A Low-Cost and Robust Landsat-Based Approach to Study Forest Degradation and Carbon Emissions from Selective Logging in the Venezuelan Amazon

Abstract

Selective logging in the tropics is a major driver of forest degradation by altering forest structure and function, including significant losses of aboveground carbon. In this study, we used a 30-year Landsat time series (1985–2015) to analyze forest degradation and carbon emissions due to selective logging in a Forest Reserve of the Venezuelan Amazon. Our work was conducted in two phases: the first, by means of a direct method we detected the infrastructure related to logging at the sub-pixel level, and for the second, we used an indirect approach using buffer areas applied to the results of the selective logging mapping. Pre- and post-logging forest inventory data, combined with the mapping analysis were used to quantify the effects of logging on aboveground carbon emissions for three different sources: hauling, skidding and tree felling. With an overall precision of 0.943, we demonstrate the potential of this method to efficiently map selective logging and forest degradation with commission and omission errors of +7.6 ± 4.5 (Mean ± SD %) and -7.5% ± 9.1 respectively. Forest degradation due to logging directly affected close to 24,480 ha, or about ~1% of the total area of the Imataca Forest Reserve. On average, with a relatively low harvest intensity of 2.8 ± 1.2 trees ha-1 or 10.5 ± 4.6 m3 ha-1, selective logging was responsible for the emission of 61 ± 21.9Mg C ha-1. Lack of reduced impact logging guidelines contributed to pervasive effects reflected in a mean reduction of ~35% of the aboveground carbon compared to unlogged stands. This research contributes to further improve our understanding of the relationships between selective logging and forest degradation in tropical managed forests and serves as input for the potential implementation of projects for reducing emissions from deforestation and forest degradation (REDD+).

Full text

remote sensing
Article
A Low-Cost and Robust Landsat-Based Approach to Study
Forest Degradation and Carbon Emissions from Selective
Logging in the Venezuelan Amazon
Carlos Pacheco-Angulo 1, * , Wenseslao Plata-Rocha 2, Julio Serrano 1, Emilio Vilanova 3,
Sergio Monjardin-Armenta 2, Alvaro González 1and Cristopher Camargo 1


Citation: Pacheco-Angulo, C.;
Plata-Rocha, W.; Serrano, J.; Vilanova,
E.; Monjardin-Armenta, S.; González,
A.; Camargo, C. A Low-Cost and
Robust Landsat-Based Approach to
Study Forest Degradation and
Carbon Emissions from Selective
Logging in the Venezuelan Amazon.
Remote Sens. 2021,13, 1435. https://
doi.org/10.3390/rs13081435
Academic Editor:
Carlos Portillo-Quintero
Received: 11 March 2021
Accepted: 2 April 2021
Published: 8 April 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1Facultad de Ciencias Forestales y Ambientales, Universidad de los Andes, Mérida 5110, Venezuela;
jserrano@ula.ve (J.S.); alvarog@ula.ve (A.G.); ccamargo@unet.edu.ve (C.C.)
2Facultad de Ciencias de la Tierra y el Espacio, Universidad Autónoma de Sinaloa, Sinaloa M-80029, Mexico;
wenses@uas.edu.mx (W.P.-R.); sa.monjardin12@info.uas.edu.mx (S.M.-A.)
3Department of Environmental Science, Policy, and Management, University of California,
Berkeley, CA 94707, USA; evilanova@berkeley.edu
*Correspondence: carlosa@ula.ve or pachecocar@gmail.com; Tel.: +58-04121021374
Abstract:
Selective logging in the tropics is a major driver of forest degradation by altering forest
structure and function, including significant losses of aboveground carbon. In this study, we used a
30-year Landsat time series (1985–2015) to analyze forest degradation and carbon emissions due to
selective logging in a Forest Reserve of the Venezuelan Amazon. Our work was conducted in two
phases: the first, by means of a direct method we detected the infrastructure related to logging at the
sub-pixel level, and for the second, we used an indirect approach using buffer areas applied to the
results of the selective logging mapping. Pre- and post-logging forest inventory data, combined with
the mapping analysis were used to quantify the effects of logging on aboveground carbon emissions
for three different sources: hauling, skidding and tree felling. With an overall precision of 0.943, we
demonstrate the potential of this method to efficiently map selective logging and forest degradation
with commission and omission errors of +7.6
±
4.5 (Mean
±
SD %) and
−
7.5%
±
9.1 respectively. Forest
degradation due to logging directly affected close to 24,480 ha, or about ~1% of the total area of the
Imataca Forest Reserve. On average, with a relatively low harvest intensity of 2.8
±
1.2 trees ha
−1
or
10.5
±
4.6 m
3
ha
−1
, selective logging was responsible for the emission of
61 ±21.9 Mg C ha−1.
Lack of
reduced impact logging guidelines contributed to pervasive effects reflected in a mean reduction of
~35% of the aboveground carbon compared to unlogged stands. This research contributes to further
improve our understanding of the relationships between selective logging and forest degradation in
tropical managed forests and serves as input for the potential implementation of projects for reducing
emissions from deforestation and forest degradation (REDD+).
Keywords:
carbon; climate change; forest degradation; Landsat; REDD+; selective logging; Venezue-
lan Amazon; TerraAmazon; Imataca Forest Reserve
1. Introduction
More than 400 million hectares (ha) of natural tropical forests have been designated
as production forests globally [
1
–
3
]. Moreover, about 40% of sawn wood traded annually
in tropical regions has an origin in natural forests [
4
], often under a “selective logging”
approach in which large trees of a relatively low number of tree species are harvested
in rotation cycles of 30 years on average [
2
,
5
,
6
]. With some exceptions, one of the main
features of selective logging across the tropics has been the insufficient adoption of reduced
impact methods with negative environmental effects on forest structure and function [
7
,
8
].
Forest degradation is a change process caused by anthropogenic and/or environ-
mental forces that result in alterations within any given forest, negatively affecting the
Remote Sens. 2021,13, 1435. https://doi.org/10.3390/rs13081435 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2021,13, 1435 2 of 25
structure and or function of the stand and site, and thereby lowering the capacity to sustain
a continuous supply of products and/or services [
9
]. Selective logging, a major degrada-
tion driver in tropical forests, may account for at least half of the total carbon emissions
coming from tropical forest degradation, representing ~6% of total tropical greenhouse
gas emissions [
10
,
11
]. Logging can cause significant losses in the aboveground carbon
in tropical forests up to 10–50% [
12
,
13
], and in some cases can add up to 123% more
forest-area damage above what has been reported for deforestation alone [
14
]. Moreover,
in tropical regions of Latin America and Asia, selective logging has been responsible for
more than 70% of the area of total forest degradation [
15
], and is often consider a preamble
to deforestation processes [
16
–
18
]. Consequently, in recent years, forest degradation has
become a major topic of discussion within the international scientific community as a
factor of great importance in the global carbon cycle [
11
,
19
,
20
]. This process has been
addressed in the United Nations Framework Convention on Climate Change (UNFCCC)
since the Bali agreement in 2007 (CP. 13), when the concept of reducing emissions from
deforestation (RED) was expanded to REDD+ to include forest management activities [
21
],
and currently with the so called Nationally Determined Contributions to the UN Paris
Climate Agreement [22].
Venezuela has a long history of natural forest management under long-term concession
contracts that started in the early 1970s [
23
,
24
]. By 1992, almost 3.2 million ha were allocated
to more than 30 forest management units (FMUs) and had management plans approved by
the national government [
25
]. However, a remarkable decline in the use of forests as reliable
sources of timber has been evident in the last 30 years. According to the last available
official data from 2018, only about 2.5% of the wood legally consumed in the country came
from FMUs in an estimated area of 246,313 ha of forests with formal management plans
mostly in the Amazon region of the country [26].
Forests in the Venezuelan Amazon account for at least 83% of the national forest
cover [
27
]. In this region, forests have been managed via various legal protection schemes,
from strictly protected areas such as national parks to others with sustainable use objectives
as the case of forest reserves, forest lots and forest areas under protection [
28
,
29
]. One of
the most important reserves in the region is the Imataca Forest Reserve (IFR), which harbor
important levels of biodiversity [
30
]. In addition, different studies have estimated that
the region represents one of the most carbon-rich areas in Venezuela with an average of
205
±
15 Mg C ha
−1
in the aboveground biomass (AGB) [
31
,
32
]. At the same time, the
IFR is considered a hot spot of deforestation [
33
] and was recently labelled as one of the
deforestation fronts of the tropical belt [
34
]. In the last two decades, approximately 1144 ha
of forests have been lost every year due to mining activities (55.5%), land use change for
livestock (26.5%) and agriculture expansion (17.9%) [
35
]. Yet, from the close to 4 million ha
that IFR covers, about 97% is still covered by different forest-types, mostly dominated by
lowland “terra firme” forests [35].
Selective logging, via legal, yet mostly unplanned conventional timber harvesting
operations was formally authorized in IFR around the 1980’s decade and has become a
major factor of anthropogenic disturbance since then [
35
]. In a standard logging operation
in this region, large trees with diameters at breast height (dbh) > 40 cm and with an average
height between 20
−
30 m from a few commercially valuable species are harvested, where
for every tree harvested close to 11 additional trees can be severely affected [
36
]. Moreover,
without adequate planning, the impacts of these operations can double the background
rates of tree mortality when compared to unlogged stands while also causing a significant
reduction in aboveground carbon [13].
To quantify carbon emissions caused by forest degradation as a result of selective log-
ging, two analytical techniques can be used: the first combines logging rates, management
plans, and high-resolution imagery for activity data (AD) and the gain/loss approach for
emission factors (EFs) [
10
,
37
]. The second combines remote detection of medium-resolution
images for AD and an assessment of the changes in carbon stocks for EFs [
38
,
39
]. For the
second technique, the AD can be obtained by a direct method, identifying and mapping

Remote Sens. 2021,13, 1435 3 of 25
canopy damage [
40
–
45
], or by mapping canopy damage in combination with intact forests
and regeneration patches [46,47]. Using an indirect method that implies identifying selec-
tive logging in the images, and the addition of buffer areas via geographic information
system (GIS) tools, the area of degraded forests due to logging can be quantified [
18
,
48
,
49
].
With this method, the intact forest concept and evaluation criteria are applied to catego-
rize intact and non-intact forests, which discriminate against forests with different carbon
stocks [
16
,
46
]. The EFs can then be obtained using data collected from field inventories mea-
sured before and after logging, with the general change in the carbon reserves calculated
using the difference between these two measurements [39].
In this study, we propose an analytical approach based on a Landsat time
series [40,41,50,51],
developed for the Monitoring System of Deforestation of the Amazon (TerraAmazon) [
52
].
One of the main goals of our approach is to produce a local-based computing analytical
approach capable of functioning under conditions of limited connectivity. We believe this
to be an important advantage to conduct an assessment of forest degradation produced
by logging in Venezuela’s Imataca Forest Reserve (IFR) for a 30-year period between
1985 and 2015 under different conditions. Overall, because of the general lack of reduced
impact logging guidelines being applied [
29
], we expected that despite a relatively low
logging intensity typically between 10 to 12 m
3
ha
−1
[
36
,
53
,
54
], compared to an average of
20 m3ha−1
for the Amazon [
55
], carbon losses could be significant. We put our results in
the context of other studies where the effects of logging on carbon have been addressed.
2. Materials and Methods
2.1. Study Area
Imataca Forest Reserve (IFR), located in southeastern Venezuela between Bolívar and
Delta Amacuro states, was officially created in 1961 and has a total area of 3,821,900 ha,
which represents 8.1% of the total area of the Venezuelan Amazon (~46.9 M ha) (Figure 1a).
The reserve has been divided into 23 different management units distributed across three
major zones (north, central, and south) (Figure 1b). 90% of these units with extensions
between 120,000 and 340,000 ha have been designated for permanent forest production,
while the rest is officially allocated for mining activities and/or conservation of biodiversity.
Between 1985 and 2012, about half of the timber-production areas at IFR was managed
under a private concession model where national government granted management rights
after an official management plan was approved with cutting cycles ranging from 25 to
40 years [
29
]. In recent years, with the enactment of the Forests and Forest Management
Law of 2008, a policy shift began with regards to how forest management should be
planned and applied in Venezuela. Along with government agencies and the newly created
National Forest Company (ENAFOR), guidelines for developing new forest management
plans were put in place to gradually shift from the model of private concessions to a more
government-dominated approach. At present, the company supervises the management for
all production forests in the country and is directly responsible for an active management
operation in the Imataca Forest Reserve.
The IFR has a northeast-southwest pattern in the distribution of precipitation from
1000 mm to 3000 mm per year approximately. The average annual temperature is around
25 and 27
◦
C, evapotranspiration ranges from 1250 mm to 1400 mm per year. Overall, in
this area we find lowland tropical humid forests, seasonal evergreen forests, deciduous
forests and swamp forests. From the standpoint of species diversity around 2800 species of
plants, 450 species of birds, 153 species of mammals, 90 species of reptiles, 62 species of
amphibians and 242 species of fish have been identified [30].

