy assessment procedure. An error matrix created for matching random sample its counterpart from a reference representing actual environment. Accuracy based on matrix, however, several drawbacks. Among these non-spatial and general character global statement like 95% entire classification; moreover, time-consuming cost-intensive process. As consequence, easily omitted which, course, undesirable and may lead are unfit application hand. For assessing data, not the only consideration. More generally, phrase quality used refer the extent which characteristics meet requirements the application aimed user. A high indicates relatively information value considered - good use. Uncertainty key-issue in and, therefore, set. During life cycle uncertainties introduced and propagated often unknown way. For investigating uncertainty, effective measures need designed. To this end, relevant consider purpose to which measures employed. Here, focus exploratory perspective. Exploratory analysis aims acquiring insight stability possible classifications data. this purpose, knowledge about underlying is imperative. exploratory analysis, iterative process, needing not only uncertainty but also effective ways convey Visualisation generally a useful communication potentially information. In thesis a class presented, exploratory analysis together with ways cartographic visualisation uncertainty. The during characterised probability vectors yielded by-product most probabilistic procedures. emphasis laid maximum a?x posteriori where every pixel vector probabilities is calculated specifies each distinguished class being true class. reflect differences resulting classification stored gis serve basis derivation of weighted entropy. Besides efforts reduction the amount present set. a posteriori rules dealt allow introduction of priori levels sophistication -thereby exceeding simple approaches embraced existing image processing packages. Another strategy within realm dealing spatial is based idea decision allows optimal decision-making given uncertain classes. Combining theory (defining uncertainty related occurrence particular class) utility the desirability consequences resulting actions taken assuming that contributes selection best under given conditions. This particularly interesting when huge sets under uncertain circumstances far-reaching wrong decisions (e.g. agricultural fraud detection European Union). Both probabilistic results procedure other quality information visualisation order develop a framework metadata. Static well more dynamic offer grips user who but persuasive assess classification. Commercial packages still failing sound consideration spatial data stake, fact incited participants camotius project look functionality “uncertainty-sensitive system”. Such system Dutch situation extra value added by remotely always beyond all doubt; explicit evaluation these data inherent reveals true value. Two case studies have stressed role planning purposes by demonstrating monitor changes extent greenhouses over space and time, making inventories area. inclusion uncertainty information approach appeal made to several improve processing results. It stated that a will encouraged if clearly demonstrable. components been scrutinised methodological part of formalised demonstration programme could a blueprint commercial packages. downloaded from: http://cartography.geog.uu.nl/research/phd