Technical ReportPDF Available

Abstract and Figures

This project aimed to develop a geodatabase of Industrial Tree Plantations (ITPs) in Caraga Region using Remote Sensing (RS) and Geographic Information System (GIS). The geodatabase is expected to aid in the characterization of ITPs in terms of their types, locations, spatial arrangements, and total area. It also aims to provide a form of documentation of the spatial-temporal aspects of ITP growth and development, and management dynamics. An important part of the geodatabase development is mapping the species types, location and extent of ITPs. The project did this by applying machine learning techniques to available RS datasets and complemented by ground surveys. Another objective of the project is to determine areas suitable for establishing new ITPs through conduct of suitability analysis; and to conduct accessibility analysis of log production flow with the use of geodatabase. Among the major accomplishments of the project are: (i.) the maps and statistics of ITPs in Caraga Region generated through the analysis of satellite and airborne remote sensing images; (ii.) a PostgreSQL+PostGIS geodatabase of ITPs in the region, including an online geodatabase visualization portal accessible at https://geoitp.ccgeo.info; (iii.) the maps and statistics of areas suitable for ITPs; and (iv.) a characterization and analysis of the spatial location, accessibility, and capability of wood processing plants (WPPs) for log production vis-à-vis existing Falcata plantations in the region. Aside from the ITP geodatabase, the project has generated a significant number of maps and other data products. For these to be accessible and utilized by the public, these products have been uploaded to the Mindanao Integrated Data Sharing Environment (MInDSEt), an online data portal managed by the Caraga Center for Geo-Informatics, of Caraga Center for Geo-Informatics, Caraga State University, Butuan City, Philippines. Interested users can access the project outputs at http://mindset.ccgeo.info:82/organization/industrial-tree-plantation-itp-research-and innovation-center).
Content may be subject to copyright.
ii
Terminal Report
Accelerated R&D Program for Capacity Building of Research and
Development Institutions and Industrial Competitiveness: Niche
Centers in the Regions for R&D (NICER) Program:
INDUSTRIAL TREE PLANTATIONS RESEARCH AND
INNOVATION CENTER (ITP CENTER) FOR UPGRADING THE
WOOD-BASED INDUSTRY
Project Title: DEVELOPMENT OF A GEODATABASE OF
INDUSTRIAL TREE PLANTATIONS IN CARAGA REGION
USING REMOTE SENSING AND GIS
Project Leader:
Engr. Jojene R. Santillan
Implementing Agency:
Caraga State University
Ampayon, Butuan City
Funding Agency:
Department of Science and Technology (DOST)
Monitoring Agency:
Philippine Council for Agriculture, Aquatic, and Natural
Resources Research and Development (PCAARRD)
September 30, 2021
iii
Summary Sheet
Program Title
Accelerated R&D Program for Capacity Building of Research
and Development Institutions and Industrial Competitiveness:
Niche Centers in the Regions for R&D (NICER) Program:
Industrial Tree Plantations Research and Innovation Center (ITP
Center) for Upgrading the Wood-Based Industry
Program Leader
For. Roger T. Sarmiento
Project Title
Project 1. Development of a Geodatabase of Industrial Tree
Plantations in Caraga Region Using Remote Sensing and GIS
Project Leader
Engr. Jojene R. Santillan
Project Staff
Engr. Jun Love E. Gesta (Information Systems Researcher II)
Ms. Marcia Coleen N. Marcial (Science Research Specialist I)
Engr. Arnaldo C. Gagula (Faculty Project Staff, L2)
Mr. Edsel Matt O. Morales (Faculty Project Staff, L2)
Engr. Meriam M. Santillan (Faculty Project Staff, L3)
Implementing
Agency
Caraga State University, Ampayon, Butuan City
Cooperating
Agencies
Department of Environment and Natural Resources-13
Local Government Units (LGUs) of Caraga Region
Funding Agency
Department of Science and Technology (DOST)
Monitoring
Agency
Philippine Council for Agriculture, Aquatic, and Natural
Resources Research and Development (PCAARRD)
Project Duration
October 1, 2019 March 31, 2021
April 1, 2021 September 30, 2021
Total Budget
Year 1: P 4,343,144.97
Year 2: P 2,636,978.02
Total: P 6,980,122.99
iv
Acknowledgement
The project would like to acknowledge the following entities for their significant
contributions and support:
DOST and Science for Change Program for the financial support, especially to
Mr. Gilbert Poralan Jr., Program Manager, Niche Centers in the Regions for
R&D (NICER) Program;
DOST PCAARRD for project monitoring and management support, especially
to Dr. Daisy Espiritu-Cabral, ISP Manager Industrial Tree Plantation(ITP);
Department of Environment and Natural Resources (DENR) Caraga Region;
Provincial Environment and Natural Resources Office (PENRO) Agusan del
Norte and Agusan del Sur;
Community Environment and Natural Resources Office (CENRO) throughout
Caraga Region;
National Mapping and Resource Information Authority (NAMRIA);
Philippine Statistics Authority (PSA);
European Space Agency (for the Sentinel-1 and Sentinel-2 satellite images);
US Geological Survey (for the Landsat satellite images); and
DOST ASTI PEDRO/Datos for Planet satellite images.
The following individuals are also acknowledged for their assistance in the satellite
and other spatial data processing and analysis, and in the field data collection activities
of the project:
Engr. Monalaine Bermoy
Engr. Joy Casinginan
Engr. Mark Dave Plaza
Jeziel Ata
Mary Grace Sotto
Mary Jane Tiempo
Irez Abines
Jerec Carpio
Kent Lloyd Mañoza
Jeffrey Jamelo
Princess Amoncio
Jenilyn Lingatong
Ann Marjorie Betio
Juliet Cabusao
Yereh-El Gono
Russel Pastor
Aries del Monte
Marie Angel Puyo
Rachelle Soliman
John Carl Escasio
The Project Leader also acknowledges the dedication and commitment of its
project personnel, namely Engr. Junlove Gesta and Marcia Coleen Marcial; and the
faculty project staff, namely Engr. Meriam Makinano-Santillan, Engr. Arnaldo C.
Gagula, and Mr. Edsel Matt O. Morales. The motivation and support of Dr. Rowena P.
v
Varela, the former Program Leader and CSU Vice President for Research, Innovation
and Extension, are also recognized.
Finally, we would like to thank our research partner, the Caraga Center for Geo-
Informatics (CCGeo), for office space, field survey equipment, and computing facilities
provided to Project 1.