Remote Sens. 2021,13, 1435 4 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of 25
The management Unit V located in the central region of IFR was selected as a study
case. The unit has a total area of 180,000 ha and was first granted management rights to
the company Industria Técnica de Maderas C.A (INTECMACA) in 1982. The unit was
originally divided into 25 logging compartments of an approximate equal area each ac-
cording to a 25-year cutting cycle. One component of our work is based on pre- and post-
logging inventories conducted at seven of these compartments namely: research plot I (RP
I), research plot II (RP II), experimental development plot (EDP), compartment 1 (C1),
compartment 2 (C2), compartment 3 (C3) and compartment 4 (C4). After operations were
halted by the end of 1990’s, the management was later transferred in 2012 to the Empresa
Nacional Forestal (ENAFOR), where two additional compartments were allocated for tim-
ber harvest: Santa Maria I (STM I) and Santa Maria II (STM II) (Figure 1c).
Figure 1. (a) Relative location of the Imataca Forest Reserve at the national scale and within the
Venezuelan Amazon region); (b) relative location of Unit V within Imataca Forest Reserve’s man-
agement units; (c) location of the logging compartments used in this study.
Depending on the characteristics of the overall planning process and the harvest in-
tensity, selective logging at Unit V can be classified into three types [18]. First, unplanned
conventional logging (CL) was characteristic of RP I, RP II and EDP between 1985 and
1988 without a formal management plan. Secondly, planned managed logging (ML) oc-
curred in two ways: in the first (ML1), a pre-commercial inventory of trees was carried
Figure 1.
(
a
) Relative location of the Imataca Forest Reserve at the national scale and within the
Venezuelan Amazon region); (
b
) relative location of Unit V within Imataca Forest Reserve’s manage-
ment units; (c) location of the logging compartments used in this study.
The management Unit V located in the central region of IFR was selected as a study
case. The unit has a total area of 180,000 ha and was first granted management rights
to the company Industria Técnica de Maderas C.A (INTECMACA) in 1982. The unit
was originally divided into 25 logging compartments of an approximate equal area each
according to a 25-year cutting cycle. One component of our work is based on pre- and
post-logging inventories conducted at seven of these compartments namely: research plot
I (RP I), research plot II (RP II), experimental development plot (EDP), compartment 1
(C1), compartment 2 (C2), compartment 3 (C3) and compartment 4 (C4). After operations
were halted by the end of 1990’s, the management was later transferred in 2012 to the
Empresa Nacional Forestal (ENAFOR), where two additional compartments were allocated
for timber harvest: Santa Maria I (STM I) and Santa Maria II (STM II) (Figure 1c).
Depending on the characteristics of the overall planning process and the harvest
intensity, selective logging at Unit V can be classified into three types [
18
]. First, unplanned
conventional logging (CL) was characteristic of RP I, RP II and EDP between 1985 and 1988
without a formal management plan. Secondly, planned managed logging (ML) occurred in
two ways: in the first (ML1), a pre-commercial inventory of trees was carried out, followed

Remote Sens. 2021,13, 1435 5 of 25
by a general planning of logging roads and landing sites. These activities were part of a
more detailed management plan in which each compartment should be-in theory-logged
every year. This method was applied in C1, C2, C3 and C4 between 1990 and 1995 [
56
]. The
second ML method (ML2) is in many ways similar to the ML1 case, especially with regards
to how logging operations were applied. However, a major difference is that planning
was organized at the landscape scale, thus watersheds and small-watersheds were used
as management units. Two additional features that were also unique for this approach
were that, on one hand, commercial trees were spatially mapped to facilitate planning of
logging roads and landings. On the other, the minimum harvest diameter was modified
from a common threshold of 40 cm for all species to dbh > 50 cm for high-wood density
(WD) species, 60 cm for medium WD and 70 cm low WD species. This method was used in
the case of STM I and STM II between 2012 and 2015 [
57
]. The mean area of each logging
compartment was 2640 ±848.5 ha (Mean ±SD).
2.2. Landsat Time Series
Fifty (50) Landsat 4, 5, 7 and 8-time series images were used, corresponding to route
233 and row 55 (Table 1). These were obtained from the collection of the US Geological
Survey (http://glovis.usgs.gov/ (accessed on 10 March 2021)), with a processing level L1T.
The time period for these datasets was selected approximately between one and two years
after logging occurred, as rapid canopy closure after disturbance and lower understory
revegetation may inhibit logging detection [41,58,59].
Table 1. General description of the Landsat time series used in this study.
Year Landsat
Sensor
Day and
Month Year Landsat
Sensor
Day and
Month Year Landsat
Sensor
Day and
Month
1986 TM5 28-November 1992 TM4 14-Dec
1996
TM5 20-Sep
TM5 30-December TM5 22-Dec TM5 6-Oct
1987
TM5 26-July
1993
TM5 7-Jan TM5 18-Aug
TM5 12-September TM4 12-Mar TM5 22-Oct
TM5 14-October TM4 5-Apr TM5 9-Dec
TM5 15-November TM5 7-Jun
1997
TM5 22-Aug
1988
TM5 7-April TM5 17-Dec TM5 23-Sep
TM5 10-June
1994
TM5 23-Mar TM5 25-Oct
TM5 12-July TM5 11-Jun 2013 ETM+ 13-Oct
TM5 6-November TM5 14-Aug 2014 ETM+ 29-Aug
1989 TM5 22-December
1995
TM5 27-Apr 2015 ETM+ 4-Jan
1990 TM4 8-February TM5 1-Aug OLI 9-Sep
TM5 23-May TM5 2-Sep 2016 OLI 27-Sep
1991
TM5 16-June TM5 4-Oct OLI 13-Oct
TM5 5-July
1996
TM5 18-Jul 2017 OLI 23-Apr
TM5 25-October TM5 3-Aug OLI 29-Aug
1992 TM4 17-September TM5 5-September
2.3. Field Data
Two independent datasets composed by information collected from temporary and
permanent ground plots were used in support of the analysis:
•
The first group was obtained from INTECMACA inventories conducted between 1986
and 1995. Two groups of permanent plots were established in RP II: four plots of
0.5 ha (100
×
50 m) located in unlogged forests and four plots of 2 ha (500
×
40 m)
in logged forests. The data include all living trees with diameters at breast height
(dbh) > 10 cm [13,60].
•
The second set of data was obtained from ENAFOR inventories conducted between
2012 and 2015. A group of temporary and permanent plots was systematically estab-
lished in the STM I and STM II logging subunits; a total of 65 plots of 1 ha
(1000 ×10 m)
with subplots of 0.01 ha (10 m
×
10 m) were measured in the pre- and post-logging

Remote Sens. 2021,13, 1435 6 of 25
periods [
61
,
62
] (Figure A1 in Appendix A). In all cases, a complete taxonomic identifi-
cation was made to every individual to account for species composition and diversity.
In addition, official reports from the logging companies were used to collect data on
the number and size of the harvested trees (diameter, height and species) [
63
]. Total volume
per tree was estimated using the Smalian scale formula (cm
3
), so total volume of wood
harvested at the compartment level could be estimated.
2.4. Analytical Approach
Our analytical approach consisted of five different phases as follows: we first mapped
selective logging, followed by a validation process of the resulting maps. The third phase
consisted of the construction and validation of the forest degradation maps, followed by
the estimation of aboveground biomass (AGB) and carbon to close with the estimation of
committed carbon emissions (CCE) (Figure 2).
Remote Sens. 2021, 13, x FOR PEER REVIEW 6 of 25
• The first group was obtained from INTECMACA inventories conducted between
1986 and 1995. Two groups of permanent plots were established in RP II: four plots
of 0.5 ha (100 × 50 m) located in unlogged forests and four plots of 2 ha (500 × 40 m)
in logged forests. The data include all living trees with diameters at breast height
(dbh) > 10 cm [13,60].
• The second set of data was obtained from ENAFOR inventories conducted between
2012 and 2015. A group of temporary and permanent plots was systematically estab-
lished in the STM I and STM II logging subunits; a total of 65 plots of 1 ha (1000 × 10
m) with subplots of 0.01 ha (10 m × 10 m) were measured in the pre- and post-logging
periods [61,62] (Figure A1 in Appendix A). In all cases, a complete taxonomic identi-
fication was made to every individual to account for species composition and diver-
sity.
In addition, official reports from the logging companies were used to collect data on
the number and size of the harvested trees (diameter, height and species) [63]. Total vol-
ume per tree was estimated using the Smalian scale formula (cm3), so total volume of
wood harvested at the compartment level could be estimated.
2.4. Analytical Approach
Our analytical approach consisted of five different phases as follows: we first
mapped selective logging, followed by a validation process of the resulting maps. The
third phase consisted of the construction and validation of the forest degradation maps,
followed by the estimation of aboveground biomass (AGB) and carbon to close with the
estimation of committed carbon emissions (CCE) (Figure 2).
Figure 2.
Analytical approach used to analyze forest degradation produced by selective logging in the Imataca Forest Reserve,
Venezuelan Amazon. Green boxes depict all activities related to the source of the data used (i.e., Landsat imagery and forest
inventory plots). Blue boxes summarize the analytical components of the study, while gray ones show the mapping outputs.
Yellow boxes refer to the final estimations of aboveground biomass (AGB), emission factors (EFs) and emissions.
2.4.1. Mapping Selective Logging
The TerraAmazon system was used to map selective logging. This system involves a
configuration that includes the creation of a PostgreSQL database, definition of the concep-
tual model, access control, phase control, project and control rules, definition of classes,
definition of the control rules, and the definition of the control area [
64
]. The Landsat time
series datasets were then exported to generate a linear spectral mixing model (LSMM) for

