Estimating Mediterranean forest parameters using multi seasonal Landsat 8 OLI imagery and an ensemble learning method

• Published in: Remote sensing of environment Vol. 199; pp. 154 - 166
• Main Authors: Chrysafis, Irene, Mallinis, Giorgos, Gitas, Ioannis, Tsakiri-Strati, Maria
• Format: Journal Article
• Language: English
• Published: New York Elsevier Inc 15-09-2017; Elsevier BV
• Subjects: Dry season; Ensemble learning; Environment; Estimation; Forest inventory; Forests; Landsat; Landsat satellites; Machine learning; Minimal depth; Multi-temporal imagery; Parameter estimation; Random forest regression; Remote sensing; Satellite imagery; Seasons; Statistical analysis; Studies; Time series; Variable importance measure; Variable selection; Wood; Variable importance measure; Random forest regression; Minimal depth; Multi-temporal imagery; Forest inventory; Remote sensing; Variable selection
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2017.07.018

The overall aim of this study was to evaluate the use of seasonal time-series Landsat 8 Operational Land Imager (OLI) satellite imagery in estimating forest stand parameters in a heterogeneous Mediterranean environment. Within this framework, the random forest regression algorithm was used to model the relationship between spectral information and tree density, basal area, and wood volume, based on single-date, single-season (dry, wet), and multi-temporal (May–December) imagery. The variable importance (VIMP) measure and the minimal depth (MD) order statistic were also investigated with regard to improved prediction accuracy and the identification of relevant variables. In general, the multi-temporal and dry-season models were more accurate than the single-date models. The models resulting from the MD variable selection from the dry season imagery were the most accurate with a coefficient of determination of up to 0.54 for tree density, 0.72 for basal area, and 0.68 for volume.
•Random Forest regression models for single-date, seasonal and multi-temporal images•Variable importance (VIMP) and Minimal depth (MD) variable selection procedures•July imagery was most efficient in predicting basal area and wood volume•The SWIR spectral information presented as the most important variable•MD variable selection for dry season imagery indicated as the most accurate model