Extracting the full value of the Landsat archive: Inter-sensor harmonization for the mapping of Minnesota forest canopy cover (1973-2015)

摘要

Remote sensing estimates of forest canopy cover have frequently been used to support a variety of applications including wildlife habitat modeling, monitoring of watershed health, change detection, and are also correlated to various aspects of forest structure and ecosystem function. Although data from the long running Landsat earth observation program (1972-present) have been previously utilized to characterize forest canopy cover, the variability in spatial and spectral resolutions between the Landsat sensors has generally limited analyses to readily comparable imagery from the TM and ETM + sensors, which omits large portions of the full temporal record. In this study, we present an R package, LandsatLinkr, which automates the processes for harmonizing Landsat MSS and OLI imagery to the spatial and spectral qualities of TM and ETM + imagery, allowing for the generation of annual cloud-free composites of tasseled cap spectral indices across the entire Landsat archive. We demonstrate the utility of LandsatLinkr products, further enhanced through the LandTrendr segmentation algorithm, for characterizing forest attributes through time by developing annual forest masks and maps of estimated canopy cover for the state of Minnesota from 1973 to 2015. The forest mask model had an overall accuracy of 87%, with omission and commission errors for the forest class of 17% and 10%, respectively, and 9% and 16% for non-forest classification. Our resulting maps depicted a significant positive trend in forest cover across all ecological provinces of Minnesota during the study period. A random forest model used to predict continuous canopy cover had a pseudo R-2 of 0.75, with a cross validation RMSE of 5%. Our results are comparable to previous Landsat-based canopy cover mapping efforts, but expand the evaluation time period as we were able to utilize the entire Landsat archive for assessment.