Remote Sens. 2021,13, 1435 7 of 25
each image to be used in the detection of selective logging [
14
,
43
,
45
,
48
,
49
]. The LSMM
uses the red (0.63–0.69 µm), near infrared (0.76–0.90 µm) and mid-infrared (1.55–1.75 µm)
bands from Landsat 4, 5, 7 and 8 [
44
], from which the samples of soil cover, vegetation and
shade can be extracted to estimate the proportions in each pixel and in their respective
images (Equation (1)). The soil fraction image was then used to estimate the area affected
by selective logging [43,44,48,49,65].
ri=a∗vegi+b∗soili+c∗shadowi+ei(1)
where
ri= is the response of the pixel in band i;
a,b, and c= proportions of vegetation, soil, and shade, respectively;
vege
i
,soil
i
and shadow
i
= spectral responses of the components of vegetation, soil and shade,
for each band respectively;
ei= is the error in band i.
A cloud mask and cloud shadow were applied to each image of the soil fraction, using
thresholds of the minimum and maximum values of the blue band (0.45–0.51
µ
m) to detect
shadows and the infrared thermal band (10.60–11.19
µ
m) to detect clouds [
64
,
66
]. Selective
logging was detected by means of a binary classification of areas with and without selective
logging based on the processed images of the soil fraction [
44
,
49
]. A value of zero was
assigned to those pixels with soil fractions lower than 37% (i.e., areas without evidence of
selective logging), and a value of one corresponded to the pixels with soil fractions between
37 and 100% (i.e., areas with signs of selective logging). A decision tree algorithm was then
used, for which the 37% limit was statistically defined based on 150 points visually selected
from a processed soil fraction image [
49
]. Once the binary images were obtained, these
were added to generate the mapping of selective logging.
2.4.2. Validation of the Selective Logging Maps
To determine the quality and degree of agreement between the mapping of selec-
tive logging and field conditions, maps were validated via the comparison with an ex-
ternal source that is considered a realistic representation of the characteristics on the
ground [67–70].
Thus, we applied a systematic sampling approach [
50
,
71
] to 36 blocks
of 100 ha (1 km
2
) [
72
], which represented approximately 11% of the study area. This
sampling technique allowed us to precisely and quickly estimate the error of the analy-
sis [
67
,
73
]. By using a visual on-screen interpretation of a minimum cartographic area of
1 ha in each sampling block [
72
] we can generate logged and unlogged forest datasets
(Figure A2
in
Appendix A).
These were considered the ground-truth data and were then
used to analyze the thematic quality of the selective logging map. A confusion matrix was
generated and the errors of omission and commission with the level of global accuracy
were also calculated.
To confirm the logged classification, a spatially non-localized analysis was used
by comparing the proportions of the area in each sample block in the selective logging
map and in the ground-truth data. These proportions were compared via simple linear
regression, with the rationale that if the mapping of selective logging and ground-truth
data were similar, adjustment values would be high and the coefficient (R
2
), would be
close to 1 [
67
,
74
]. This method has been widely used in the mapping of forest fires [
75
,
76
],
analyses of land use [74] and deforestation [77].
2.4.3. Construction and Validation of the Forest Degradation Maps
Using an indirect method, mapping of forest degradation was performed for each
logging compartment. In doing so, we estimated the average radius between log landings
in each of the soil fraction images as proposed by Monteiro et al. [
48
]. In our case, this value
was 600 meters (m), so 300 m was used as a threshold to estimate the approximate area of
forest degradation caused by logging, and a square buffer was applied to the mapping of

Remote Sens. 2021,13, 1435 8 of 25
selective logging using GIS tools. Maps of forest degradation were validated by comparing
these with the area logged reported in the management plans, which allowed for the
calculation of commission and omission errors [68].
2.4.4. Estimation of Aboveground Biomass (AGB) and Carbon
Forest inventory data for pre- and post-logging conditions were used to estimate AGB
and carbon, following the approach by the Global Observation of Forest and Land Cover
Dynamics panel (GOFC-GOLD) [
46
] for the establishment of REDD+ projects. The AGB
per tree in each ground plot was estimated using the pantropical allometric regression
from Chave et al. [
78
] (Equation (2)). All the estimates were generated for each plot and
the estimations were scaled to 1 ha when necessary. Values of aboveground carbon (Mg C
ha−1) were assumed to be 50% of AGB [79].
(AGB)est =exp−1.803 −0.976E+0.976ln(ρ)+2.673 ln(D)−0.0299(ln(D))2(2)
where:
AGB = is the aboveground biomass of the individual trees expressed in kilograms
(kg); E= is a water stress factor that shows an important covariance with the diameter-
height ratio in tropical trees and includes information on seasonal temperature (ST) and the
climatic water deficit (CWD). Based on the geographic location of each plot, Eand CWD
were derived from a 2.5 arc-minute resolution raster file available at http://chave.ups-tlse.
fr/pantropical_allometry.htm. (accessed on 10 March 2021);
ρ
= is the density of the wood
in g cm−3, with data assigned for each taxonomic group from the pantropical database of
Zanne et al. [80] and Chave et al. [81]; D = is the diameter of each tree in cm.
2.4.5. Estimation of Committed Carbon Emissions (CCE)
A stock-difference method was used to estimate emissions related to selective logging
following the 2019 refinement of the 2006 IPCC guidelines [
39
]. In doing this, the following
assumptions were considered:
•
To simplify the carbon accounting process, the committed emissions approach was
used, in which all carbon removed is assumed to be emitted at the time of its removal
via logging [37,39].
•
Emissions were estimated in each compartment by multiplying the area affected by
degradation (activity data) with the difference of the carbon content in the pre- and
post-logging period (emission factor) [39].
•
The different harvesting activities were classified in the selective logging map into: log
landings, caused when the forest is cleared for the purpose of temporary log storage
before final transportation; logging roads, built to transport timber from log landings
to sawmills; and logging gaps, created by tree felling and skid trails, resulting in
damage or death to other standing trees [
18
,
46
]. These categories were associated with
the emissions in each compartment to determine the overall emission contribution for
each activity.
•
Using the reported values of timber extracted from each compartment, carbon losses
from logging were estimated by calculating the equivalent carbon of the volume of
extracted roundwood, which considered the wood specific density to obtain AGB. A
factor of 0.5 was used to estimate the amount of carbon [79,82].
•
To adapt our data categories to the gain and loss method proposed by the IPCC [
39
] and
used in Pearson et al. [
37
] and Ellis et al. [
10
], we linked the data to three main sources of
emissions as follows: (1) roundwood extracted and felling emissions; (2) logging gaps
and skidding emissions; (3) log landings and roads with hauling emissions.
•
We express carbon emissions in three ways: (1) emissions per area (Mg ha
−1
) by divid-
ing all emissions from each compartment using the estimated area of degradation [
83
];
(2) emissions per volume of harvested roundwood (Mg m
−3
), dividing all emissions
from logging in each compartment by the total volume extracted [
10
]; (3) the carbon

Remote Sens. 2021,13, 1435 9 of 25
impact factor (CIF) (Mg Mg
−1
) also called “mean carbon export ratio” [
82
], represents
the emissions of each compartment relative to the emissions of the total volume of
extracted roundwood.
3. Results
3.1. Accuracy of Maps of Selective Logging and Forest Degradation
The results from the error matrix are presented in terms of area proportions. The row
totals of the error matrix (Sum = 0.113) provide what was represented in the ground-truth
samples in each class and constitutes the total proportion of selective logging. Conversely,
the column totals provide the estimated proportions according to the ground-truth data,
and this was estimated in 0.013 for the logged class. Multiplying this value by the total
number of pixels on the map (i.e., 330,386), the result is 4158 pixels, or approximately 374 ha.
The area of agreement between the mapping of selective logging and the ground-truth
data was 2452 pixels or close to 221 ha. There was an underestimation of 1706 pixels or
around 154 ha that was confused with the unlogged class. On the other hand, the logged
class had the highest user error (commission), with a proportion of 0.146, and a producer
error (omission) of 0.4103 because greater proportions of areas were included and excluded
in this class, respectively (Table 2). The proportion of these errors was observed in nine of
the 36 blocks that were used in the validation, three for the case of ENAFOR and six for
INTECMACA (Figure A3 in Appendix A). The global precision of the selective logging
map was 0.943.
Table 2.
The error matrix, omission and commission errors, and overall precision in terms of
estimated area proportion.
Ground Truth (Proportion)
Class Logged Unlogged Total
Logged 0.007 0.001 0.009
Unlogged 0.005 0.099 0.105
Total 0.013 0.101 0.113
Error
Commission 0.146 0.049
Omission 0.410 0.013
Global Precision 0.943
The values of the ground-truth proportions and those of the mapping of selective
logging showed an overall good fit. The proportions presented high similarity, as indicated
by a high value in the coefficient of determination after a simple linear regression
(R2= 0.82)
(Figure A4 in Appendix A). In relation to the area of forest degradation obtained from
the map, this represented approximately 13.6% (24,484 ha) of the entire area of Unit V.
Analyzing this area for each compartment, six had commission errors, ranging from
+1.6% in compartment C2 (71 ha) to +14.7% in compartment RP II (325 ha), while three
compartments had errors of omission, ranging from
−
0.3% in compartment EDP (5 ha) to
−17.7% in compartment C1 (292 ha) (Figure A5 in Appendix A).
3.2. Mapping Selective Logging at Imataca Forest Reserve
Figure 3shows the full process to detect and outline selective logging activities for the
case of the EDP compartment (i.e., experimental development plot), logged during the first
three months of 1988.

Remote Sens. 2021,13, 1435 10 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 10 of 25
from the map, this represented approximately 13.6% (24,484 ha) of the entire area of Unit
V. Analyzing this area for each compartment, six had commission errors, ranging from
+1.6% in compartment C2 (71 ha) to +14.7% in compartment RP II (325 ha), while three
compartments had errors of omission, ranging from −0.3% in compartment EDP (5 ha) to
−17.7% in compartment C1 (292 ha) (Figure A5 in Appendix A).
3.2. Mapping Selective Logging at Imataca Forest Reserve
Figure 3 shows the full process to detect and outline selective logging activities for
the case of the EDP compartment (i.e., experimental development plot), logged during the
first three months of 1988.
The total area directly affected by selective logging-related activities was around 2535
ha, from which compartment C2 had the largest area with 539 ha (21.2% of total area) and
compartment C1 the lowest with 89 ha (3.5%), with an overall mean area of forest degra-
dation of 282 ± 124 ha (±SD). The number of log landings built was, on average, 2.1 ± 0.6
(±SD) per 100 ha, with a total area of 2926 ± 496 m2 and an area affected by logging of 59 ±
16 m2 ha−1. In terms of the extension of logging roads, we estimated an average of 16 ± 2.3
m ha−1 in length, and a disturbed area of roads close to 583 ± 109 m2 ha−1. Finally, the av-
erage area affected in logging gaps was estimated at 174 ± 111 m2 ha−1 (Table 3). Spatial
patterns of the different areas affected by logging were highly heterogeneous across all
compartments. More regular arrangements, especially in terms of road building and dis-
tribution, seem clearer in the case of the oldest logging operations (e.g., RP1, RP2, EDP) in
contrast with the most recent logged areas (i.e., STM I, STM II) (Figure 4).
Figure 3. Stepwise process for the mapping of selective logging for the case of compartment EDP. Composition 453 (a)
which was used to generate the soil fraction of the linear spectral mixing model (LSMM), a cloud mask (b) was then
applied, and later a decision tree classification was used to obtain a binary image of the logged and unlogged areas (c).
This image was then categorized to analyze the three main logging activities (d).
Figure 3.
Stepwise process for the mapping of selective logging for the case of compartment EDP. Composition 453 (
a
) which
was used to generate the soil fraction of the linear spectral mixing model (LSMM), a cloud mask (
b
) was then applied, and
later a decision tree classification was used to obtain a binary image of the logged and unlogged areas (
c
). This image was
then categorized to analyze the three main logging activities (d).
The total area directly affected by selective logging-related activities was around
2535 ha, from which compartment C2 had the largest area with 539 ha (21.2% of total
area) and compartment C1 the lowest with 89 ha (3.5%), with an overall mean area of
forest degradation of 282
±
124 ha (
±
SD). The number of log landings built was, on
average,
2.1 ±0.6 (±SD) per 100 ha,
with a total area of 2926
±
496 m
2
and an area af-
fected by logging of 59
±
16 m
2
ha
−1
. In terms of the extension of logging roads, we
estimated an average of 16
±
2.3 m ha
−1
in length, and a disturbed area of roads close
to
583 ±109 m2ha−1.
Finally, the average area affected in logging gaps was estimated at
174 ±111 m2ha−1
(Table 3). Spatial patterns of the different areas affected by logging were
highly heterogeneous across all compartments. More regular arrangements, especially
in terms of road building and distribution, seem clearer in the case of the oldest logging
operations (e.g., RP1, RP2, EDP) in contrast with the most recent logged areas (i.e., STM I,
STM II) (Figure 4).