vi
Table of Contents
Summary Sheet ......................................................................................................... iii
Acknowledgement ..................................................................................................... iv
Abstract ..................................................................................................................... ix
Introduction ............................................................................................................. 10
Rationale ............................................................................................................. 10
Objectives of the Project ...................................................................................... 13
Significance of the Project .................................................................................... 13
Literature Review .................................................................................................... 15
Mapping Tree Plantations with Remote Sensing .................................................. 15
Plantation Stand Age Estimation with Remote Sensing........................................ 16
Design and Development of GIS Database for Industrial Tree Plantations ........... 17
Methodology ............................................................................................................ 18
Overview .............................................................................................................. 18
ITP Mapping using Sentinel-2 Images .................................................................. 18
Detailed Mapping of ITPs using Unmanned Aerial System (UAS) ........................ 24
ITP Stand Age Estimation: Case Studies for Falcata Plantations Using Sentinel-2
and Landsat 8 OLI Images ................................................................................... 27
Case Study 1: Falcata Age Estimation Using Sentinel-2 ................................... 28
Case Study 2: Falcata Age Estimation Using Landsat 8 OLI Image .................. 29
Development of an online/web-based GIS database ............................................ 30
Database Development Using PostgreSQL+PostGIS ....................................... 30
Enabling WMS and WFS Access to the Geodatabase ...................................... 31
ITP Geodatabase Visualization Portal .............................................................. 32
ITP-related Spatial Data Layers, Maps and Other Project Outputs Access via
MInDSEt ........................................................................................................... 32
ITP Suitability Modeling and Mapping .................................................................. 33
The MaxEnt Model ........................................................................................... 33
ITP Presence Data ........................................................................................... 33
Environmental Variables ................................................................................... 35
Environmental Variable Selection ..................................................................... 35
MaxEnt Modeling .............................................................................................. 37
Spatial Analysis of Log Production Flow .............................................................. 38
Major Results and Findings ..................................................................................... 40
ITP Mapping Results ............................................................................................ 40
Coarse Resolution Mapping Using Sentinel-2 Image ........................................ 40
Refined and corrected the ITP Maps Using High Resolution Google Earth Images
......................................................................................................................... 42
vii
Illustrated Statistics of ITPs in Caraga Region based on Refinement and
Correction ......................................................................................................... 44
Detailed Maps and Statistics of ITPs in Caraga Region .................................... 52
UAS Mapping ....................................................................................................... 61
Falcata Stand Age Estimation Using Sentinel-2 Image ........................................ 65
Relationship between stand age and Sentinel-2 Bands and Vegetation Indices 65
Estimated Stand Age of Falcata Plantations Using Sentinel-2 Image ............... 65
Concluding Remarks on Falcata Stand Age Estimation Using Sentinel-2 Image
......................................................................................................................... 69
Falcata Stand Age Estimation Using Landsat 8 OLI Image .................................. 70
Multivariate Regression Models for Age Estimation .......................................... 70
Accuracy of the Age Estimation Models............................................................ 70
Concluding Remarks on Falcata Age Estimation Using Landsat 8 OLI ............. 72
Field Mapping of WPPs and Furniture Makers ..................................................... 74
ITP Geodatabase ................................................................................................. 83
ITP Geodatabase Built with PostgreSQL+PostGIS ........................................... 83
ITP Geodatabase Access WMS and WFS ........................................................ 91
Web Visualizations of Geodatabase Layers ..................................................... 93
Online Access to ITP-related Spatial Data Layers, Maps and Other Project
Outputs via MInDSEt ........................................................................................ 94
Results of ITP Suitability Modeling and Mapping Using MaxEnt ........................... 97
MaxEnt Model Results ...................................................................................... 97
MaxEnt Model Output Raster Grids ................................................................ 101
ITP Suitability Maps based on MaxEnt Model Outputs ................................... 103
ITP Suitability Statistics .................................................................................. 108
Results of Log Production Flow Analysis .......................................................... 121
Log Production Potential of Falcata Plantations in Caraga Region ................. 121
Analysis of Annual Log Requirement of Wood Processing Plants................... 122
Service Area Analysis of WPPs ...................................................................... 125
Falcata Plantations within WPP Service Areas ............................................... 129
Location-Allocation Analysis of WPPs and Falcata Plantation ........................ 131
Summary of Major Findings and Conclusions ........................................................ 136
Literature Cited ...................................................................................................... 139
List of Appendices ................................................................................................. 143
Appendix A. Project 6Ps outputs .................................................................... 143
Appendix B. ITP Geodatabase Quick Overview and Instructions for Access .. 143
Appendix C. Provincial-level Industrial Tree Plantation (ITP) Maps ................ 143
Appendix D. Municipal-level ITP Maps ........................................................... 143
Appendix E. ITP Suitability Maps .................................................................... 143
viii
Appendix F. Presented and Published Research papers, including papers for
publication ...................................................................................................... 143
Appendix G. Undergraduate theses of students working on the project .......... 143
Appendix H. End User License Agreements (EULAs)..................................... 143
Appendix I. Policy brief prepared by the project .............................................. 143
Appendix A. Project 6Ps Outputs ........................................................................... 144
ix
Abstract
This project aimed to develop a geodatabase of Industrial Tree Plantations (ITPs) in
Caraga Region using Remote Sensing (RS) and Geographic Information System
(GIS). The geodatabase is expected to aid in the characterization of ITPs in terms of
their types, locations, spatial arrangements, and total area. It also aims to provide a
form of documentation of the spatial-temporal aspects of ITP growth and development,
and management dynamics. An important part of the geodatabase development is
mapping the species types, location and extent of ITPs. The project did this by applying
machine learning techniques to available RS datasets and complemented by ground
surveys. Another objective of the project is to determine areas suitable for establishing
new ITPs through conduct of suitability analysis; and to conduct accessibility analysis
of log production flow with the use of geodatabase.
Among the major accomplishments of the project are: (i.) the maps and statistics of
ITPs in Caraga Region generated through the analysis of satellite and airborne remote
sensing images; (ii.) a PostgreSQL+PostGIS geodatabase of ITPs in the region,
including an online geodatabase visualization portal accessible at
https://geoitp.ccgeo.info; (iii.) the maps and statistics of areas suitable for ITPs; and
(iv.) a characterization and analysis of the spatial location, accessibility, and capability
of wood processing plants (WPPs) for log production vis-à-vis existing Falcata
plantations in the region.
Aside from the ITP geodatabase, the project has generated a significant number of
maps and other data products. For these to be accessible and utilized by the public,
these products have been uploaded to the Mindanao Integrated Data Sharing
Environment (MInDSEt), an online data portal managed by the Caraga Center for Geo-
Informatics, of Caraga Center for Geo-Informatics, Caraga State University, Butuan
City, Philippines. Interested users can access the project outputs at
http://mindset.ccgeo.info:82/organization/industrial-tree-plantation-itp-research-and-
innovation-center).
10
Introduction
Rationale
The Caraga Region (Figure 1) has long been the biggest producer of major
forest products in the Philippines. In 2019, the region contributed 64% (or 607,959 m3)
of the country’s total log production of 948,104 m3 (DENR FMB, 2019). Such
contribution has been consistently held by the region in the past years (Figure 2).
Figure 1. Caraga Region, Mindanao, Philippines. (Credits: Arnaldo C. Gagula, Caraga State University)
Figure 2. Log production statistics for 2008-2019. Source: Philippine Forest Statistics, Forest
Management Bureau.
-
200,000
400,000
600,000
800,000
1,000,000
1,200,000
1,400,000
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019
Log Production, in cubic meter
Caraga Region Philippines
11
Figure 3. A Falcata plantation in Butuan City, Agusan del Norte, Mindanao, Philippines. (Credits: ITP
Center Project 1)
One of the major contributors of the region’s log production are the Industrial Tree
Plantations or ITPs that have been established in the various localities in the Agusan
and Surigao provinces (Paler et al., 1998). Of the region’s 2019 total log production,
99.95% (or 607,656 m3) came from ITPs. Among the ITP species, Falcata is the most
widely planted (Figure 3, Figure 4), providing 555,966 m3 of logs produced in 2019,
which is more than 50% of the nationwide total log production (DENR FMB, 2019).
With the constant need for logs by the wood-based industries, and the fact that
forest products cannot be sufficiently provided by natural forests (Arguirre-Salado, et
al., 2015), the role of ITPs will remain to be significant in the coming years, more so by
the issuance of Executive Order No. 23 in February 2011 which declared a moratorium
on the cutting and harvesting of timber in natural and residual forests nationwide
(Philippine Star, 2018).
Despite their importance in the country’s log production, ITPs in the Caraga Region
are not well characterized in terms of their types, locations, spatial arrangements, and
total area. The spatial-temporal aspects of their growth and development, and
management dynamics are also not well documented. The limited availability of these
information makes it difficult to understand many relevant questions related to ITPs in
the region such as:
Where are the ITPs located in Caraga Region? What are the extents of these
ITPs?
What ITP species are currently planted?
Are there areas available for establishing new ITPs?
12
Where are the processing plants currently located? How efficient are these
processing plants in log production? Where should we establish processing
plants to ensure efficient log production?
Figure 4. Falcata plantations (delineated in cyan color) in Butuan City as can be seen from the Google
Earth application.
