A suggestion for the evaluation of maps of small streams in forests

Abstract

Numerous remotely sensed stream-mapping methods have been developed over the last several decades to extend small stream protection. While the accuracy assessment of the final products has been considered to be very essential, the estimation of the positional accuracy of stream map products has not been well addressed because it is not easy to acquire the field reference data, especially for streams located in remote forest areas. More importantly, it is also because there does not exist any prominent statistical method to estimate the positional accuracy of stream maps. Since most positional accuracy assessment methods for the remote sensing data product are designed for at least two dimensional features, the linear feature of streams does not fit with those methods. As a result of this challenge, the previously introduced estimation methods for stream maps' accuracy have not dealt with the positional accuracy with regard to two error aspects, commission and omission errors. In this study, I suggest a road-crossing stream points (RCSPs) sampling method, which is the practical version of the contour line-based line transect sampling method to collect field stream position, and review the previously introduced estimation methods and the feasibility of the application of the error matrix, which has not been inspected before. I found that the error matrix has good reliability in evaluating stream network positions even if it has the pseudo-infinite problem. Among the error matrix-based accuracy estimates, the producer accuracy, user accuracy, andK 1 (KHAT-inf) are feasible to evaluate the positional accuracy of stream maps, but Congalton's overall accuracy cannot be applied to stream maps. The sampling method and the accuracy assessment method that are newly introduced in this study were applied to evaluate three D8 stream maps, and it was proved that these new methods are feasible.