Measurement Error in Remotely Sensed Optical Data
Description
Affected by measurement error, at-sensor radiance has inherent variation over time, space and spectrum, violating the stationary data assumption in regression-based applications in earth observation. These measurement errors may bias the relationships of remotely sensed data with natural resources such as forests. A short time series of Landsat 8 Collection 2 Level 2 data is generated based on forest sample plots, as replicate measurements of the remote sensing system. Using this data, we propose an approach to correcting the biased effect of remotely sensed data for improving the precision of forest inventory.
Files
Steps to reproduce
A short time series of satellite images were downloaded from Earth Explorer, and candidate predictors were calculated, including principal components of the original spectral bands, textural features of the first principal component, vegetation indices including Enhanced Vegetation Index (EVI), Generalized Difference Vegetation Index (GDVI), Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Specific Leaf Area Vegetation Index (SLAVI) and Simple Ratio (SR). Firewood volume is calculated by sorting undecayed dead standing woods, fallen deadwood and suitable living trees. A general volume model for dry climate zone was used for per-tree volume. The plot-level totals and density (m3/ha) were obtained by aggregation and expansion.
Institutions
- International Center for Bamboo and Rattan