PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2016
605
Basal Area and Diameter Distribution Estimation
Using Stereoscopic Hemispherical Images
Mariola Sánchez-González, Miguel Cabrera, Pedro Javier Herrera, Roberto Vallejo, Isabel Cañellas, and Fernando Montes
Abstract
In recent years, proximal sensing data has increasingly been used
to optimize forest inventories. In this paper we present a forest
inventory methodology based on stereoscopic hemispherical im-
ages. An automated pixel-based approach and a user-guided “re-
gion growing” approach have been developed for image matching.
To estimate the basal area, number of trees and mean diameter,
the sampling probability is determined for each tree. The accu-
racy and precision of the estimates derived from stereoscopic
hemispherical images was analyzed for a set of National Forest
Inventory plots. The results revealed that tree matching depends
on the species, the distance to the target tree and the diameter.
The Pearson correlation coefficient was 0.86 for the mean diame-
ter and 0.89 for the basal area, whereas for the number of trees per
hectare it was 0.59. The proposed methods may be used in large
scale forest inventories as a cost-efficient way of obtaining data on
diameter distribution and basal area from field surveys following
a two-stage scheme combined with remote sensing techniques.
Introduction
The changes being brought about by human activities in
forest ecosystems may lead to shifts in species distributions
(Hernández
et al
., 2014a), biological invasions (Hernández
et
al
., 2014b) or population decline processes (Sangüesa-Barre-
da
et al
., 2015). Under this scenario, forest management
requires knowledge of forest conditions and forest dynamics
over short time intervals. Forest inventory techniques should
achieve a tradeoff between cost efficiency and measurement
precision. In recent years, remote sensing data has increas-
ingly been employed in forest monitoring. These techniques
provide continuous data coverage of large areas periodically,
which is particularly useful for the early detection of changes
in forest ecosystems (Wulder
et al
., 2004). Since the launch of
high and medium resolution, multispectral satellite sensors
and the increase of aerial lidar applications, the possibilities
offered by these techniques for forest structure assessment
have been explored (Gómez
et al
., 2011; Gómez
et al
., 2012;
Næsset and Gobakken, 2008). However, above-canopy remote
sensing techniques present limitations for the estimation of
diameter distribution and number of trees because of the dif-
ficulty involved in distinguishing individual tree crowns as
well as interception by upper canopy layers. Hence, a second
stage of sampling, that typically consists of a 200 m × 200 m
to 1000 m × 1000 m grid (or other similar sampling design)
covering all the forest area, is required to provide precise field
measurements of the target variables in order to build regres-
sion estimators (Vauhkonen
et al
., 2011). Thus, interest has
focused on proximal sensing techniques.
To date, hemispherical photographs have mainly been used
in forest ecology to study radiation and Leaf Area Index (
LAI
)
(Frazer
et al
., 2001; Jonckheere
et al
., 2004).
LAI
estimation is
based on the inverse Poisson model (Weiss
et al
., 2004), which
establishes the gap fraction as a function of the zenith direc-
tion. Image segmentation is aimed at separating the visible sky
from the foliage elements (Ishida, 2004; Nobis and Hunziker,
2005; Schwalbe
et al
., 2009). In forest environments, differ-
ent approaches have been proposed to correct the effect of
interception by the woody parts of plants (Nilson and Kuusk,
2004), foliage clumping or the effect of slope on the canopy
geometry (España
et al
., 2008; Montes
et al
., 2007). Montes
et
al.
(2008) showed that the spatial pattern of the trees can be
derived from the angular variance of the gap fraction retrieved
from a single hemispherical image. The main problem with
methods based on a single hemispherical image is the difficul-
ty to determine the distance to features identified in the image.
The Utility Model MU2005-01738 (ForeStereo) patented by
the Spanish Forest Research Centre of the Spanish National
Institute for Agriculture and Food Research and Technology
(
INIA-CIFOR
) consists of a forest measurement system based on
stereoscopic hemispherical images, which allow the distance
to the matched features to be determined (Herrera
et al
., 2011;
Herrera
et al
., 2009). Similarly to other photogrammetric
techniques (Clark
et al
., 2000; Dick
et al
., 2010; Forsman
et al
.,
2012; Korpela
et al
., 2007; Varjo
et al
., 2006), the information
derived from the stereoscopic hemispherical images can be
used to estimate tree diameters and height and build taper
equations (Rodríguez-García
et al
., 2014). However, estima-
tion of basal area, density or structural attributes from the
information derived from the images still presents a challenge
due to the difficulty involved in determining the sampling
probability (Rodríguez-García
et al
., 2014). The stereoscopic
hemispherical images provide information on the visible light
spectrum, which is used to retrieve distances through ste-
reo-vision matching methods. Despite the differences between
the stereoscopic hemispherical images and the Terrestrial Laser
Scanning (
TLS
), which directly measures distances, the prob-
lem of stand variable estimation is similar for both techniques.
The main sources of bias are the instrument bias (the tree
detection capability reduces with distance due to the angular
resolution of the sensor), the assessment of diameter at breast
height (
DBH
) due to poor detection or irregular stems, and the
complete shadowing of trees by other trees, branches, foliage
elements and shrubs (Seidel and Ammer, 2014). In recent
years, methods for retrieving diameter, basal area, and density
data through
TLS
have been developed (Brunner and Gizachew,
Mariola Sánchez-González, Isabel Cañellas, and Fernando
Montes are with INIA-CIFOR,Ctra. de la Coruña km 7,5,
28040 Madrid, España (
).
Miguel Cabrera is with Aranzada Gestión Forestal, C/Alonso
Heredia, 31, 28028, Madrid, España.
Pedro J. Herrera is with the Universidad Francisco de Vitoria,
Ctra. Pozuelo-Majadahonda, km.1800, 28223 Madrid, España.
Roberto Vallejo is with the Ministerio de Agricultura, Ali-
mentación y Medio Ambiente, Gran Vía de San Francisco, 4,
28005, Madrid, España,
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 8, August 2016, pp. 605–616.
0099-1112/16/605–616
© 2016 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.8.605
