Combining Camera Relascope-Measured Field Plots and Multi-Seasonal Landsat 8 Imagery for Enhancing the Forest Inventory of Boreal Forests in Central Russia

Abstract

The study considers a forest inventory for the mean volume, basal area, and coniferous/deciduous mapping of a large territory in central Siberia (Russia), employing a camera relascope at arbitrary sized sample plots and medium resolution satellite imagery Landsat 8 from the leaf-on and leaf-off seasons. The research bases are on field plots and satellite data that are acquired for the real operational forest inventory, performed for industrial purposes during summer⁻fall 2015. Sparse Bayesian regression was used to estimate linear regression models between field-measured variables and features derived from satellite data. Coniferous/deciduous mapping was done, applying maximum likelihood classification. The study reported the root mean square error for the mean volume and basal area under 25% for both the plot level and compartment level. The overall accuracy of the forest-type classification in coniferous, mixed coniferous/deciduous, and deciduous classes was 71.6%. The features of Landsat 8 images from both seasons were selected in almost every model, indicating that the use of satellite imagery from different seasons improved the estimation accuracy. It has been shown that the combination of camera relascope-based field data and medium-resolution satellite imagery gives accurate enough results that compare well with previous studies in that field, and provide fast and solid data about forests of large areas for efficient investment decision making.

Keywords

• remote sensing
• Operational Land Imager (OLI)
• Trestima
• forest classification
• Sparse Bayesian Regression