Massachusetts Forest Cover

For my senior honors thesis, I worked to update a Landsat forest cover map for Massachusetts to 2015. At Clark University forest cover maps for the state have been produced back to the 1970’s in ~10 year intervals based on data availability. Landsat 8 data for summer months of 2015 were used, with 15 images (3 per Landsat scene) in total. CLASlite was used, a software produced by the Carnegie Institution for Science, which classifies raw imagery into fractional cover (soil, photosynthetic vegetation, and non-photosynthetic vegetation) and forest cover maps. CLASlite is a semi-automated approach, user input is only data and thresholds for classification and masking, and most importantly, it’s FREE!

CLASlite has a change detection step included, however we chose to use our own method of doing change detection, as we wanted to use multiple dates of Landsat 8 imagery to classify forest for greater confidence. Each Landsat scene had 3 images, which each were classified into forest cover maps from fractional cover using CLASlite. Confidence of forest was calculated using the equation, $Confidence=\frac{N_{f}-N_{n_f}}{N_{f}+N_{n_f}}$, where $N_{f}$ is the number of times a pixel was classified as forest and $N_{n_f}$ is the number of times a pixel was classified as non-forest. This confidence map was then reclassified to forest cover where values 0 to 0.6 are non-forest, and 0.6 to 1 are forest.

Results

Forest Cover

Training Points were generated at 968 locations, stratified by Ecoregion. A basemap from MassGIS with WorldView imagery also from 2015 was used to validate that forest cover was accurately classified. Rate of accuracy varied by Ecoregion, with overall accuracy at 87%.
Forest and Non-Forest cover persistence, gain, and loss shown. This shows where forest cover has grown and where it has lost area.

Statistics

Avatar
Tyler Anderson
Graduate Student

Graduate Student at Clark University interested in Remote Sensing