Periodic inventories through ground surveys are traditionally used for determining
the types, locations, spatial arrangements, and total area of forest resources. Since
these approaches are often difficult to conduct, time consuming and expensive,
Remote Sensing (RS) and Geographic Information System (GIS) are commonly used
nowadays as better and efficient alternatives. RS and GIS are two technologies that
have been widely used for mapping and characterization of forest resources, including
tree plantations. RS, in particular, is a viable approach for forest resource mapping and
inventories over large areas (Chen et al., 2016; Fagan et al., 2018), including localities
that are difficult and expensive to monitor when using ground surveys. Images
acquired by sensors on-board airborne and satellite platforms can be processed to
map the location and extent of vegetation, including their associated spatio-temporal
changes, even up to the species level (e.g., Koukoulas and Blackburn, 2005). The
information derived from the images can be further analysed and integrated with other
spatial and non-spatial datasets using GIS to derive new information, such as
determining the environmental characteristics of mapped tree locations, detecting and
determining the rate of change of forest resources, and finding areas suitable for
13
establishing tree plantations and processing plants (e.g., Arguirre-Salado, et al., 2015;
Williamson and Nieuwenhuis, 1993), among many others.
Obtaining the above information and integrating it with other datasets is valuable
for efficient management and monitoring of ITPs in the region. It can also be used for
market projections, economic forecast models in relation to climate change scenarios,
and for other purposes requiring wood industry-based information. On the other hand,
monitoring the extent of ITPs is critical for understanding environmental and
socioeconomic impacts (Torbick et al., 2016).
Objectives of the Project
The main objective of this project is to develop a geo-database of ITPs in
Caraga Region using RS and GIS.
The specific objectives are as follows:
1. To map the species types, location and extent of ITPs by applying machine
learning techniques to available RS datasets, images acquired by unmanned
aerial mapping system (UAS), and complemented by ground surveys;
2. To develop an online/web-based GIS database of ITP types, location and
extent integrated with other spatial and non-spatial datasets such as
administrative boundaries, land classification, road network, wood processing
plant locations, and other relevant datasets;
3. To determine areas suitable for establishing new ITPs through conduct of
suitability analysis with the use of GIS database; and
4. To conduct accessibility analysis of log production flow from source to
processing plants with the use of GIS database.
Part of the ITP Mapping process is to provide a means to estimate plantation
stand age as well as number of trees per plantation using remote sensing data.
Significance of the Project
The Project will generate an online geodatabase which will provide the
concerned government agencies (e.g., DENR) with science-based information, such
as the current/most recent location and extents of ITPs and areas for establishment of
new ITPs. This information can then be used to formulate strategies and policies for
either nurturing or improving the status ITPs, and therefore upgrading the wood-based
economy, especially in Caraga Region. The region, being the biggest supplier of
timber, will contribute to the Philippine economy if the region’s wood-based economy
will be improved.
The geodatabase developed by the project will also be beneficial to the Caraga
Region, Wood-based Industry Players, Local Government Units (LGUs), and
Researchers:
Caraga Region: The geodatabase will provide fundamental information
necessary for the formulation and implementation of policies and ordinances
to improve the wood-based industry in the region. Likewise, the improved
14
wood-based industry will generate employment and, therefore, more taxes
will be paid to sustain the government programs.
Wood-based Industry Players: The geodatabase will also guide the wood
industry players in making informed decisions related to their enterprise.
Among the potential uses of the geodatabase are: providing the industry
players of an overview map of current/most recent location and extent of
ITPs in the region; easier identification of localities where to establish new
ITPs; and prospecting ITP locations that have harvestable or soon to be
harvestable logs.
LGUs: The ITPS geodatabase can provide relevant information to support
policy reforms and for effective management measures and regulations
towards upgrading the wood-based industry. The information generated
from the project can be used as basis for training and education among tree
farmers and other wood-based industry players.
Researchers: Researchers will be provided an important source of datasets
and information necessary to conduct research and innovations to upgrade
the wood-based industry.
15
Literature Review
Mapping Tree Plantations with Remote Sensing
Mapping the spatial distribution and temporal dynamics of tree plantations is a
key role that RS has played in recent years (Torbick et al., 2016). As with any type of
vegetation, the use of RS in mapping tree plantations involves various considerations,
processes and techniques. Among them are the type of image to be used, the
classification method to extract vegetation extent from the images, and the accuracy
of image classification (Xie et al., 2008).
There are three known categories of mapping tree plantations based on the
type of image used: 1) optical image-based approach, 2) Synthetic Aperture Radar
(SAR) image-based approach and 3) SAR-optical image integration/fusion approach
(Chen et al., 2016).
Many studies have utilized low to medium resolution optical satellite images
(e.g., those acquired by MODIS, ASTER, Landsat, SPOT, and Sentinel-2) to map tree
plantations such as rubber (Li and Fox, 2011; Li and Fox, 2011; Liu et al. 2012), oil
palm (Broich et al., 2011; Carlson et al., 2013), acacia (Win et al., 2009; Larson, 1993),
and bamboo (Vina et al., 2008; Xu et al., 2012). There are many limitations, however,
on the use of optical images made it difficult to mapping tree plantation. One of these
is the similar spectral characteristics between natural forests and forest plantations
(Torbick et al., 2016), and the persistent cloud cover that occurs in images of tropical
areas (Kou et al., 2015). This widely known limitation of optical data in the tropics
remains an obstacle for automated mapping of tree plantations over large areas
(Torbick et al., 2016).
To address the limitations brought by cloud cover, Synthetic Aperature Radar
(SAR) images have been utilized to map tree plantations (e.g., Rosenqvist, 1996).
Torbick et al. (2016) contended that images acquired by SAR sensors is not only
advantageous because of the sensor’s ability to penetrate clouds, but also because of
its sensitivity to structural information (biomass, density, vertical layering) which are
useful in mapping tree plantations.
There have also been a growing number of studies that utilized the SAR-optical
image integration/fusion approach in mapping tree plantations. Chen et al. (2016)
argued that integration of both SAR (e.g, ALOS PALSAR) and optical data (e.g.,
MODIS and Landsat) could provide more comprehensive information about vegetation
canopy (which can be provided by optical data) and vegetation structure (which can
be provided by SAR data), which may improve mapping of forests through reducing
commission and omission errors. The study of Torbick et al (2016) has made use of
this comprehensive information when they fused Sentinel-1, Landsat-8, and PALSAR-
2 images to map plantations at the regional scale in Myanmar and Indonesia. Chen et
al (2016) integrated PALSAR 25-m and multi-temporal Landsat images to obtain
accurate map of tropical forests and rubber plantation in Hainan Island. In a similar
manner, Dong et al (2013) mapped deciduous rubber plantations through integration
of PALSAR and multi-temporal Landsat imagery and achieved highly accurately
results. Recently, Sentinel-1 and Sentinel-2 images have been used by Erinjery et al.
(2018) to address the highly challenging task of discriminating vegetation types of
tropical rainforests due to large environmental heterogeneity, high topographical
variability and near constant cloud cover.
16
On the other hand, the emergence of airborne Light Detection and Ranging
(LiDAR) technology has paved the way for accurate measurements of three-
dimensional forest vegetation characteristics, (Lefsky et al, 2002; Hudak et al., 2002;
Lovell et al., 2003; Coops et al., 2007). LiDAR sensors not only directly measure the
three-dimensional distribution of plant canopies but also subcanopy topography, such
that high-resolution topographic maps and highly accurate estimates of vegetation
height, cover, and canopy structure can be generated (Lefsky et al., 2002). LiDAR has
also been shown to accurately estimate Leaf Area Index (LAI) and aboveground
biomass even in areas where passive optical and active radar sensors are not feasible
to use (Lefsky et al., 2002). The integration of LiDAR data with remotely-sensed
images has also made possible the accurate mapping of individual tree location, height
and species in forestlands (Koukoulas and Blackburn, 2005; García et al., 2018). The
addition of Landsat images to LiDAR data was shown to improve results of mapping
forest canopy heights (García et al., 2018).