Remote Sens. 2021,13, 1435 11 of 25
Table 3.
General description of the area disturbed by construction of log landings, roads, and logging gaps in each
compartment by logging approach in Unit V, Imataca Forest Reserve, Venezuela.
Compartments a
Unplanned Conventional
Logging
(CL)
Planned Logging 1
(ML1)
Planned Logging 2
(ML2)
RP I RP II EDP C 1 C 2 C 3 C 4 STM I STM II
Loglandings
Number of log landings
per 100 ha logged 1.7 2.6 2.5 2.1 1.2 1.1 2.7 2.1 2.9
Mean area of log
landings (m2)±SD
2690
±1183
2317
±935
2825
±1073
2614
±1018
3485
±1774
3650
±1563
3324
±1345
3133
±2139
2297
±925
Log landings area per ha
logged (m2ha−1)46.1 59.7 69.4 55.7 41.2 40.4 90.4 64.6 65.8
Logging roads
Length of logging road
per hectare logged
(m ha−1)
14 19 20 13 14 16 17 16 17
Logging road area per
hectare logged
(m2ha−1)
404 617 769 448 605 642 625 601 533
Logginggaps
Area of gaps per total
area logged (m2ha−1)184 278 254 41 378 64 106 153 105
a
Compartments: RP I: Research plot I; RP II: Research plot II; EDP: Experimental Development Plot; C1: Compartment 1; C2: compartment
2; C3: Compartment 3; C4: Compartment 4; STM I: Santa Maria I; and STM II: Santa Maria II.
3.3. Analysis of Forest Degradation
The total area of forest degradation caused by selective logging activities was estimated
in 24,484 ha. The average area of forest degradation by compartment was 2720
±
911 ha
(Mean
±
SD), with a maximum value of 3457 ha for compartment C4 and a minimum
of 1360 ha in compartment C1 (Figure 4; Table 4). In total, 65,036 trees were harvested
in all compartments with a total volume of 247,034 m
3
, for a mean logging intensity of
2.8 ±1.2
trees ha
−1
or 10.5
±
4.6 m
3
ha
−1
. Average ground area damaged per tree logged
was estimated in 334
±
121.9 m
2
, equivalent to 91
±
41.9 m
2
for each cubic meter of timber
harvested. The mean of the total area affected by logging represented 8.2
±
1.8 % of the
entire degraded area, 0.6
±
0.2 % of this area corresponded to the construction of log
landings, 5.8 ±1.1 % to roads, and 1.7 ±1.1 % to logging gaps (Table 4).
3.4. Aboveground Carbon Density
The values of forest carbon density for pre- and post-logging varied in the different
compartments, with the pre-logging ranging from 268 Mg C ha
−1
(C4 compartment) to
312 Mg C ha
−1
(C2 compartment), with an average of 308
±
76 (
±
SD) Mg C ha
−1
. After
logging, aboveground carbon density varied from 222 Mg C ha
−1
(RP I compartment) to
263 Mg C ha
−1
(STM I compartment), with an average of 229
±
30 (
±
SD) Mg C ha
−1
. The
average difference between the two conditions was estimated in 35% (Table 5).

Remote Sens. 2021,13, 1435 12 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 11 of 25
Figure 4. Mapping of selective logging activities and estimation of forest degradation in each compartment. Maps in (a),
(b,c) represent the unplanned conventional logging (CL). Figures from (d–g) are representative of the planned managed
logging 1 (ML1). Finally, (h,i) reflects the second planned logging approach (ML2). Codes for logging compartments: RP
I: Research plot I; RP II: Research plot II; EDP: Experimental Development Plot; C1: Compartment 1; C2: compartment 2;
C3: Compartment 3; C4: Compartment 4; STM I: Santa Maria I; and STM II: Santa Maria II.
Table 3. General description of the area disturbed by construction of log landings, roads, and logging gaps in each com-
partment by logging approach in Unit V, Imataca Forest Reserve, Venezuela.
Compartments a
Unplanned Conventional
Logging
(CL)
Planned Logging 1
(ML1)
Planned Logging 2
(ML2)
RP I RP II EDP C 1 C 2 C 3 C 4 STM I STM II
Log landings
Number of log landings
per 100 ha logged 1.7 2.6 2.5 2.1 1.2 1.1 2.7 2.1 2.9
Figure 4.
Mapping of selective logging activities and estimation of forest degradation in each compartment. Maps in (
a
),
(
b
,
c
) represent the unplanned conventional logging (CL). Figures from (
d
–
g
) are representative of the planned managed
logging 1 (ML1). Finally, (
h
,
i
) reflects the second planned logging approach (ML2). Codes for logging compartments: RP I:
Research plot I; RP II: Research plot II; EDP: Experimental Development Plot; C1: Compartment 1; C2: compartment 2; C3:
Compartment 3; C4: Compartment 4; STM I: Santa Maria I; and STM II: Santa Maria II.
3.5. Committed Carbon Emissions (CCEs) from Selective Logging
Mean CCEs per volume harvested was 2.9
±
1.1 (mean
±
SD) Mg C m
−3
, those from
construction of logging roads were 1.9
±
0.7 Mg C m
−3
(64% of all emissions), logging gaps
~0.5
±
0.4 Mg C m
−3
(17% of all emissions), logged roundwood with
~0.3 ±0.01 Mg C m−3
(10% of all emissions), and those from log landings ~0.2
±
0.1 Mg C m
−3
(9% all emissions).
The maximum value was 4.4
±
1.1 Mg C m
−3
in the EDP compartment, and the minimum
was 1.7
±
0.4 Mg C m
−3
from STM I compartment (Figure 5A). Mean CCEs per area for
all selective logging activities was 64.2
±
22.2 Mg C ha
−1
. Logging road construction
accounted for 43.9
±
13.7 Mg C ha
−1
(68% of all emissions), logging gaps represented
11.3
±
9.2 Mg C ha
−1
(18% of total), log landings accounted for 5.9
±
2.0 Mg C ha
−1
(9%), and logged roundwood represented 3.2
±
1.4 Mg C ha
−1
(5% of all emissions). The

Remote Sens. 2021,13, 1435 13 of 25
maximum value was estimated at 91.4
±
24.4 Mg C ha
−1
in the compartment RP II, while
the minimum was 23.8 ±4.1 Mg C ha−1from the STM I compartment (Figure 5B).
Table 4.
General characteristics of the harvesting operations and the area affected in each compartment classified by activity
and logging approach.
Compartments a
Unplanned Logging
(CL)
Planned Logging 1
(ML1)
Planned Logging
2 (ML2)
RP I RP II EDP C 1 C 2 C 3 C 4 STM I STM II
Year of logging 1985 1987 1988 1990 1991 1992 1995 2012 2015
Area of forest degradation (ha) 2099 2525 1995 1360 4480 2892 3457 2571 3105
Total number of trees logged 4000 11,458 5566 6116 10,240 5125 12,249 4345 5937
Total logged volume (m3)15,433 44,207 13,650 24,566 33,804 25,695 47,073 16,639 25,967
Logging intensity 1 (trees ha−1)1.9 4.5 2.8 4.5 2.3 1.8 3.5 1.7 1.9
Logging intensity 2 (m3ha−1)7.4 17.5 6.8 18.1 7.5 8.9 13.6 6.5 8.4
Log landings (%) 0.5 0.6 0.7 0.6 0.4 0.4 0.9 0.6 0.7
Logging roads (%) 4.0 6.2 7.7 4.5 6.1 6.4 6.3 6.0 5.3
Logging gaps (%) 1.8 2.8 2.5 0.4 3.8 0.6 1.1 1.5 1.0
Total area affected by logging (%) 6.3 9.6 10.9 5.4 10.2 7.5 8.2 8.2 7.0
Ground damage per tree
logged (m2)333 211 391 121 448 421 232 484 368
Ground damage per m3of timber
harvested (m2)86 55 160 30 136 84 60 127 84
a
Compartments: RP I: Research plot I; RP II: Research plot II; EDP: Experimental Development Plot; C1: Compartment 1; C2: compartment
2; C3: Compartment 3; C4: Compartment 4; STM I: Santa Maria I; and STM II: Santa Maria II.
Table 5.
Density of carbon in the aboveground biomass by logging compartment in pre- and post-
logging and average (including the mean ±standard deviation).
Forest C Density (Mg C ha−1)
Compartments Logging Method Pre-Logging Post-Logging Difference (%)
RP I CL 299 ±76 222 ±30 35
RP II CL 300 ±91 219 ±23 37
EDP CL 293 ±76 220 ±30 34
C 1 ML1 280 ±29 225 ±37 24
C 2 ML1 312 ±76 229 ±30 36
C 3 ML1 289 ±7 246 ±21 17
C 4 ML1 268 ±29 226 ±37 18
STM I ML2 283 ±7 263 ±21 8
STM II ML2 272 ±29 224 ±37 21
Average 308 ±76 229 ±30 35
CCEs for all nine compartments were highly variable. The CCEs expressed in terms
of the Carbon Impact Factor (CIF), ranged between a minimum of 5.0 and a maximum
of 13.2 Mg C Mg C
−1
for compartments C1 and EDP respectively, and their baseline was
8.6 ±3.7 Mg C Mg C−1
. On a volume basis, CCEs ranged between 1.8 and 4.4 Mg m
−3
for
the case of C1, C4, STM I and EDP compartments, and their baseline was
2.9 ±1.1 Mg m−3.
On an area basis, CCEs ranged between 21.9 and 86.1 Mg ha
−1
for STM I and RP II
compartments respectively, and their baseline was 61
±
22 Mg ha
−1
. The effect of the
intensity of logging on CCEs was very similar in all three ways of reporting emissions
(Figure 6). In terms of the logging approach, logging intensity values for the EDP and
RP I compartments, where a conventional unplanned (CL) approach was common, were
relatively low (between 6 and 8 m
3
ha
−1
), but the effects on carbon were above the baselines
of the three ways of reporting emissions. In the case of the RP II compartment, logging
intensity was high (17.5 m
3
ha
−1
) but below the CCEs baselines expressed either by

Remote Sens. 2021,13, 1435 14 of 25
volume or CIF. Planned selective logging 1 (ML1) used in compartments C1, C3 and C
4 had values below the baseline for all three ways of reporting emissions. However, the
intensities increased from 8.3 m
3
ha
−1
for compartment C3, 13.6 m
3
ha
−1
for compartment
C4 and 18.1 m
3
ha
−1
for compartment C1, while in compartment C2 this intensity was low
(7.5 m
3
ha
−1
) but its CCE values were still the highest. For planned selective logging 2
(ML2) applied in the compartment STM I, all values when reporting emissions were below
the baseline, with an intensity of 5.2 m
3
ha
−1
, while compartment STM II the CCEs values
by volume and CIF were above the baseline with an intensity of 8.4 m3ha−1(Figure 6).
Remote Sens. 2021, 13, x FOR PEER REVIEW 14 of 25
Figure 5. Estimated carbon emissions for each logging activity in the compartments. Emissions
values are expressed as carbon emitted per volume of wood extracted (Mg C m−3) (A) and per area
(Mg C ha−1) (B). Compartments: RP I: Research plot I; RP II: Research plot II; EDP: Experimental
Development Plot; C1: Compartment 1; C2: compartment 2; C3: Compartment 3; C4: Compart-
ment 4; STM I: Santa Maria I; and STM II: Santa Maria II.
CCEs for all nine compartments were highly variable. The CCEs expressed in terms
of the Carbon Impact Factor (CIF), ranged between a minimum of 5.0 and a maximum of
13.2 Mg C Mg C−1 for compartments C1 and EDP respectively, and their baseline was 8.6
± 3.7 Mg C Mg C−1. On a volume basis, CCEs ranged between 1.8 and 4.4 Mg m−3 for the
case of C1, C4, STM I and EDP compartments, and their baseline was 2.9 ± 1.1 Mg m−3. On
an area basis, CCEs ranged between 21.9 and 86.1 Mg ha−1 for STM I and RP II compart-
ments respectively, and their baseline was 61 ± 22 Mg ha−1. The effect of the intensity of
logging on CCEs was very similar in all three ways of reporting emissions (Figure 6). In
terms of the logging approach, logging intensity values for the EDP and RP I compart-
ments, where a conventional unplanned (CL) approach was common, were relatively low
(between 6 and 8 m3 ha−1), but the effects on carbon were above the baselines of the three
ways of reporting emissions. In the case of the RP II compartment, logging intensity was
high (17.5 m3 ha−1) but below the CCEs baselines expressed either by volume or CIF.
Planned selective logging 1 (ML1) used in compartments C1, C3 and C 4 had values below
the baseline for all three ways of reporting emissions. However, the intensities increased
from 8.3 m3 ha−1 for compartment C3, 13.6 m3 ha−1 for compartment C4 and 18.1 m3 ha−1
for compartment C1, while in compartment C2 this intensity was low (7.5 m3 ha−1) but its
CCE values were still the highest. For planned selective logging 2 (ML2) applied in the
compartment STM I, all values when reporting emissions were below the baseline, with
Figure 5.
Estimated carbon emissions for each logging activity in the compartments. Emissions values are expressed
as carbon emitted per volume of wood extracted (Mg C m
−3
) (
A
) and per area (Mg C ha
−1
) (
B
). Compartments: RP I:
Research plot I; RP II: Research plot II; EDP: Experimental Development Plot; C1: Compartment 1; C2: compartment 2; C3:
Compartment 3; C4: Compartment 4; STM I: Santa Maria I; and STM II: Santa Maria II.

