Physically based inversion modeling for unsupervised cluster labeling, independent forest classification, and LAI estimation using MFM-5-Scale

Abstract

Unsupervised clustering is important for regional- to national-scale forest inventories where supervised training data are impractical or unavailable. However, labeling clusters in terms of land-cover classes can be labour intensive and problematic, and clustering and related methods do not provide biophysical-structural information (BSI). Canopy reflectance models such as 5-Scale are powerful forest remote sensing tools; however, 5-Scale can only be run in forward mode and is not invertible to obtain the required forest information. This problem was solved using multiple-forward-mode (MFM) coupled with 5-Scale to enable MFM-5-Scale inversion of land cover and BSI using a look-up table (MFM-LUT) approach that matches satellite image reflectance values with modeled reflectance values that have associated land cover and BSI, such as density, leaf area index (LAI), and crown dimensions, as well as subpixel-scale component fractions. MFM requires no training data or a priori BSI and can optionally be stratified (generalized) by species, structural, hierarchical, mixed forest, and other class definitions. In this paper, MFM-5-Scale was used with Landsat thematic mapper (TM) imagery at the Boreal Ecosystem-Atmosphere Study (BOREAS) southern study area (SSA) modeling subarea (MSA) in Saskatchewan, Canada. MFM-5-Scale was used to label unsupervised cluster sets (n = 17 and 97) from a previous land-cover classification by progressive generalization (CPG), with the best results obtained from independent, stand-alone MFM classification (87%, 76%, and 71% for the three hierarchies of 16 forest type, species, and density classes) validated against the provincial (SERM) forest inventory map and also compared with a standard maximum likelihood (ML) classification. Further, MFM-5-Scale estimated LAI at 24 BOREAS plots within ±0.57 LAI compared with ground-based tracing radiation and architecture of canopies (TRAC) LAI validation data. BSI is not provided by CPG clustering or ML. Based on this and other studies, we conclude that MFM provides an inversion modeling context for sophisticated forest radiative transfer models to retrieve a higher level of land cover and BSI, with detailed LUTs providing a rich set of forest information suitable for query, analysis, and follow-on simulation studies. These methods can augment existing regional- to national-scale remote sensing based inventories by providing a robust cluster labeling and BSI capability or can provide stand-alone capabilities over a variety of applications and scales.