Effects of Mismatches of Scale and Location between Predictor and Response Variables on Forest Structure Mapping

Abstract

Measurement error introduced by mismatches of scale and location between response and predictor variables is one of the major sources of error in forest structure mapping. It affects the evaluation of analytical models, compromises the results of accuracy assessments, and reduces the accuracy of mapping products. Using forest structure attributes measured in a specifically-designed ground plot system, we isolated the measurement error from the total mapping errors that are related to multiple factors, and examined the distribution and magnitude of this error caused by a scale mismatch between a relatively larger forest unit and a relatively smaller forest unit, as well as location mismatch of a specific distance between two forest units of the same size. We demonstrated the effects of measurement error on the analytical models and resulting maps for three common mapping scenarios linking ground data with remote sensing imagery. Our results indicated that this scale- and location-related error can be analyzed using the Classical and Berkson error models in most practical mapping exercises involving data measured on-ground and remotely-sensed imagery, and that the distinct error pattern of each type of measurement error can be used to identify the major error source. Based on this analysis, we can adjust the plot design or adjust the resolution of imagery, and select the optimal analytical method to achieve the best mapping result.