Remote Sens. 2021,13, 1435 15 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 15 of 25
an intensity of 5.2 m3 ha−1, while compartment STM II the CCEs values by volume and CIF
were above the baseline with an intensity of 8.4 m3 ha−1 (Figure 6).
Figure 6. Committed carbon emissions (CCEs) for nine compartments in Unit V of the Imataca forest reserve, reported as
carbon impact factor (CIF) (A), volume of timber extracted (B) and area (C). The purple symbols are compartments with
unplanned conventional logging (CL), the turquoise ones with planned managed logging 1 (ML1) and the yellow with
planned managed logging 2 (ML2). Black dashed line and gray polygons show the trend and 95% confidence intervals of
a linear regression; horizontal gray dashed line show the average emissions considered as a baseline in each case.
Figure 6.
Committed carbon emissions (CCEs) for nine compartments in Unit V of the Imataca forest reserve, reported as
carbon impact factor (CIF) (
A
), volume of timber extracted (
B
) and area (
C
). The purple symbols are compartments with
unplanned conventional logging (CL), the turquoise ones with planned managed logging 1 (ML1) and the yellow with
planned managed logging 2 (ML2). Black dashed line and gray polygons show the trend and 95% confidence intervals of a
linear regression; horizontal gray dashed line show the average emissions considered as a baseline in each case.
4. Discussion
4.1. Potential of the Analytical Approach
In this study we were able to develop a reliable and useful method to analyze selective
logging via a local network using the Amazon Deforestation Monitoring System (TerraA-
mazon), with an architecture developed in a client-server environment, and with a TerraLib

Remote Sens. 2021,13, 1435 16 of 25
database which uses PostgreSQL as a Database Management System [
52
]. This analytical
approach allowed us to estimate the area of forest degradation caused by selective logging
with a high level of accuracy that subsequently help estimating the amount of carbon
emissions produced.
Our results show the potential for this method to efficiently map selective logging
automatically as shown in other studies [
44
,
48
,
49
,
84
,
85
]. With a relatively high global
precision (GP = 0.943) that is within the range recommended by the GOFC-GOLD [
46
] for
the development of forest cover monitoring maps, and slightly higher than that reported in
the Brazilian Amazon (0.92) [
49
], and a high determination coefficient (0.82) that shows a
close fit between the model and real proportions of the logging classes [
67
,
74
,
77
], we believe
this is a promising and powerful tool to study forest degradation in tropical countries with
severe connectivity limitations as the case of Venezuela.
Nonetheless, our study also confirms that detecting and mapping forest degradation
with optical remote sensing data is a complex task, because the pixels that indicate forest
degradation are an intricated mix of different land cover with diverse signals (i.e., vege-
tation, dead trees, bark, tree branches, soil, shadows, etc.) [
45
]. In addition, evidence of
logging can rapidly disappear in less than two years after logging due to canopy closure and
understory revegetation potentially limiting the overall accuracy of the method [
18
,
41
,
58
].
Mapping of forest degradation caused by selective logging yielded acceptable re-
sults, with average commission and omission errors of +7.6
±
4.5 % (mean
±
SD) and
−7.5% ±9.1
respectively, well below the uncertainty threshold of
±
20% of the estimate for
area [
46
], and of the results reported for the Brazilian Amazon (18% and 20%) [
18
] and the
entire Amazon region (12% and 32%) [40].
The use of a relatively simple GIS model where the average of the log yarding radius
was used, as in other studies [
48
,
49
], obeys to the fact that the Landsat time series for this
area of the Amazon was not robust enough. Large proportions of clouds and cloud shadows
covering the images [
86
], and the fairly low number of images available during the 1980s
and 1990s [
87
] limited the use of spectral indices specialized in forest degradation such as
the Normalized Degradation Fraction Index (NDFI) [
18
,
41
] and Continuous Degradation
Detection (CODED) [
40
], or other more specialized statistical indices such as the Forest
Degradation Index (FDI), based on a Multi Criteria Decision Analysis (MCDA) approach
using the Analytic Hierarchy Process (AHP) technique [
88
,
89
]. Indeed, a similar result was
found in a recently study on forest degradation in the Amazon using the NDFI [
40
], where
forest degradation was marginally detected in our study area, sometimes overlapping with
deforested areas.
The radius of the storage yards used by this indirect method (300 meters) to determine
the buffer areas is slightly lower than in the case of other forest-types with lower densities
of commercial tree species in the central and southern Amazon where 350 m was used as a
threshold [
48
]. Our number of 300 m is, however, higher than the 180 m buffer zone used
in a study carried out in dense tropical forests in the south-central Amazon [
49
]. Although
stands of Unit V at IFR are often dense to moderately dense with an average volume of
commercial trees of 33 m
3
ha
−1
[
57
], compared to 20 m
3
ha
−1
for transitional forests [
48
]
and the 38 m
3
ha
−1
for dense forests [
49
], we argue that an intermediate threshold value
can reflect the fact that logging intensity was overall low in this unit of the Imataca Forest
Reserve (IFR).
4.2. Selective Logging Detection
The use of Landsat images with a spatial resolution of 30-m has been a common
approach in other studies analyzing the effects of selective logging [
41
,
45
,
51
,
90
–
92
], so
these results are useful to compare the degree of agreement with our estimates. For
instance, the mean mapped size of the log landings in our study area was 2926
±
496 m
2
(mean ±SD),
17% higher than the originally planned area of 2500 m
2
(50
×
50 m) [
56
,
61
].
Despite this, the mean size mapped is within the range (1–4 pixels) of log landings detection
in the soil fraction image [48], since approximately 3.3 pixels were detected.

Remote Sens. 2021,13, 1435 17 of 25
In our study area, log landings were approximately nine times larger than those
reported in southern Brazilian Amazonia (339
±
31 m
2
) for reduced impact logging (RIL)
operations. Regarding the number of log landings, we found that for every 100 ha, a
mean of 2.1
±
0.6 landings was created, equivalent to a disturbed area of 59
±
16 m
2
ha
−1
.
These estimates largely differ from those reported for RIL in the Brazilian Amazon by
Feldpausch et al. [82],
where for every 100 ha, a higher density of log landings was created
(6.2
±
0.4), but with a much lower area disturbed of 20.8
±
1.2 m
2
ha
−1
. We interpret
these results as a reflection of the overall poor planning and practical operationalization of
logging activities in the case of IFR in comparison with RIL operations in other parts of
the Amazon.
With regards to the effects of construction of logging roads, the management plans
indicate a maximum width of 10 meters for main roads and 5 meters in the case of secondary
roads plus an extra 10 m portion at each side for shoulder and ditch purposes in both type of
roads [
56
,
61
,
93
,
94
]. Our mapping analysis shows that the mean length of all roads mapped
was 16
±
2.3 m ha
−1
and the disturbed area was 583
±
109 m
2
ha
−1
. If we consider
that all the roads were established with an average of 30 m, the equivalent disturbed
area that should have been expected would be around 483
±
69 m
2
ha
−1
, indicating a
potential overestimation of ~21%. This overestimation can be interpreted as a result of the
inaccuracies in the measurements of these areas due to the effect of the pixel size [18].
The average disturbance of logging gaps indicates a large variability in the total area
disturbed (i.e., 174
±
111 m
2
) likely a direct response of the different densities of commercial
trees that can be found in this type of forests and the subsequent effects that felling and
hauling one or more logged trees can have on the overall structure of the unlogged portion
of the forest stands [36,94].
4.3. Relationship between Logging Intensity and Degradation
We found no significant differences between the CL and ML1 harvesting modalities
(3.1 ±1.3
trees ha
−1
and 3.0
±
1.2 trees ha
−1
respectively). However, the ML2 modality was
significantly different (1.8
±
1.2 trees ha
−1
), likely a consequence of the increase made to the
minimum harvest diameters (MHD) for this compartment [
57
]. Overall, logging intensity was
relatively low and similar to other management units in the IFR [
53
,
54
]. Compared to other
areas in the Amazon, logging intensity at IFR is lower (e.g., 4.4 trees ha
−1
—
Jackson et al. [95];
4.5 trees ha
−1
—Johns et al. [
96
]; 6.4 trees ha
−1
—Verissimo et al. [
97
]), which can be explained
by the differences in species composition and the abundance of commercial species among
these areas.
Considering the different approaches of selective logging applied in our study case,
the proportion of aboveground biomass (AGB) affected by logging was 35
±
1.7% for the
CL modality, below the 60% reported for other conventional logging cases in Amazonian
forests [
18
,
59
]. In the case of planned logging, the average AGB damaged was 24
±
8.5%
for ML1 and 14.5
±
9.7% for the ML2 case, above and below respectively compared with
an overall 20% reported for this type of logging [
18
,
59
]. In general, all three modalities
of logging were within the ranges of effects to living biomass (10–46%) found in other
studies [97–99].
Of the total area of forest degradation (24,484 ha), the mean area affected by selective
logging that was detected by the direct method was 8.2
±
1.8%, slightly lower than the
10.2
±
1.2% [
100
] and 13
±
4.5 (SD) % [
82
], reported for the eastern and southern Brazilian
Amazonia respectively, where reduced impact techniques were applied.
Logging disturbances were not homogeneous in each compartment and within the
different types of logging. In some cases, however, by increasing the minimum harvest
diameter (MHD), the logging intensity declined but not the proportion of area disturbed,
such as the case of ML2 where MHDs were stratified by specific wood density. A potential
explanation for these discrepancies can be found in the fact that while a lower number of
trees were harvested, their average size also increased. Thus, without an adequate planning