In addition to LiDAR, recent technological advancements have enabled the use
of Unmanned Aerial Vehicles (UAVs) as an alternative remote sensing platform for
forest mapping and inventory offering a distinctive combination of very high resolution
data capture at a significantly lower survey cost (Wallace et al., 2012). Using UAV for
tree plantation mapping purposes is found to be advantageous compared with satellite
remote sensing and aerial photogrammetry because UAV can be deployed easily and
frequently to satisfy the requirements of rapid monitoring, assessment and mapping in
natural resources at a user-defined spatio-temporal scale (Feng et al., 2015). The
sensor on board UAV acquires finer images which can capture details of ground
objects to assist for fully accurate vegetation mapping than that of sensors on-board a
satellite (Feng et al., 2015).
Plantation Stand Age Estimation with Remote Sensing
Due to its significant role in the Philippines’ log production, efficient and
accurate mapping of ITPs is essential for better monitoring and sustainable
management of these plantations. One of the most valuable parameters to monitor is
stand age, as it can provide information on the location of plantations at specific ages.
Such information can facilitate the estimation of stand density and forecasting of
volume of harvestable logs, including determining which of the stands are already
harvestable or will become harvestable in one or more years.
Stand age estimation over large areas is laborious and expensive if done
traditionally through conduct of field surveys such that remote sensing-based
approaches have become a popular alternative and a highly active area of research
over the last decades (Schumacher et al., 2020; Chen et al., 2018; Trisasongko and
Paull, 2020; Spracklen and Spracklen, 2021; Chen et al., 2012). The use of time series
remotely sensed data is a common approach in stand age estimation. Such approach
was employed by Chen et al. (2018) for rubber plantation age estimation where three
major steps were employed, consisting of building yearly Normalized Difference
Vegetation Index time series, modeling tree growth, and mapping the stand age of the
plantations. In Spracklen and Spracklen (2021), object-based classification of Sentinel-
2 time series data was applied to delineate acacia plantations into 6 age classes with
an overall accuracy of 70%, with young plantation consistently separated from older.
While time-series-based analysis of satellite data has been found to have great
potential in stand age estimation, its application for ITP age estimation maybe
problematic in areas like the Philippines where cloud cover is persistent in optical
17
satellite images. There is difficulty in using time series images because plantation
areas maybe cloud-covered in one or more images. On the other hand, ITPs like
Falcata plantations are widely planted by small holder farmers, and vast of lands are
converted into plantations over the years. This implies that Falcata plantations have
varying planting years, and hence, there stand ages are also varying. As such,
regression analysis techniques can be employed to the best available, cloud free
images to develop models that relate image bands with stand age (Chen et al., 2018).
Design and Development of GIS Database for
Industrial Tree Plantations
A Geographic Information System (GIS) is basically a computer-based system
that provides the different sets of capabilities to handle geo-referenced data such
as data input, data management (data storage and retrieval), manipulation and
analysis, and generation of output (Aronoff, 1989). GIS plays a crucial role in forest
resource inventories as this kind of activity generates enormous amount of data sets
that needs to be stored and effectively managed (Avanitis et al., 2000).
One good example application of GIS for ITP-related activity is reported by Ellis
et al. (2000) where GIS was used as database management application for
agroforestry planning and tree selection in the State of Florida, USA. For this
application, GIS was primarily used as a decision support system (DSS) aid in the
dissemination of information on different agroforestry opportunities and potential tree
species to landowners, farmers and extension agents. The GIS-based DSS enabled
the user to select a location of interest which is linked to spatial data on climate and
soils characteristics for the state of Florida. An important part of the application is a
relational database of over 500 trees and 50 tree attributes. According to the authors,
“the application is built with a modular and flexible framework in which spatial data of
different scales and/or regions as well as plant data may be easily incorporated” (Ellis
et al., 2000).
Tasoulas et al (2013) has also developed a GIS application for urban forestry
management planning which aims to help cities manage forestry projects efficiently
and reduce management costs. The application utilized Geoserver,
PostGIS/PostgreSQL, OpenLayers, GDAL, PROJ.4, and Entity Framework 5.0.
18
Methodology
Overview
The methodology adopted by the project includes four major activities, namely
ITP mapping, GIS database development, ITP suitability mapping, and spatial analysis
on the accessibility and efficiency of log production flow. Each of the activity is
discussed in detail in the next sub-sections.
At the minimum, the following ITP species which are used by the lumber, veneer,
and plywood industry (DOST PCAARRD, 2016) are targeted to be mapped:
Falcata (Paraserianthes falcataria (L.) Nielsen)
Gmelina/Yemane (Gmelina arborea Roxb)
Mangium (Acacia mangium Wild)
Bagras (Eucalyptus deglupta Blume)
ITP Mapping using Sentinel-2 Images
The species types, location and extent of ITPs in Caraga Region were mapped
using Sentinel-2 and high-resolution Google Earth images. Machine learning
techniques such as Support Vector Machine (SVM), Artificial Neural Networks (ANN),
and Random Forests (RF) were first evaluated for their applicability and accuracy in
mapping ITPs from Sentinel-2 image. Ground truth datasets (e.g., actual locations of
ITPs and other land-cover features) were used as reference in training the classifiers,
and in determining their accuracies. Images acquired by an Unmanned Aerial Mapping
System (UAS) were also used as ground truth data. The role of UAS in the mapping
process is especially important, particularly in accuracy assessment, as using a UAS
can lessen the time and manpower required to gather actual locations of ITPs.
The ITP mapping workflow consisted of 3 stages.
Coarse Spatial Resolution ITP Mapping
Coarse resolution maps of ITPs (spatial resolution of 10-m) of Caraga Region were
derived through processing and analysis of recent (year 2019-2020) Sentinel-2 images
using the best classifier (i.e., the classifier with the highest accuracy among SVM, ANN
and RF which was determined through an experimental analysis).
Fourteen (14) Sentinel-2 Level 2A images acquired during the period of August
2019 May 23, 2020 were subjected to supervised image classification and analysis
for coarse resolution mapping of ITPs in Caraga Region (Table 1, Figure 5). The
images do not require to be pre-processed as they are already orthorectified and pixel
values are already in terms of surface reflectance. Envi 5.0 software was used for
image classification and analysis.
Only the four (4) spectral bands of the image with 10-m spatial resolution were
used in the classification and analysis. These bands correspond to images formed
from the electromagnetic radiation (EMR) with visible to near-infrared (VNIR)
wavelengths that were reflected by earth surface features and captured by the sensors
of Sentinel-2 satellite.
19
The Maximum Likelihood algorithm was used for classifying each pixels of the
image into different land-cover types, including Falcata, Gmelina, Mangium and
Bagras (Table 2).
The decision to use Maximum Likelihood was supported by the results of an
experimental analysis which evaluated various classifiers that included Maximum
Likelihood, Linear Support Vector Machine, Polynomial Support Vector Machine,
Radial Basis Function (RBF) Support Vector Machine (SVM), Artificial Neural Network,
and Random Forest. The 6 classifiers were trained and used to classify a 9 km x 9km
image subset containing a portion of Butuan City (Figure 6). The accuracy of each
classifier to map Falcata and Non-Falcata were then evaluated using randomly-
selected 1000 Falcata and 1000 Non-Falcata points. The “best” classifier was
Maximum Likelihood, gaining the highest overall classification accuracy of 90.9%
(Figure 7). The output classification map for the test area is shown in Figure 8.
