Method for assessing the thematic and positional accuracy of seagrass mapping

• Published in: Marine geodesy Vol. 20; no. 2-3; pp. 175 - 193
• Main Authors: Bruce, E. M., Eliot, I. G., Milton, D. J.
• Format: Journal Article
• Language: English
• Published: Taylor & Francis Group 01.01.1997
• Subjects: geographic information system; Marine; positional accuracy; seagrass; Shark Bay; thematic mapping; ISW, Australia, Western Australia, Shark Bay
• ISSN: 0149-0419; 1521-060X
• DOI: 10.1080/01490419709388104

Maps showing the density and distribution of seagrasses in Shark Bay, Western Australia, were prepared through interpretation of vertical aerial photographs. The hard-copy maps were digitized and geo-referenced by registration against Landsat TM images using geographic information systems (GIS). Since habitat mapping of marine environments can generate a range of errors related to the techniques used in data capture, this process raised questions concerning the potential sources of error and the accuracy of the resulting maps. The uncertainty of data reliability is a critical consideration in the application of spatial data to environmental management. This prompts the need to determine map accuracy. A method was developed to assess the spatial accuracy of the seagrass mapping using GIS. The assessment involved: first, identification of potential sources of error; second, line transect surveys in the field for comparison with the mapped distribution; third, application of techniques to describe levels of attribute and positional error, including an error matrix, confidence limits, kappa statistics, and line intersect sampling; and fourth, estimation of the overall reliability of the spatial information. The error matrix demonstrated a low level of accuracy in the description of seagrass density, due to misclassification of low-biomass areas as sand and difficulties in distinguishing between similar categories. The kappa statistic can be used to assess the level of attribute accuracy and determine a more appropriate classification. The line intersect technique enabled calculation of boundary offset and accounted for the irregular nature of the field surveys. Under a reliability of 100 m positional accuracy, calculated using line intersect techniques, was high. Low levels of overall attribute accuracy demonstrated the importance of applying reliability measures to individual categories. This approach, combined with a measure of positional error, provides a more detailed indication of data reliability.