Towards a portable, scalable, open source model of tree cover derived from Landsat spectra
Abstract
Tree cover is a key parameter used in a variety of applications, including ecosystem and fire behavior modeling, wildlife management, and is the primary way by which a variety of biomes are classified. At large scales, quantification of tree cover can help elucidate changes in deforestation and forest recovery and understand the relationship between climate and forest distributions. To determine tree cover at large scales, remote sensing-based methods are required. There exist a variety of products at various scales and extents, including two global products, Hansen et al.'s treecover2000 product and Sexton et al.'s Landsat Vegetation Continuous Fields (VCF) product. While these products serve an important role, they are only available for a limited set of dates: treecover2000 is available for the year 2000, and Landsat VCF for 2000 and 2005. In this analysis, we created a single model of tree cover as a function of Landsat spectra that is both calibrated and validated using small footprint LiDAR estimates of tree cover, trained across multiple Landsat scenes. Our model was found to be accurate and portable across space and time largely due to using a large amount of LiDAR - Landsat pixel pairs across multiple Landsat scenes to capture both sensor and scene heterogeneity. We will be releasing the model itself, rather than time-limited products, to allow other users to apply the model to any reflectance-calibrated Landsat scene from any time period.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2016
- Bibcode:
- 2016AGUFM.B34A..08G
- Keywords:
-
- 0428 Carbon cycling;
- BIOGEOSCIENCESDE: 0480 Remote sensing;
- BIOGEOSCIENCESDE: 1632 Land cover change;
- GLOBAL CHANGEDE: 1694 Instruments and techniques;
- GLOBAL CHANGE