During its application to classify the 14 Sentinel-2 images, the Maximum
Likelihood classifier was trained using representative number of pixels of the land
cover classes listed in Table 1. This means that for each image, there are independent
sets of training data for each land cover class. In Envi software terminology, they are
all called “training regions of interests (ROIs)”. As an example, for image “T51NYJ-
55615”, there are separate training ROIs for bagras, banana, barren, etc. The same is
true for the rest of the images. The training ROIs are composed of pixels that
corresponds to the actual locations of a particular land cover class on the ground. For
example, if the land cover is bagras, then the pixels consisting the “bagras” training
ROIs are pixels that corresponds to bagras on the ground. To train and assess the
accuracy of the classifiers, ground truth datasets of ITP locations and other land-cover
features were utilized. They were obtained through geo-tagging using a handheld
GPS, and through acquisition of images using a UAS particularly for areas that are
difficult to access by walking.
All these training ROIs were saved as “end members”, and used as inputs for
implementation of an end-member classification approach using the Maximum
Likelihood algorithm in Envi 5.0 software. In this case, an “end member” is composed
of basic statistics of surface reflectance of each land cover class (minimum, maximum,
mean, standard deviation) including a covariance matrix indicating how the surface
reflectance of the land cover class varies across the four bands of the Sentinel-2
image. These statistics are needed by the Maximum Likelihood classifier to determine
the “likelihood” or the degree of probability that a pixel in the Sentinel-2 image is
belonging to a particular land-cover class. The end-member classification approach
was advantageous as it allows independent classification of all images using the same
sets of training ROIs or “end members”. The classified images were then subjected to
majority analysis as a post-classification analysis to remove spurious pixels (i.e.,
individual pixels belong to a class but are located within the extent of a larger class).
The post-classified images were then mosaicked into a single classified image
covering the Caraga Region. From this result, all those pixels classified as Falcata,
Gmelina, Mangium and Bagras were separated and preliminary area statistics were
calculated.
Due to limitations of the Maximum Likelihood classifier (i.e., capable only of
90.9% accuracy of classifying ITPs, in particular Falcata), the classification result
cannot be considered as final. There is high possibility of both underestimation and
overestimation of the mapped ITPs. To reduce these possibilities, the output of the
coarse resolution mapping was subjected to refinement and correction in Stage 2.
20
Table 1. Sentinel-2 L2A images and its date of acquisition.
Sentinel-2 Image
Date of Acquisition
T51NYJ-55615
August 10, 2019
T51NZJ-55615
August 10, 2019
T51PYK-55615
August 10, 2019
T51PYL-55615
August 10, 2019
T51PYL-40937
August 10, 2019
T51PYM-40937
August 10, 2019
T51PZK-55615
August 10, 2019
T51PZL-55615
August 10, 2019
T51PZL-40937
August 10, 2019
T51PZM-40937
August 10, 2019
T52PBQ-55615
August 10, 2019
T51NZJ-20443
November 15, 2019
T51PZK-20443
November 15, 2019
T51PZK-20445
May 23, 2020
Figure 5. A mosaic of 10-m spatial resolution Sentinel-2 Level 2A (L2A) images covering the Caraga
Region.
21
Table 2. Land cover classes considered in the Maximum Likelihood classification of Sentinel-2
images .
Bagras
Fishpond
Nipa
Banana
Forest
Oil Palm
Barren
Gmelina
Sago
Built-up
Grassland
Sand
Coconut
Mangium
Shrubs and Other Trees
Cropland
Mangrove
Water
Exposed Riverbed
Marsh Vegetation
Falcata
Mud
Figure 6. A subset of Sentinel-2 image used to evaluate the accuracy of various machine learning
classifiers for mapping ITPs.
22
Figure 7. Summary of overall accuracies of machine learning classifiers that were evaluated.
Figure 8. Image classification result for the test area using Maximum Likelihood.
90.9 89.75 89.75 89.8
76.6
89.65
65
70
75
80
85
90
95
Maximum
Likelihood Linear SVM Polynomial
SVM RBF SVM Neural Net Random
Forest
Over all Accuracy (%)
Classifier
23
High Resolution ITP Map Refinement and Validation
The ITP maps derived from the first stage were refined using high resolution
satellite images available in Google Earth (Figure 9). The stage 1 results were overlaid
in Google Earth and were validated by manually checking if the classification is correct
or not, or if the extent of the classified ITP needs to be refined. The refinements and
corrections were done only for areas of Caraga Region where ITPs were mapped in
Stage 1. Manual digitizing of ITPs was done in cases where the classifiers failed to
accurately detect ITPs.
Ideally, high resolution images would provide the best result for ITP mapping.
However, it is often difficult to know where to start mapping the ITPs from the high
resolution images. This is in addition to the fact that manual digitizing of ITP location
and extents in high resolution images would require considerable amount of time and
labor. Given that there is already a coarse resolution map of ITPs from the Sentinel-2
images, the ITP refinement and validation can be focused on areas where there is high
potential for ITPs to occur.
To aid in the refinement and validation, field validation surveys were conducted.
The data from these surveys were also used to generate interpretation keys of how
ITPs appear in high resolution Google Earth images. These interpretation keys were
used as reference by the team doing the refinement and correction. Unfortunately, due
to the travel restrictions and health concerns posed by the COVID-19 Pandemic, the
field surveys were not completed in many parts of Caraga Region. This resulted to
being not able to completely validate, refine and correct the ITP location and extents
determined from Sentinel-2 images.
Since the ITP Maps are especially important pieces of reference information for the
ITP industry, providing them the Sentinel-2-derived ITP Map may be inappropriate
considering that it was not thoroughly validated. For this reason, the ITP map
containing only the portions that were validated, refined and corrected will be released
to the end user.
Figure 9. Illustration of the refinement and correction of ITP Map. The image on the left shows the result
of the Sentinel-2 image classification (in red). The yellow polygons on the image on the right is the
refined/corrected extents.
24
Detailed Mapping of ITPs using Unmanned Aerial
System (UAS)
This stage of the mapping process involves acquiring images using a UAS for
3D mapping of ITPs. Using Structure-from-Motion photogrammetric approach, 3D
characteristics of ITPs such as stand density, tree heights and stand volume) can be
extracted from the images.
The 3D mapping was limited only to certain areas due to the limitations of the
UAS (e.g., allowable flying height, battery/power capacity, etc). This activity was more
of an exploratory application of UAS for 3D ITP mapping.
For this activity, two UAS were used, namely Phantom 4 RTK and Mavic2Pro.
The UAS mapping was conducted in 33 sites in Butuan City, Agusan del Norte (Figure
11). These sites represent Falcata plantations that are of different growth stages.
DroneDeploy was used for flight planning. Reconnaissance was performed first, prior
to the actual image acquisitions. All images were processed using Agisoft Metashape
software. The following very high-resolution data products were generate: 3D point
cloud, Digital Surface Model (DSM), and orthomosaic. These products were used as
refinement and validation data for Sentinel-2 -based ITP mapping, for plantation age
estimation analysis, as well as for additional R&D activities such as tree crown
delineation.
Figure 10. UASM image acquisitions using the Phantom 4 RTK and Mavic2Pro drones.
25
Table 3. Location of sites, number of plantations, and DroneDeploy flight parameters for UAS mapping.
Sit
e
Barangay
No. of
Plantatio
n
Plantatio
n Age
Flight
Directio
n
(degree
s)
Estimate
d No. of
Images
Total
Acquisitio
n Time
Total
Area
(has.
)
Mappin
g Flight
Speed
(m/s)
min
.
sec
.
