Improving Fractional Land Cover Data Sets from EOS MODIS
Abstract
Vegetation cover is a key component in earth system and carbon modeling and studies of fire occurrence, ecosystems, soils and hydrology. Global conservation planning and monitoring and carbon accounting require information about vegetation cover and change. With these needs in mind, the MODIS Vegetation Continuous Fields products (VCF) were created to provide annual global maps of fractional vegetation cover at 250 m from 2000 to the present. Three complementary continuous layers are created: percent tree cover, percent non-tree vegetation, and percent non-vegetated. These nested metrics are accompanied by a water mask and per-pixel quality layers. VCF products are being widely used across a range of applications. Improvements have recently been made in these products, currently versioned as MODIS Collection 6. In particular, they are unique in providing separate tree and herbaceous cover estimates needed for determining plant functional types in climate models. Hands-off, automated machine learning methods are used to create VCF products. Models are based on a training library of vegetation cover values, manually classified from Landsat scenes representing a wide range of global ecosystems. Bagged linear models for each year are built using automated open source statistical software from training and daily MODIS surface reflectance and thermal data. Models are then applied to MODIS data to create each annual VCF product. Comparisons of MODIS VCF products with field-based fractional tree cover and other remote sensing fractional vegetation cover products demonstrates good agreement, with RMS deviations of 9-23 percent vegetation cover varying by density and ecoregion. These products demonstrate an increased accuracy (6.5% measured as mean absolute error) and consistency to be gained over earlier methods such as mixture modeling and neural networks when deriving variables from global remote sensing data.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.B31I2593D
- Keywords:
-
- 1632 Land cover change;
- GLOBAL CHANGEDE: 1640 Remote sensing;
- GLOBAL CHANGEDE: 1855 Remote sensing;
- HYDROLOGYDE: 1942 Machine learning;
- INFORMATICS