Remote Sens. 2021,13, 1435 18 of 25
for liana removal prior to logging or the sufficient application of directional feeling during
harvesting there is potential for a much larger disturbance effect [101].
4.4. Pre and Post Logging Aboveground Biomass (AGB) and Committed Emissions (CCE)
If the AGB is averaged prior to logging, the value obtained is 308
±
76.3 Mg ha
−1
,
which coincides with several studies in humid tropical lowland forests of the Venezuelan
Amazon [
12
,
13
,
31
,
32
,
102
]. Logging across all compartments reduced pre-disturbance AGB,
on average, by 35% with higher values found for the unplanned logging approaches as
expected (Table 3; Figure 6). While the characteristics of the logging at each compartment
are unique, it is encouraging to see a lower reduction in carbon losses when selective
logging includes a better preparation. However, it must be noted that a lower damage
is not always a response of better planning. Instead, it can also be a direct response to
the spatial aggregation of commercially valuable timber trees along with topographical
conditions and other biophysical/economic factors that are particular at each site and at the
time when logging occurred (e.g., 1985 vs. 2015). For instance, in a recent review conducted
across the tropics, Putz et al. [
101
] found that an average of 57% (range 22–97%) of the area
in logging blocks was not directly affected by timber harvests, with more forests being left
intact in areas farther from roads, on slopes >40%, and within 25 m of perennial streams.
In addition, our study, as many others focusing on the impacts of selective logging in the
tropics (e.g., Verissimo et al. [
97
]; Gerwing, [
98
]; Veríssimo et al. [
99
]) is mostly based on
aggregated means of logging intensity, and that often can be a relatively weak reflection of
the conditions on the ground [101].
In our analysis, CCEs both by sources (Figure 5) and those expressed by area, vol-
ume and impact factor (Figure 6) were higher compared to other parts of the tropics
(e.g., [98–100]).
This corroborates what has been discussed regarding the intensity and
the logging methods used across these areas of the IFR, highlighting the widespread low
efficiency compared to other cases where reduced impact logging is formally applied.
Although beyond the scope of our work, the absence of formal criteria and indicators for
monitoring forest management practices in Venezuelan managed forests that has been
demonstrated previously [
29
], our results add new evidence about the inadequate planning
of logging activities, which may further limit the potential of forest management to serve
as a climate change mitigation tool.
5. Conclusions
Compared to other regions of the Amazon basin and the tropics in general, our
research reveals that in the northeastern Venezuelan Amazon, while the overall harvesting
intensity has remained low over long periods of time, the disturbances associated to logging
were considerably high. Selective logging activities showed rather poor planning and low
efficiency, reflected in the fact that, regardless of the metric used (area-based, volume-based
or CIF), carbon emissions were higher than most studies focusing on similar questions.
Digital image processing and GIS techniques used within the TerraAmazon system,
and in conjunction with the 2019 Refinement IPCC Guidelines, enabled us to develop,
for the first time in Venezuela, a low-cost and robust analytical approach to study the
relationships between selective logging, forest degradation and carbon emissions. Our
work also reveals that the use of spectral contrast along with Landsat time series, and
ground-based data are excellent tools for the analysis of forest degradation, the evaluation
of the impact at the canopy level due to the different activities and modalities of selective
logging, and for the estimation of carbon emissions. This study is a step forward to
improve and plan other more detailed analytical techniques (e.g., Lidar-Light Detection and
Ranging) and for establishing a baseline of carbon emissions in the context of sustainable
forest management.
Author Contributions:
Conceptualization, C.P.-A.; methodology, C.P.-A.; software, C.P.-A. and C.C.;
validation, C.C. and A.G., formal analysis, C.P.-A., W.P.-R., J.S., and E.V.; investigation, C.P.-A., W.P.-R.
and J.S.; writing—original draft preparation, C.P.-A., W.P.-R., J.S., E.V., and S.M.-A.; writing review

Remote Sens. 2021,13, 1435 19 of 25
and editing, C.P.-A., W.P.-R., and E.V.; visualization, S.M.-A. and A.G.; supervision, C.P.-A.; All
authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Not applicable.
Acknowledgments:
The authors would like to thank to the coordinators of the project “Sustainable
Forest Lands Management and Conservation Under an Eco-Social Approach (GCP/VEN/011/GFF)”,
funded by the Global Environment Facility (GEF), for logistical support in field trips; and ENAFOR
for sharing information from management plans and data from forest inventory plots. We also
thank Jose Ignacio Azuaje from Ministerio del Poder Popular para el Ecosocialismo for his support
providing data and information about the management process for Imataca Forest Reserve.
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A
Remote Sens. 2021, 13, x FOR PEER REVIEW 19 of 25
also reveals that the use of spectral contrast along with Landsat time series, and ground-
based data are excellent tools for the analysis of forest degradation, the evaluation of the
impact at the canopy level due to the different activities and modalities of selective log-
ging, and for the estimation of carbon emissions. This study is a step forward to improve
and plan other more detailed analytical techniques (e.g., Lidar—Light Detection and
Ranging) and for establishing a baseline of carbon emissions in the context of sustainable
forest management.
Author Contributions: Conceptualization, C.P.-A.; methodology, C.P.-A.; software, C.P.-A. and
C.C.; validation, C.C. and A.G., formal analysis, C.P.-A., W.P.-R., J.S., and E.V.; investigation, C.P.-
A., W.P.-R. and J.S.; writing—original draft preparation, C.P.-A., W.P.-R., J.S., E.V., and S.M.-A.;
writing review and editing, C.P.-A., W.P.-R., and E.V.; visualization, S.M.-A. and A.G.; supervi-
sion, C.P.-A.; All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Not applicable.
Acknowledgments: The authors would like to thank to the coordinators of the project “Sustainable
Forest Lands Management and Conservation Under an Eco-Social Approach (GCP/VEN/011/GFF)”,
funded by the Global Environment Facility (GEF), for logistical support in field trips; and ENAFOR
for sharing information from management plans and data from forest inventory plots. We also
thank Jose Ignacio Azuaje from Ministerio del Poder Popular para el Ecosocialismo for his support
providing data and information about the management process for Imataca Forest Reserve.
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A
Figure A1. Location of field data.
Figure A1. Location of field data.

Remote Sens. 2021,13, 1435 20 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 20 of 25
Figure A2. Blocks of systematic sampling on the soil fraction images of ENAFOR (9/23/1997) and
INTECMACA (9/9/2015).
Figure A3. Errors of commission and omission of the evaluation of the accuracy of the selective
logging map.
Figure A2.
Blocks of systematic sampling on the soil fraction images of ENAFOR (9/23/1997) and
INTECMACA (9/9/2015).
Remote Sens. 2021, 13, x FOR PEER REVIEW 20 of 25
Figure A2. Blocks of systematic sampling on the soil fraction images of ENAFOR (9/23/1997) and
INTECMACA (9/9/2015).
Figure A3. Errors of commission and omission of the evaluation of the accuracy of the selective
logging map.
Figure A3.
Errors of commission and omission of the evaluation of the accuracy of the selective
logging map.

Remote Sens. 2021,13, 1435 21 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 21 of 25
Figure A4. Linear regression between mapping of selective logging percentage data and ground-
truth data.
Figure A5. Area logged reported in management plans vs. area mapped (A) and general differ-
ences (error %) (B).
Figure A4.
Linear regression between mapping of selective logging percentage data and ground-truth
data.
Remote Sens. 2021, 13, x FOR PEER REVIEW 21 of 25
Figure A4. Linear regression between mapping of selective logging percentage data and ground-
truth data.
Figure A5. Area logged reported in management plans vs. area mapped (A) and general differ-
ences (error %) (B).
Figure A5.
Area logged reported in management plans vs. area mapped (
A
) and general differences
(error %) (B).

Remote Sens. 2021,13, 1435 22 of 25
References
1. Blaser, J.; Sarre, A.; Poore, D.; Johnson, S. Status of Tropical Forest Management 2011; ITTO Technical Series; International Tropical
Timber Organization: Yokohama, Japan, 2011.
2.
Petrokofsky, G.; Sist, P.; Blanc, L.; Doucet, J.-L.; Finegan, B.; Gourlet-Fleury, S.; Healey, J.R.; Livoreil, B.; Nasi, R.;
Peña-Claros, M.; et al.
Comparative effectiveness of silvicultural interventions for increasing timber production and sustaining conservation values in
natural tropical production forests. A systematic review protocol. Environ. Evid. 2015,4, 8. [CrossRef]
3.
Sist, P.; Rutishauser, E.; Peña-Claros, M.; Shenkin, A.; Hérault, B.; Blanc, L.; Baraloto, C.; Baya, F.; Benedet, F.; da Silva, K.E.; et al.
The Tropical managed Forests Observatory: A research network addressing the future of tropical logged forests. Appl. Veg. Sci.
2015,18, 171–174. [CrossRef]
4.
Piponiot, C.; Rödig, E.; Putz, F.E.; Rutishauser, E.; Sist, P.; Ascarrunz, N.; Blanc, L.; Derroire, G.; Descroix, L.; Guedes, M.C.; et al.
Can timber provision from Amazonian production forests be sustainable? Environ. Res. Lett. 2019,14, 64014. [CrossRef]
5.
Piponiot, C.; Rutishauser, E.; Derroire, G.; Putz, F.E.; Sist, P.; West, T.A.P.; Descroix, L.; Guedes, M.C.; Coronado, E.N.H.;
Kanashiro, M.; et al.
Optimal strategies for ecosystem services provision in Amazonian production forests. Environ. Res. Lett.
2019,14, 124090. [CrossRef]
6.
Putz, F.E.; Romero, C. Futures of Tropical Production Forests; Center for International Forestry Research (CIFOR): Bogor, Indonesia,
2015. [CrossRef]
7.
Putz, F.E.; Zuidema, P.A.; Synnott, T.; Peña-Claros, M.; Pinard, M.A.; Sheil, D.; Vanclay, J.K.; Sist, P.; Gourlet-Fleury, S.;
Griscom, B.; et al.
Sustaining conservation values in selectively logged tropical forests: The attained and the attainable. Conserv.
Lett. 2012,5, 296–303. [CrossRef]
8.
Putz, F.E.; Zuidema, P.A.; Pinard, M.A.; Boot, R.G.A.; Sayer, J.A.; Sheil, D.; Sist, P.; Elias; Vanclay, J.K. Improved Tropical Forest
Management for Carbon Retention. PLoS Biol. 2008,6, e166. [CrossRef] [PubMed]
9.
FAO. Forest degradation and improvement. In Proceedings of the Second Expert Meeting on Harmonising Forest-Related
Definitions for use by Various Stakeholders, Rome, Italy, 11–13 September 2002.
10.
Ellis, P.W.; Gopalakrishna, T.; Goodman, R.C.; Putz, F.E.; Roopsind, A.; Umunay, P.M.; Zalman, J.; Ellis, E.A.; Mo, K.;
Gregoire, T.G.; et al.
Reduced-impact logging for climate change mitigation (RIL-C) can halve selective logging emissions from tropical forests. For.
Ecol. Manag. 2019,438, 255–266. [CrossRef]
11.
Pearson, T.R.H.; Brown, S.; Murray, L.; Sidman, G. Greenhouse gas emissions from tropical forest degradation: An underestimated
source. Carbon Balance Mana. 2017,12, 3. [CrossRef]
12. Houghton, R.A. Aboveground forest biomass and the global carbon balance. Glob. Chang. Biol. 2005,11, 945–958. [CrossRef]
13.
Vilanova, E.; Ramírez-Angulo, H.; Torres-Lezama, A. El almacenamiento de carbono en la biomasa aérea como un indicador del
impacto de la extracción selectiva de maderas en la Reserva Forestal Imataca, Venezuela. Interciencia 2010,35, 659–665.
14.
Asner, G.P.; Knapp, D.E.; Broadbent, E.N.; Oliveira, P.J.C.; Keller, M.; Silva, J.N. Selective Logging in the Brazilian Amazon.
Science 2005,310, 480–482. [CrossRef] [PubMed]
15.
Hosonuma, N.; Herold, M.; De Sy, V.; De Fries, R.; Brockhaus, M.; Verchot, L.; Angelsen, A.; Romijn, E. An assessment of
deforestation and forest degradation drivers in developing countries. Environ. Res. Lett. 2012,7, 44009. [CrossRef]
16.
Mollicone, D.; Achard, F.; Federici, S.; Eva, H.D.; Grassi, G.; Belward, A.; Raes, F.; Seufert, G.; Stibig, H.-J.; Matteucci, G.
An incentive mechanism for reducing emissions from conversion of intact and non-intact forests. Clim. Chang.
2007
,
83, 477–493. [CrossRef]
17.
Pacheco, C.; Aguado, I.; Mollicone, D. Las causas de la deforestación en Venezuela: Un estudio retrospectivo. Biollania
2011
,
10, 281–292.
18.
Souza, C. Monitoring of Forest Degradation: A Review of Methods in the Amazon Basin. In Global Forest Monitoring from Earth
Observation; Archard, F., Hansen, M.C., Eds.; CRC Press Taylor & Francis Group: Boca Raton, FL, USA, 2013; pp. 171–194.
19.
Achard, F.; Stibig, H.J.; Eva, H.D.; Lindquist, E.J.; Bouvet, A.; Arino, O.; Mayaux, P. Estimating tropical deforestation from Earth
observation data. Carbon Manag. 2010,1, 271–287. [CrossRef]
20.
Herold, M.; Roman-Cuesta, R.; Mollicone, D.; Hirata, Y.; Van Laake, P.; Asner, G.; Souza, C.; Skutsch, M.; Avitabile, V.;
MacDicken, K. Options for monitoring and estimating historical carbon emissions from forest degradation in the context of
REDD+. Carbon Balance Manag. 2011,6, 13. [CrossRef]
21.
UNFCCC. Decision 2/CP.13. Reducción de las Emisiones Derivadas de la Deforestación en los Países en Desarrollo: Métodos
Para Estimular la Adopción de Medidas, Bali. 2007. Available online: http://unfccc.int/resource/docs/2007/cop13/spa/06a01s.
pdf#page=8 (accessed on 19 January 2021).
22.
Nationally determined contributions (NDCs). United Nations Framework Convention on Climate Change. In Proceedings of
the COP 21 Climate Agreement. United Nations Framework Convention on Climate Change, Paris, France, 30 November–11
December 2015.
23.
Kammesheidt, L.; Lezama, A.T.; Franco, W.; Plonczak, M. History of logging and silvicultural treatments in the western
Venezuelan plain forests and the prospect for sustainable forest management. For. Ecol. Manag. 2001,148, 1–20. [CrossRef]
24.
Torres-Lezama, A.; Ramírez-Angulo, H.; Vilanova, E.; Barros, R. Forest resources in Venezuela: Current status and prospects for
sustainable management. Bois Forêts Des Trop. 2008,295, 21–33.
25.
Centeno, J. Estrategia Para el Desarrollo Forestal en Venezuela; Fondo Nacional de Investigación Forestal: Caracas, Venezuela, 1995.
26. MINEC. Anuario Estadísticas Forestales 2017; MINEC: Caracas, Venezuela, 2018.

