Quantifying restored riparian buffer delineation accuracy and detectability in multitemporal imagery

We used multitemporal imagery to map tree cover and quantify the accuracy and detectability of planted riparian buffers. We used data with two spatial resolutions and found that classification of 1 m National Agriculture Imagery Program (NAIP) imagery using a texture-based automatic delineation process matched visual interpretation in terms of both accuracy of tree cover mapping and detectability. In contrast, our method struggled to detect vegetation changes in the riparian buffers using moderate resolution 30 m Land Change Monitoring, Assessment, and Projection (LCMAP) data. Overall pixel-based map accuracy of buffer tree cover using the 1 m automatic process averaged 96 %, while the 30 m LCMAP-based maps had an average accuracy of 37 %. Using the 1 m NAIP imagery, we found on average it took 6 years for planted buffer trees to reach a 50 % detectability level, and 11 years to reach 100 % detectability. Detectability of riparian tree cover increased from 13 % to 98 % 11 years after planting using the 1 m process while the 30 m process consistently had around 9 % detectability. We found correlation between soil types associated with frequent flooding and weaker detection of buffer tree cover. Median slope did not have a strong correlation with delineation accuracy but did influence the magnitude and timing of detectability. The delineation process utilized in this study and the identification of potential impact factors on accuracy and detectability will support efforts to delineate riparian buffer tree cover in other regions.