1
Amparo
1
7
-80
38
4
28
2
8
2
Ampayon
1
1
93
31
4
8
1
8
3
Basag
4
4,2,4,6
-92
197
13
48
11
8
4
Basag
1
4
180
37
4
22
2
8
5
Basag
1
5
180
22
3
45
1
8
6
Basag
1
2
93
39
4
44
2
8
7
Basag
2
7,5
-80
123
8
52
7
8
8
Basag
1
6
-7
136
8
57
7
8
9
Basag
1
3
-6
36
4
34
2
8
10
Cabcabon
2
6,6
-180
115
9
2
6
8
11
Camayaha
n
1
10
-93
80
7
31
4
8
12
Lemon
2
2,4
-2
76
6
54
4
8
13
Lemon
1
8
18
98
7
27
6
8
14
Manila de
Bugabus
1
4
-83
35
4
34
2
8
15
Dulag
1
10
175
120
8
59
7
8
16
Dulag
1
8
180
72
6
31
4
8
17
San Mateo
2
8, 3
-90
71
6
47
4
8
18
San Mateo
1
1
41
59
5
48
3
8
19
Sumilihon
3
5,3,3
1
138
9
58
8
8
20
Sumilihon
1
5
6
42
4
49
2
8
21
Sumilihon
2
6,7
86
100
7
45
5
8
22
Pigdaulan
1
3
-88
46
4
57
2
8
23
Pigdaulan
1
5
173
102
7
33
5
8
24
Maguinda
1
1
-58
16
3
23
0
5
25
San Mateo
1
1
-58
16
3
23
0
5
26
Maguinda
1
1
-154
16
3
3
0
6
27
Florida
1
2
149
25
3
50
1
8
28
Basag
1
2
-4
16
3
19
1
8
29
Basag
1
7
-104
33
4
17
2
8
26
30
Basag
1
7
-4
16
3
19
1
8
31
Pigdaulan
1
8
-143
16
3
23
1
7
32
Dulag
1
8
159
18
3
21
1
8
33
Dulag
1
10
-180
19
3
36
1
8
Figure 11. Map of selected Falcata plantations for UAS image acquisitions.
Figure 12. Example flight plans prepared using DroneDeploy.
27
ITP Stand Age Estimation: Case Studies for Falcata
Plantations Using Sentinel-2 and Landsat 8 OLI
Images
Two case studies were conducted to test a satellite remote sensing approach
for plantation stand age estimation. We focused on Falcata plantations, and we
selected Butuan City as our study site (Figure 13). Specifically, we utilized Sentinel-2
and Landsat 8 OLI images.
Figure 13. Map of the case study site for the Falcata stand age estimation using remote sensing. Shown
in detail are examples of the 148 Falcata plantations used in the analysis.
28
Case Study 1: Falcata Age Estimation Using Sentinel-2
SENTINEL-2 IMAGE
The study made use of Level-2A Sentinel-2 image dataset acquired on August
10, 2019 which was used in the ITP mapping. The dataset is composed of 10-m, 20-
m, and 60-m spatial resolution image bands. However, a subset of the 10-m bands
enclosing the study area was only used in subsequent steps. Vegetation indices (Vis)
were also calculated and include the Atmospherically Resistant Vegetation Index
(ARVI), Difference Vegetation Index (DVI), Green Normalized Difference Vegetation
Index (GNDVI), Normalized Difference Vegetation Index (NDVI), Normalized
Difference Water Index (NDWI), Perpendicular Vegetation Index (PVI), Simple Ratio
Index (RVI), Soil Adjusted Vegetation Index (SAVI), and Sentinel-2 Red-Edge Position
Index (S2REP).
FALCATA STAND AGE DATA REGRESSION ANALYSIS
Field surveys were conducted on October 20-29, 2020 to gather geographic
locations and age of 136 homogeneous Falcata plantations or tree farms in the study
area. Data products generated from the UAS image acquisitions were also used to
supplement the field survey data.
Plantation/tree farm owners or caretakers present during the survey were
asked of the planting dates as basis for the age. The ground truth data points were
overlaid on Google Earth images (dated 19 August 2020; same date as the Sentinel-
2 image) to create polygon shapefiles which were then rescaled to 10-m to fit the
Sentinel-2 image spatial resolution. The rescaled polygon shapefiles were carefully
examined to remove pixels that are certainly not Falcata. The polygon shapefiles were
then divided proportionally and in a random manner into ~60% (n=81) for correlation
regression analysis and ~40% (n=55) for validation for each age classification (Table
4).
Table 4. Stand area polygon statistics for the 10-m Sentinel-2 image correlation and
regression analysis.
Stand Age,
years
No. of Polygons
Number of Pixels/Polygon
Min
Max
Sum
Mean
1
4
5
123
241
40.17
2
8
5
81
262
26.2
3
14
9
123
756
42
4
40
9
337
2190
53.42
5
27
8
700
2214
82
6
29
13
332
2025
67.5
7
6
19
115
330
55
8
5
19
109
393
56.14
10
3
26
306
367
122.33
CORRELATION AND REGRESSION ANALYSIS
The mean values of band reflectance and VIs were calculated for each polygon
and utilized in the correlation and regression analysis. Pearson’s correlation
coefficients of stand age versus band reflectance and VIs were computed to determine
the relationship between the two variables. For stand area model development,
Univariate exponential regression analysis was selected as it best represented the
relationship between the stand age and the image-derived variables. Such decision
29
was aided by scatterplots which indicated that stand age revealed a curvilinear
relationship with most of the bands and VIs, an observation like that of [8]. The
regression models developed were then evaluated based on their coefficient of
determination (R2) values, and then used to predict stand age using the validation
dataset. Root Mean Square Errors (RMSE) were computed for each model as basis
for evaluated their accuracy.
Case Study 2: Falcata Age Estimation Using Landsat 8 OLI Image
LANDSAT 8 OLI IMAGE
A 2021 Landsat-8 image (path=112, row=54; acquired at 04/03/2021) was
used in this case study. The image used contains the bands 1 to 7. Four vegetation
indices bands were also generated in this study, namely NDVI, GNDVI, RVI, and DVI.
Other studies have considered the incorporation of VIs in estimating stand age
information. Moreover, elevation and slope were also considered additional data layers
in this study. The value of a single pixel for a Falcata stand was then extracted from
these data layers for the prediction (n=179) and validation (n=125) dataset.
MULTIVARIATE REGRESSION
The backward data entry method in JASP software was employed in this study
to develop the multivariate regression models. This data entry method has been
considered to be useful for fine-tuning regression models to select the best predictors
available (Goss-Sampson, 2003). Before the model development, the data outliers
were removed through the one-sigma rule. Linear and nonlinear regression models
were developed using the following different initial input explanatory variables for the
Falcata stand age: 1) bands 1 to 7; 2) bands 1 to 7 and 4 derived VIs; 3) bands 1 to 7,
4 derived VIs, and the elevation and slope data layers. For the nonlinear regression
models, the exponential and reciprocal transformations were considered to be tested
to account for the curvilinear relationship between the stand age and Landsat-8 bands.
The validation dataset was then applied to the model to assess the prediction
accuracy, with their RMSE values computed. Scatter plots of the actual age versus the
predicted age by the models were also plotted to observe the performance of the
models.
30
Development of an online/web-based GIS database
An online web-based GIS database was developed using a combination of
database management system (DBMS) software and web GIS technologies such as
PostgreSQL+PostGIS, Geoserver, Web Map Service (WMS), Web Feature Service
(WFS), and Leaflet, among others. The database includes the ITP types, location and
extent that were derived from the ITP Mapping activity. Other spatial and non-spatial
datasets were integrated such as administrative boundaries, land classifications, road
network, wood processing plant locations, and other relevant datasets.
A series of field data collection activities was conducted to validate and obtain
the geographic coordinates of WPPs and furniture makers/stores. A list from the
Department of Trade and Industry (DTI) was used as reference for this activity.
The database was designed such that it can be used for efficient management
and monitoring of ITPs in the region, including mapping out the activities and services
of the various key players, the flow of product, including the role of ITP farms in carbon
sequestration and disaster risk reduction; market projections; generation of economic
forecast models in relation to climate change scenarios; and for other purposes
requiring wood industry-based information.
Database Development Using PostgreSQL+PostGIS
The database was developed using the existing PostgreSQL database
management system software installed in the Mindanao Integrated Data Sharing
Environment (MInDSEt) data server that is being hosted and maintained by the Caraga
Center for Geo-Informatics (CCGeo) of Caraga State University.