Remote Sens. 2021,13, 1435 23 of 25
27.
Food and Agriculture Organization of the United Nations (FAO). Global Forest Resources Assessment 2020 Key Findings 2020; FAO:
Rome, Italy, 2020.
28.
Torres, A. La cuidada movilización de los recursos forestales. La industria forestal. Medio humano, establecimientos y actividades.
In Geo Venezuela; Tomo 3; Fundación Polar: Caracas, Venezuela, 2008; pp. 382–438.
29.
Vilanova, E.; Ramírez-Angulo, H.; Ramírez, G.; Torres-Lezama, A. Compliance with sustainable forest management guidelines in
three timber concessions in the Venezuelan Guayana: Analysis and implications. For. Policy Econ. 2012,17, 3–12. [CrossRef]
30.
Berroterán, J.L. Reserva Forestal Imataca, Ecología y Bases Técnicas Para el Ordenamiento Territorial; Ministerio del Ambiente y de los
Recursos Naturales: Caracas, Venezuela, 2003.
31.
Avitabile, V.; Herold, M.; Heuvelink, G.B.M.; Lewis, S.L.; Phillips, O.L.; Asner, G.P.; Armston, J.; Ashton, P.S.; Banin, L.;
Bayol, N.; et al.
An integrated pan-tropical biomass map using multiple reference datasets. Glob. Chang. Biol.
2016
,22, 1406–1420.
[CrossRef] [PubMed]
32.
Delaney, M.; Brown, S.; Lugo, A.; Torres-Lezama, A.; Bello-Quintero, N. The distribution of organic carbon in major components
of forests located in five life zones of Venezuela. J. Trop. Ecol. 1997,13, 697–708. [CrossRef]
33.
Pacheco, C.; Aguado, I.; Mollicone, D. Identification and characterization of deforestation hot spots in Venezuela using MODIS
satellite images. Acta Amaz. 2014,44, 185–196. [CrossRef]
34.
Pacheco, P.; Mo, K.; Dudley, N.; Shapiro, A.; Aguilar-Amuchastegui, N.; Ling, P.Y.; Anderson, C.; Marx, A. Deforestation Fronts:
Drivers and Responses in a Changing World; WWF: Gland, Switzerland, 2021.
35.
Pacheco-Angulo, C. Estimación de las emisiones evitadas directas e indirectas en los depósitos de biomasa terrestres de la
RFI. In Protocolo para la Valoración Ambiental y Económica de la Reserva Forestal Imataca; Peréz, R., Ed.; Componente 1: Sistema
Nacional Integrado de Información Forestal (SINIFF); Proyecto Ordenación Forestal Sustentable y Conservación de Bosques en la
Perspectiva Ecosocial (GCP/VEN/011/GFF); MINEC: Caracas, Venezuela, 2019.
36.
Ochoa, J. Análisis preliminar de los efectos del aprovechamiento de maderas sobre la composición y estructura de bosques en la
Guayana venezolana. Interciencia 1998,23, 197–207.
37. Pearson, T.R.H.; Brown, S.; Casarim, F. Carbon emissions from tropical forest degradation caused by logging. Environ. Res. Lett.
2014,9, 1–11. [CrossRef]
38.
Global Forest Observations Initiative (GFOI). Integration of Remote-Sensing and Ground-Based Observations for Estimation of Emissions
and Removals of Greenhouse Gases in Forests; Methods and Guidance from the Global Forest Observations Initiative; Global Forest
Observations Initiative; Group on Earth; Food and Agriculture Organization of the United Nations: Rome, Italy, 2020.
39.
IPCC. 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories; Calvo Buen.; IPCC: Bern, Switzerland, 2019.
40.
Bullock, E.L.; Woodcock, C.E.; Souza, C., Jr.; Olofsson, P. Satellite-based estimates reveal widespread forest degradation in the
Amazon. Glob. Chang. Biol. 2020,26, 2956–2969. [CrossRef]
41.
Bullock, E.L.; Woodcock, C.E.; Olofsson, P. Monitoring tropical forest degradation using spectral unmixing and Landsat time
series analysis. Remote Sens. Environ. 2020,238, 110968. [CrossRef]
42.
Ota, T.; Ahmed, O.S.; Minn, S.T.; Khai, T.C.; Mizoue, N.; Yoshida, S. Estimating selective logging impacts on aboveground
biomass in tropical forests using digital aerial photography obtained before and after a logging event from an unmanned aerial
vehicle. For. Ecol. Manag. 2019,433, 162–169. [CrossRef]
43.
Pacheco-Angulo, C.; Vilanova, E.; Aguado, I.; Monjardin, S.; Martinez, S. Carbon emissions from deforestation and degradation
in a forest reserve in Venezuela between 1990 and 2015. Forests 2017,8, 291. [CrossRef]
44.
Shimabukuro, Y.E.; Arai, E.; Duarte, V.; Jorge, A.; dos Santos, E.G.; Gasparini, K.A.C.; Dutra, A.C. Monitoring deforestation and
forest degradation using multi-temporal fraction images derived from Landsat sensor data in the Brazilian Amazon. Int. J. Remote
Sens. 2019,40, 5475–5496. [CrossRef]
45.
Souza, C.M.; Roberts, D. Mapping forest degradation in the Amazon region with Ikonos images. Int. J. Remote Sens.
2005
,
26, 425–429. [CrossRef]
46.
GOFC-GOLD. A Sourcebook of Methods and Procedures for Monitoring and Reporting Anthropogenic Greenhouse Gas Emissions and
Removals Associated with Deforestation, Gains and Losses of Carbon Stocks in Forests Remaining Forests, and Forestation; GOFC-GOLD
Report Versio; GOFC-GOLD Land Cover Project Office, Wageningen University: Wageningen, The Netherlands, 2016.
47.
Souza, C.; Firestone, L.; Silva, L.M.; Roberts, D. Mapping forest degradation in the Eastern Amazon from SPOT 4 through spectral
mixture models. Remote Sens. Environ. 2003,87, 494–506. [CrossRef]
48.
Monteiro, A.L.; Souza, C.M.; Barreto, P. Detection of logging in Amazonian transition forests using spectral mixture models. Int. J.
Remote Sens. 2003,24, 151–159. [CrossRef]
49.
Souza, J.R.; Barreto, P. An alternative approach for detecting and monitoring selectively logged forests in the Amazon. Int. J.
Remote Sens. 2000,21, 173–179. [CrossRef]
50.
Banskota, A.; Kayastha, N.; Falkowski, M.J.; Wulder, M.A.; Froese, R.E.; White, J.C. Forest Monitoring Using Landsat Time Series
Data: A Review. Can. J. Remote Sens. 2014,40, 362–384. [CrossRef]
51.
Jackson, C.M.; Adam, E. Remote sensing of selective logging in tropical forests: Current state and future directions. iForest
Biogeosci. For. 2020,13, 286–300. [CrossRef]
52.
TerraAmazon. Monitoring System of Deforestation in the Amazon; Fundação de Ciência, aplicações e Tecnologia Espacial and
Instituto Nacional de Pesquisas Espaciais: São Paulo, Brazil, 2005.
53. CODEFORSA. Informe Plan Anual de Corta No 8. Unidad N-2 Reserva Forestal Imataca; El Palmar: Edo Bolívar, Venezuela, 2003.

