A comparative analysis of modeling approaches and canopy height-based data sources for mapping forest growing stock volume in a northern subtropical ecosystem of China. Issue 1 (31st December 2022)

• Record Type: Journal Article
• Authors: Lin, Wenke; Lu, Yagang; Li, Guiying; Jiang, Xiandie; Lu, Dengsheng
• Is Part Of: GIScience & remote sensing. Volume 59:Issue 1(2022)
• Page Start: 568
• Page End: 589
• Publication Date: 2022-12-31
• Subjects: Airborne lidar -- ZiYuan-3 -- hierarchical bayesian approach -- growing stock volume -- subtropical forest; Geodesy -- Periodicals; Cartography -- Periodicals; Aerial photogrammetry -- Periodicals; Remote sensing -- Periodicals; 526.05
• DOI: 10.1080/15481603.2022.2044139
• Languages: English
• ISSNs: 1548-1603

Abstract

ABSTRACT: Lidar has been regarded as the most accurate data source for forest-growing stock volume (FGSV) estimation, but inconsistent acquisition dates of lidar data with field survey often result in poor FGSV estimation accuracy. Spaceborne stereo imagery is captured at regular intervals, providing new opportunities for mapping and updating FGSV spatial distributions. Digital Surface Model derived from spaceborne stereo imagery and Digital Terrain Model (DTM) derived from airborne lidar can be used together to produce a canopy height model (CHM) (LS-CHM), which can then be used to predict FGSV spatial distributions, but this methodology has yet to be explored. Our research attempts to compare the performance of LS-CHM and lidar-CHM (L-CHM) for FGSV modeling and to explore the advantages of using the hierarchical Bayesian approach (HBA) over traditional linear regression and random forest modeling approaches when sample size is small. Considering different forest types and topographical conditions, as well as the number of sample plots for each forest type, HBA is used to develop the FGSV estimation model, and the results are compared with those from linear regression and random forest approaches. The research results in a northern subtropical forest ecosystem indicate that overall, L-CHM provides better predictions than LS-CHM using the same modeling approaches, and L-CHM is especially valuable when FGSV is small or large, but when FGSV falls within 100–200 m 3 /ha, LS-CHM–based variables produce better modeling accuracy than L-CHM–based variables using linear regression or HBA. The HBA based on stratification of both forest type and slope aspect provides the best FGSV estimation, using either L-CHM or LS-CHM, and solves the modeling problem due to limited sample sizes for forest types. Our research provides new insights to using the combination of satellite stereo images and lidar-derived DTM for mapping and updating FGSV in a large area.