Assessment of Classification Accuracy

Introduction

The quality of the habitat information derived from remotely sensed data is determined by the quantitative process of accuracy assessment. The purpose of accuracy assessment is to identify and quantify errors in the maps by comparing the attributes of the map versus reference data at various sites. It is important that the mapmaker know how reliably a given habitat can be classified. This parameter is called "producers accuracy". The users of a map product want to know what percentage of the polygons of a particular class are correctly attributed. This parameter is called "users accuracy". Furthermore, remotely sensed imagery that may be suitable for mapping coral reef habitats can be acquired from a wide variety of platforms and imaging systems, each having their own strengths and limitations. It is important to identify the technical merits of each, one measure of which is the thematic accuracy of the map products.

To determine the overall accuracy of the mapped product, GIS data prepared by visually interpreting IKONOS satellite imagery was assessed for accuracy using conventional methodologies. It was proposed that specific areas being mapped be used as test areas for this work. A statistically robust data set composed of random field habitat observations were collected within the test areas to assess the accuracy of the mapped product. These areas were chosen based on input from the local marine biologists and coral reef managers. These groups provided advice on the location of the most diverse benthic communities and also areas of particular importance based on management strategies and marine protected areas. Thus, it was the goal of this team to collect accuracy assessment field data representing as many of the habitats that occur in these regions as possible.

The thematic accuracy of all mapped products was determined at the most general and detailed levels of the classification scheme including both the biological cover type and geomorphological structure. Eight coral reef test areas were selected based on the diversity of the habitat types and to assure that all benthic habitats throughout the main Hawaiian Islands were represented. The accuracy of all maps is, therefore, considered a conservative representation of the thematic accuracy of the habitat maps prepared using the same methods.