Remote Sens. 2021,13, 1435 24 of 25
54.
COMAFOR. Plan de Ordenación y Manejo Forestal, Unidad C-3 Reserva Forestal Imataca; Consorcio Maderero Forestal C.A. Upata,
Estado Bolívar.Varios tomos; COMAFOR: Upata, Venezuela, 1994.
55.
Piponiot, C.; Sist, P.; Mazzei, L.; Peña-Claros, M.; Putz, F.E.; Rutishauser, E.; Shenkin, A.; Ascarrunz, N.; de Azevedo, C.P.;
Baraloto, C.; et al. Carbon recovery dynamics following disturbance by selective logging in Amazonian forests. Elife
2016
,
5, e21394. [CrossRef] [PubMed]
56.
INTECMACA. Plan de Ordenación y Manejo Forestal de la Unidad V de la Reserva Forestal Imataca; Industria Técnica de Maderas,
C.A.: Caracas, Venezuela, 1989.
57.
ENAFOR. Plan de Ordenación y Manejo Forestal Unidad V, Reserva Forestal Imataca; Empresa Nacional Forestal, S.A.: Caracas,
Venezuela, 2012.
58.
Asner, G.P.; Keller, M.; Pereira Rodrigo, J.; Zweede, J.C.; Silva, J.N.M. Canopy damage and recovery after selective logging in
amazonia: Field and satellite studies. Ecol. Appl. 2004,14, 280–298. [CrossRef]
59.
Souza, C.; Cochrane, M.; Sales, M.; Monteiro, A.; Mollicone, D. Integrating Forest Transects and Remote Sensing data to Quantify
Carbon Loss due to Forest Degradation: A case study of the Brazilian Amazon. In Case Studies on Measuring and Assessing Forest
Degradation; Forest Resources Assessment WorkingPaper161; FAO: Rome, Italy, 2009; 20p.
60.
Hurtado, C. Evaluación de los proyectos de investigación de la Unidad V de la Reserva Forestal Imataca; Universidad de Los Andes,
Facultad de Ciencias Forestales, Escuela de Capacitación Forestal: Mérida, Venezuela, 1988.
61.
ENAFOR. Monitoreo de la Masa Forestal pre Aprovechamiento, Unidad de Producción Anual Santa María II; Forestal, E.N., Ed.; ENAFOR:
Upata, Venezuela, 2015.
62.
ENAFOR. Monitoreo Post Aprovechamiento de la Masa Forestal, Unidad de Producción Anual Santa María II; Forestal, E.N., Ed.;
ENAFOR: Upata, Venezuela, 2016.
63.
Azuaje, F. Segundo Informe de Avance de la Consultoría en Restauración, Conservación y Manejo Forestal Sustentable (MFS) Manejo Sus-
tentable de Tierras (MST) de Bosques en Zonas Afectadas por Procesos de Degradación, “Ecología Forestal”; Proyecto: Ordenación forestal
sustentable y conservación de bosques en la perspectiva ecosocial (GCP/VEN/011/GFF); MINEC: Caracas, Venezuela, 2018.
64. INPE-FUNCATE. TerraAmazon 4.4 User´s Guide Administrator; INPE-FUNCATE: São Josédos Campos, Brazil, 2013.
65.
Shimabukuro, Y.E.; Beuchle, R.; Grecchi, R.C.; Achard, F. Assessment of forest degradation in Brazilian Amazon due to
selective logging and fires using time series of fraction images derived from Landsat ETM+ images. Remote Sens. Lett.
2014
,
5, 773–782. [CrossRef]
66.
Vidal, D.; Corrêa, M.; Gama, A.; Guerreiro, C.; De Almeida, A.; Corrêa, M.; Cavalcante, N.; Sant’Ana, J. Testes para definição
dos parâmetros de detecção de nuvens e sombras em imagens do sensor AWIFS no plugin Cloud Detection, do aplicativo
TerraAmazon. In Proceedings of the An. XVII Simpósio Bras. Sensoriamento Remoto—SBSR, João Pessoa, Brazil, 25–29 April
2015; INPE: Florianópolis, Brazil, 2015.
67.
Chuvieco, E. Fundamentals of Satelite Remote Sensing an Environmental. An Environmental Approach; Taylor & Francis Group: Boca
Ratón, FL, USA, 2016.
68.
Congalton, R.; Green, K. Assesing the Accuracy of Remotely Sensed Data: Principles and Practices; CRC Press: Boca Raton, FL, USA, 2009.
69.
Jensen, J.R. Introductory Digital Image Processing: A Remote Sensing Perspective, 3rd ed.; Prentice-Hall: Upper Saddle River, NJ, USA, 2005.
70.
Olofsson, P.; Arévalo, P.; Espejo, A.B.; Green, C.; Lindquist, E.; McRoberts, R.E.; Sanz, M.J. Mitigating the effects of omission
errors on area and area change estimates. Remote Sens. Environ. 2020,236, 111492. [CrossRef]
71.
Olofsson, P.; Foody, G.M.; Herold, M.; Stehman, S.V.; Woodcock, C.E.; Wulder, M.A. Good practices for estimating area and
assessing accuracy of land change. Remote Sens. Environ. 2014,148, 42–57. [CrossRef]
72.
Cohen, W.; Fiorella, M.; Gray, J.; Helmer, E.; Anderson, K. An Efficient and Accurate Method for Mapping Forest Clearcuts in the
Pacific Northwest using Landsat imagery; American Society for Photogrammetry and Remote Sensing: Bethesda, MD, USA, 1998;
Volume 64.
73.
Congalton, R. comparison of sampling schemes used in generating error matrices for assessing the accuracy of maps generated
from remotely sensed data. Photogramm. Eng. Remote Sens. 1988,54, 593–600.
74.
Moreno, M.V.; Chuvieco, E. Validación de productos globales de cobertura del suelo en la España Peninsular. Rev. Teledetec.
2009
,
31, 5–22.
75.
Chuvieco, E.; Englefield, P.; Trishchenko, A.P.; Luo, Y. Generation of long time series of burn area maps of the boreal forest from
NOAA–AVHRR composite data. Remote Sens. Environ. 2008,112, 2381–2396. [CrossRef]
76.
Roy, D.P.; Landmann, T. Characterizing the surface heterogeneity of fire effects using multi-temporal reflective wavelength data.
Int. J. Remote Sens. 2005,26, 4197–4218. [CrossRef]
77.
Pacheco, C.; Aguado, I.; Lopez, J. Comparación de los métodos utilizados en el monitoreo de la deforestación tropical, para la
implementación de estrategias REDD+, caso de estudio los Llanos Occidentales Venezolanos. In Proceedings of the Anais XVI
Simpósio Brasileiro Sensoriamento Remoto—SBSR, Foz do Iguaçu, Brasil, 13–18 April 2013; pp. 2817–2826.
78.
Chave, J.; Réjou-Méchain, M.; Búrquez, A.; Chidumayo, E.; Colgan, M.S.; Delitti, W.B.C.; Duque, A.; Eid, T.; Fearnside, P.M.;
Goodman, R.C.; et al. Improved allometric models to estimate the aboveground biomass of tropical trees. Glob. Chang. Biol.
2014
,
20, 3177–3190. [CrossRef]
79.
Brown, S.; Pearson, T.; Moore, N.; Parveen, A.; Ambagis, S.; Shoch, D. Impact of Selective Logging on the Carbon Stocks of Tropical
Forests: Republic of Congo as a Case Study; Winrock International Report; USAID: Washington, DC, USA, 2005.

Remote Sens. 2021,13, 1435 25 of 25
80.
Zanne, A.E.; Lopez-Gonzalez, G.; Coomes, D.A.; Ilic, J.; Jansen, S.; Lewis, S.L.; Miller, R.B.; Swenson, N.G.; Wiemann, M.C.; Chave, J.
Data from: Towards a Worldwide Wood Economics Spectrum, Dryad, Dataset. 2009. Available online: https://doi.org/10.5061/
dryad.234 (accessed on 7 April 2021).
81.
Chave, J.; Coomes, D.; Jansen, S.; Lewis, S.L.; Swenson, N.G.; Zanne, A.E. Towards a worldwide wood economics spectrum. Ecol.
Lett. 2009,12, 351–366. [CrossRef] [PubMed]
82.
Feldpausch, T.R.; Jirka, S.; Passos, C.A.M.; Jasper, F.; Riha, S.J. When big trees fall: Damage and carbon export by reduced impact
logging in southern Amazonia. For. Ecol. Manag. 2005,219, 199–215. [CrossRef]
83.
Ellis, P.; Griscom, B.; Walker, W.; Gonçalves, F.; Cormier, T. Mapping selective logging impacts in Borneo with GPS and airborne
lidar. For. Ecol. Manag. 2016,365, 184–196. [CrossRef]
84.
Asner, G.P.; Keller, M.; Pereira Jr, R.; Zweede, J.C. Remote sensing of selective logging in Amazonia: Assessing limitations based
on detailed field observations, Landsat ETM+, and textural analysis. Remote Sens. Environ. 2002,80, 483–496. [CrossRef]
85.
Graça, P.; Santos, J.; Soares, J.; Souza, P. Desenvolvimento Metodológico Para Detecção e Mapeamento de Áreas Florestais sob
Exploração Madeireira: Estudo de Caso, Região Norte do Mato Grosso. In Proceedings of the Anais XII Simpósio Brasileiro de
Sensoriamento Remoto, Goiânia, Brasil, 16–21 April 2005; pp. 1555–1562.
86.
Herold, M. An Assessment of National Forest Monitoring Capabilities in Tropical Non-Annex I Countries: Recommendations for Capacity
Building; GOFC-GOLD Land Cover Project Office and Friedrich Schiller University: Jena, Norway, 2009.
87.
Beuchle, R.; Eva, H.D.; Stibig, H.-J.; Bodart, C.; Brink, A.; Mayaux, P.; Johansson, D.; Achard, F.; Belward, A. A satellite data set for
tropical forest area change assessment. Int. J. Remote Sens. 2011,32, 7009–7031. [CrossRef]
88.
Modica, G.; Merlino, A.; Solano, F.; Mercurio, R. An index for the assessment of degraded Mediterranean forest ecosystems. For.
Syst. 2015,24, 37. [CrossRef]
89.
Wu, B.; Meng, X.; Ye, Q.; Sharma, R.P.; Duan, G.; Lei, Y.; Fu, L. Method of Estimating Degraded Forest Area: Cases from Dominant
Tree Species from Guangdong and Tibet in China. Forests 2020,11, 930. [CrossRef]
90.
Monteiro, A.; Lingnau, C.; Souza, C. Classificação orientada a objeto para detecção da exploração seletiva de madeira na
amazônia.Object-based classification to detection of selective logging in the Brazilian Amazon. Rev. Bras. Cartogr. 2007,59, 225–234.
91.
Hethcoat, M.G.; Edwards, D.P.; Carreiras, J.M.B.; Bryant, R.G.; França, F.M.; Quegan, S. A machine learning approach to map
tropical selective logging. Remote Sens. Environ. 2019,221, 569–582. [CrossRef]
92.
Yan, G.; Margaret, S.; Jaime, P.-G.; Adrian, G. Remote sensing of forest degradation: A review. Environ. Res. Lett.
2020
,15, 103001.
93.
ENAFOR. Monitoreo de la masa Forestal Post Aprovechamiento, Unidad de Producción Anual Santa María I; Empresa Nacional Forestal
(ENAFOR): Upata, Venezuela, 2015.
94.
Ussher, E.; Gutiérrez, N.; Vilanova, E. Impacto del aprovechamiento forestal en la composición de especies de uso potencial
maderable y no maderable en la reserva Forestal Imataca, Venezuela. In Proceedings of the XIV Congreso Forestal Mundial,
Durban, Sudáfrica, 7–11 September 2015.
95.
Jackson, S.M.; Fredericksen, T.S.; Malcolm, J.R. Area disturbed and residual stand damage following logging in a Bolivian tropical
forest. For. Ecol. Manag. 2002,166, 271–283. [CrossRef]
96.
Johns, J.; Barreto, P.; Uhl, C. Os Danos da Exploracao de Madera Com e Sem Planejamento na Amazonia Oriental; IMAZON, Instituto do
Homem e Meio Ambiente da Amazônia: Belém, Brazil, 1998.
97.
Verissimo, A.; Barreto, P.; Mattos, M.; Tarifa, R.; Uhl, C. Logging impacts and prospects for sustainable forest management in an
old Amazonian frontier: The case of Paragominas. For. Ecol. Manag. 1992,55, 169–199. [CrossRef]
98.
Gerwing, J.J. Degradation of forests through logging and fire in the eastern Brazilian Amazon. For. Ecol. Manag.
2002
,157, 131–141.
[CrossRef]
99.
Veríssimo, A.; Barreto, P.; Tarifa, R.; Uhl, C. Extraction of a high-value natural resource in Amazonia: The case of mahogany. For.
Ecol. Manag. 1995,72, 39–60. [CrossRef]
100.
Asner, G.P.; Broadbent, E.N.; Oliveira, P.J.C.; Keller, M.; Knapp, D.E.; Silva, J.N.M. Condition and fate of logged forests in the
Brazilian Amazon. Proc. Natl. Acad. Sci. USA 2006,103, 12947–12950. [CrossRef] [PubMed]
101.
Putz, F.E.; Baker, T.; Griscom, B.W.; Gopalakrishna, T.; Roopsind, A.; Umunay, P.M.; Zalman, J.; Ellis, E.A.; Ruslandi; Ellis, P.W.
Intact Forest in Selective Logging Landscapes in the Tropics. Front. For. Glob. Chang. 2019,2, 30. [CrossRef]
102.
Malhi, Y.; Wood, D.; Baker, T.R.; Wright, J.; Phillips, O.L.; Cochrane, T.; Meir, P.; Chave, J.; Almeida, S.; Arroyo, L.; et al. The regional
variation of aboveground live biomass in old-growth Amazonian forests. Glob. Chang. Biol. 2006,12, 1107–1138. [CrossRef]