The server runs in Ubuntu 20.04 Long Term Support (LTS), with PostgreSQL
12.9. The PostGIS extension was installed to accommodate spatial datalayers. A blank
PostgreSQL database (named “geoitp”) was first created in the server. Then, GIS
shapefiles of each of the spatial data layers were then imported to the database using
the DB Manager in QGIS 3.6.15 (Figure 14). An example geodatabase table for
mapped Falcata plantations is shown in Figure 15.
Figure 14. Illustration of the various spatial data layers imported into the geodatabase as tables using DB
Manager of QGIS 3.16.15. The database is hosted at MInDSET at IP address 45.202.17.189.
31
Figure 15. A PostgreSQL database table for mapped Falcata plantations.
Enabling WMS and WFS Access to the Geodatabase
For easier public access, especially by agencies that requires data and
information on ITPs, enabling Web Map Service (WMS) and Web Feature Service
(WFS) access to the geodatabase is necessary.
To enable such capability, the geodatabase spatial layers were published as
WMS and WFS layers using MInDSET’s Geoserver at
http://mindset.ccgeo.info:8080/geoserver/ (Figure 16). For WMS, each of the layers
were symbolized/styled in Geoserver prior to publishing.
Figure 16. MInDSEt’s Geoserver where all ITP Geodatabase spatial layers (stored in the PostgreSQL
database) were published as either WMS or WFS layers.
32
ITP Geodatabase Visualization Portal
A web visualization portal using Leaflet (https://leafletjs.com/) was also
developed as one of the ways to visualize the various geodatabase layers. This is an
interactive web map where users can display mapped ITP plantations or tree farms
together with other spatial layers like WPP locations, land classifications (e.g.,
Alienable and Disposable Lands, Timber Lands, Protected Areas, etc.). This portal is
hosted externally from the MInDSEt and hence utilized only the GeoJSON files of the
spatial data layers.
ITP-related Spatial Data Layers, Maps and Other Project Outputs Access
via MInDSEt
Individual files of spatial data layers, maps, and other project outputs were
uploaded to the Mindanao Integrated Data Sharing Environment (MInDSEt) Portal of
Caraga State University (Figure 17) to ensure data security and integrity and for easier
access by its target end-users. These files are in the form of zippe GIS shapefiles, MS
Excel files, PDF of reports, and JPEG files of maps, among others.
Figure 17. The MInDSEt portal where the geodatabase data layers and other project outputs can be
accessed. The portal can be accessed at http://mindset.ccgeo.info:82/.
33
ITP Suitability Modeling and Mapping
Data from the web-based GIS database, particularly the spatial locations of
existing ITP species, were utilized to determine areas suitable for establishing new
ITPs particularly for Falcata, Mangium, Gmelina, and Bagras species. The suitability
mapping includes determining bioclimatically suitable areas, as well as biophysically-
suitable areas. This was performed using MaxEnt model.
The MaxEnt Model
As a presence-only Species Distribution Models (SDM), MaxEnt links species
locations (in this study, the ITP species locations) with environmental conditions and
then geographically-projecting where the species are likely to be found based on
suitable environmental conditions (Heumann et al., 2011; Heumann et al., 2013).
Moreover, as a general-purpose method for making predictions or inferences from
incomplete information (Philipps et al., 2006), MaxEnt predicts the potential distribution
of species using only the occurrence data and generates a numerical output (a map)
with values ranging from 0 to 1, which indicate the lowest to highest habitat suitability,
respectively (Xu et al., 2020). The central idea behind MaxEnt is to estimate a target
probability distribution by finding the probability distribution of maximum entropy (i.e.,
that is most spread out, or closest to uniform), subject to a set of constraints that
represent our incomplete information about the target distribution(Philipps et al.,
2006). MaxEnt has gained popular applications in predicting species distribution under
current and future climate scenarios, (e.g., Chakraborty, et al., 2016; Li et al., 2020;
Coban et al., 2020). In land suitability modeling and mapping, MaxEnt has been used
in assessing the suitable cultivation areas for Scutellaria baicalensis in China (Xu et
al., 2020) and in modeling crop suitability for upland and lowland crops in Thailand
(Heumann et al., 2011).
ITP Presence Data
Table 5 provides a summary of ITP presence data points used in MaxEnt
modeling and suitability mapping. Figure 18 and Figure 19 show the locations of these
presence data points across Caraga Region. These data were extracted from the ITP
Geodatabase.
For each species, the datapoints were divided into 2 sets, one for model
training and another set for validation of the suitability maps. The points were randomly
selected from the centroids of mapped ITP plantations/tree farms with at least 1
hectare in area using the “Random Point Generator” extension (Jenness Enterprises)
of ArcView GIS 3.2 software. The distances between points were set to be not to less
than 1 kilometer. A comma space value (CSV) file was generated containing the
Universal Transverse Mercator (UTM 51) WGS 1984 grid coordinates of each point for
use in MaxEnt.
Table 5. ITP Presence Data used in MaxEnt suitability modeling and mapping.
ITP Species
Training Data
Points
Validation Data
Points
Total
Falcata
1,065
1,060
2,125
Bagras
6
6
12
Gmelina
30
28
58
Mangium
40
38
78
34
Figure 18. Map showing the presence data points used in the MaxEnt modeling and suitability mapping
of Bagras, Gmelina and Mangium.
Figure 19. Map showing the presence data points used in the MaxEnt modeling and suitability mapping
of Falcata.
35
Environmental Variables
A total 26 environmental variables were identified, including 19 bioclimatic
variables, solar radiation, wind speed, 3 topographic variables (elevation, slope and
aspect), soil type, and land cover type (Table 6; Figure 20). These are variables are
considered to interact and support optimal development and facilitate the spatial
distribution of Falcata plantations within the landscape.
The 19 bioclimatic variables (BIO1-BIO19) with ~1 km spatial resolution and in
GeoTIFF format are derivatives of the WorldClim version 2.1 climate data for 1970-
2000; they were downloaded from https://www.worldclim.org/data/worldclim21.html.
The solar radiation and wind speed data layers were also downloaded from the same
website.
A Digital Elevation Model (DEM) derived from SAR interferometry with spatial
resolution of 5 meters was used to generate the topographic layers. The DEM obtained
from the Philippines’ National Mapping and Resource Information Authority (NAMRIA).
The DEM was resampled to 1 km to be consistent with the resolution of the bioclimatic,
solar radiation, and wind speed layers. The resampled DEM served as the elevation
layer while slope and aspect layers were derived from it using tools available in ArcGIS
10.8 software.
The soil and land-cover type layers (in shapefile format and converted to 1km
rasters) were obtained from the Department of Agriculture - Bureau of Soils and Water
Management (DA-BSWM) and NAMRIA, respectively. All the data layers were layer-
stacked using ArcGIS 10.8 software to ensure that they have the same spatial
resolution (1 km) and coverage, as well as raster dimensions. Each layer was then
exported to ASCII (*.asc) format in preparation for MaxEnt modeling.
Environmental Variable Selection
It is common practice that collinearity between environmental variables must
be quantified such that only those variables that are not collinear (i.e., not correlated)
with each other are used in the SDM. Collinearity “may cause over-adjustment
problems and is expected to increase uncertainty and decrease the statistical power
of the model” (Rojas Briceño, 2020). To facilitate the collinearity analysis, Pearson’s
correlation coefficients (r) between the variables were calculated; variables with a r ≥
0.7 are considered correlated, and only one of them should be selected and included
in the MaxEnt model (Rojas Briceño, 2020; Garcia et al., 2013). The variable selection
procedure reported in Garcia et al. (2013) was used to minimize error in choosing the
final set of environmental variables or predictors. It consists of MaxEnt pre-model or
initial runs using all environmental variables. Based on the output of the pre-model
runs, only one variable (with the highest percent contribution) from a set of highly
cross-correlated variables was included in the final Maxent model (Garcia et al., 2013).
Thus, only 15 out of 26 variables were kept for the final Maxent model for Falcata, 10
for the Bagras MaxEnt model, 14 for the Gmelina MaxEnt model, and 14 for the
Mangium MaxEnt model (Table 6).
36
Table 6. List of environmental variables for MaxEnt modeling.
Variable
Name
Description
Falcata
MaxEnt
Model
Inclusion
Bagras
MaxEnt
Model
Inclusion
Gmelina
MaxEnt
Model
Inclusion
Mangium
MaxEnt
Model
Inclusion
BIO1
Annual Mean Temperature
BIO2
Mean Diurnal Range (Mean
of monthly (max temp - min
temp))
BIO3
Isothermality [(BIO 02/BIO
07)x100)]
BIO4
Temperature Seasonality
(standard deviation x 100)
BIO5
Maximum Temperature of
Warmest Month
BIO6
Minimum Temperature of
Coldest Month
BIO7
Temperature Annual
Range (BIO 05 BIO 06)
BIO8
Mean Temperature of
Wettest Quarter
BIO9
Mean Temperature of
Driest Quarter
BIO10
Mean Temperature of
Warmest Quarter
BIO11
Mean Temperature of
Coldest Quarter
BIO12
Annual Precipitation
BIO13
Precipitation of Wettest
Month
BIO14
Precipitation of Driest
Month
BIO15
Precipitation Seasonality
(Coefficient of Variable)
BIO16
Precipitation of Wettest
Quarter
BIO17
Precipitation of Driest
Quarter
BIO18
Precipitation of Warmest
Quarter
BIO19
Precipitation of Coldest
Quarter
SRAD
Solar radiation
WIND
Wind speed
LCOV
Land-cover type
SOIL
Soil type
ELEV
Elevation
ASPECT
Aspect
SLOPE
Slope
37
Figure 20. Examples of environmental variables used in MaxEnt Modeling.
MaxEnt Modeling
MaxEnt Version 3.4.1, downloaded from
https://biodiversityinformatics.amnh.org/open_source/maxent/, was used in this study.
For each ITP species, the model run was using 15 replicates with 5000 maximum
iterations each to have adequate time for convergence. Each iteration uses a random
partition of the presence data through bootstrapping (70% for modeling, 30% for
testing), with convergence threshold of 0.00001 and 10,000 maximum background
points. Random seeding as well as jack-knife test for variable importance were also
enabled.
Each of the model’s accuracy and performance was evaluated using the Area
Under the Curve (AUC) statistic which is calculated from the Receiver Operating
Characteristic (ROC) based on training and testing data. The following differentiation
of performance levels was used: excellent (>0.9), good (0.80.9), accepted (0.70.8),
poor (0.60.7), and unsatisfactory (<0.6) (Rojas-Briceño et al., 2020).
38
A logistic output format was also selected to generate a raster grid of
continuous probability values ranging from 0 to 1 for each iteration. The average of the
15 output rasters were averaged and used for suitability mapping, with the following
suitability classes and probability value ranges applied: ‘high’ (>0.6), ‘moderate’ (0.4–
0.6), ‘low’ (0.20.4), and not suitable (<0.2) (Zhang, K. et al., 2019).
Spatial Analysis of Log Production Flow
Data on locations of ITP from the geodatabase was utilized to conduct accessibility
spatial analysis of log production flow. The analysis focused on Falcata plantations as
this species is the widely planted and has the highest log production among all log
producing species in Caraga Region.
Network partitioning and optimal routing procedures available in ArcGIS 10.8
software were applied for this purpose. The analysis made use of the centroid of ITP
plantations, land transportation (road) network, and locations of WPPs and furniture
makers.
Detailed road network of Caraga Region (provided by the Phil-LiDAR 1 and
GeoSAFER Agusan Projects of Caraga State University, and updated by the project)
was used in the network analysis. Based on their classification, each road was
assigned speed limits for motor trucks as prescribed in Republic Act No. 4136
1
. The
assumption here is that only motor trucks are used to transport logs from plantations
to WPPs. This speed limit was then used together with the road segment length (from
one node to the next node) to compute travel time, in terms of minutes. These travel
times were then used as “impedance” in the network analysis.
Table 7. Speed limits based on RA 4136.
Road Classification
Maximum Allowable Speeds for Motor
Trucks (kph)
Open country roads (e.g., national
roads)
50
City and municipal roads/streest
30
Barangay roads
20
In particular, the following analysis were performed in ArcGIS 10.8:
Service Area Analysis of WPPs The output of this analysis are polygons
showing the areas that can be serviced by the WPPs within a certain time
limit. For example, with this analysis, it is possible to determine how many
Falcata plantations and furniture makers are within 15 minutes, 30 minutes,
45 minutes, 1 hour, 2 hours, etc., of existing WPPs. Hence, the accessibility
of existing WPPs can be evaluated, and the plantation locations that can
be serviced by these WPPs can be identified.
Location-Allocation Analysis (Maximize Capacitated Coverage) This
analysis will assign a particular plantation location to the nearest WPP that
can accommodate the estimated volume of logs produced from that
plantation without exceeding the WPP’s capacity. In case the nearest WPP
can no longer accommodate a plantation, that plantation will be assigned
to the next nearest plantation whose capacity is not yet exceeded. The
1
RA 4136. An Act to Compile the Laws Relative to Land Transportation and Traffic Rules, to
Create a Land Transportation Commission and for other Purposes.
39
process is iterative until all WPPs have been assigned with plantations.
Since each WPP have limited capacities, it is possible that not all plantation
locations will be assigned to a WPP. The analysis was set such that the
WPPs will be allocated with plantations in order of decreasing ALR (i.e., the
WPP with largest ALR will be allocated first, followed by the WPP with the
second largest ALR, and so on).
In the location-allocation analysis, the capacity being referred is the WPPs
annual log requirement, expressed in cubic meters per year. Using such type of
capacity will complicate and make impossible to perform the location-allocation
analysis. It is because when ArcGIS executes this procedure, it assumes that the
capacity assigned to a WPP corresponds to the number of point locations (not volume).
As an illustration, if a WPP capacity value is 100, the analysis will look for 100 locations
that are nearest to it.
To circumvent this complexity, the WPP capacity was transformed into
hectares where a hectare has an equivalent volume of logs produced annually.
Following Micosa-Tandug (2012), a hectare of Falcata plantation was estimated to
produce 39 m3 of logs on 10-year rotations. Hence, a WPP with 91,250 m3 of annual
log requirement will have an equivalent annual capacity of 91250/39 = 2339.74 has =
2340 has (rounded-off). In some WPPs, the annual log requirement is not known, and
the only information available are the daily rated capacity (DRC) and annual capacity.
In such case, we made use of the annual capacity as equivalent to the annual log
requirement.
Having assigned a unit of hectare for the WPP capacity, the next step is to
convert the Falcata plantation polygons into individual points such that each point is
equivalent to 1 hectare of plantation. Since the Falcata plantation polygons have
varying area, they were rasterized to 1 hectare (or 100 m x 100 m) pixel size. The
centers of all pixels were then exported as point vector file and used as Falcata
plantation locations in the location-allocation analysis. This means that each point
corresponds to 1 hectare of Falcata plantation. This procedure discretizes the Falcata
plantations from a continuous polygon of varying area to individual points
corresponding to the same area of plantation.
40
Major Results and Findings
ITP Mapping Results
Coarse Resolution Mapping Using Sentinel-2 Image
Figure 21 shows the result of the coarse resolution mapping of ITPs in Caraga
Region using Sentinel-2 images. It can be noticed from the map that there is a large
distribution of Falcata, Gmelina and Mangium. The preliminary (unrefined) statistics
show that there are approximately 147,443.94 hectares of Falcata, 130,086.27
hectares of Gmelina, 33,590.40 hectares of Mangium, and 7,929.06 hectares of
Bagras. These statistics should be used with caution as the ITP map was not
thoroughly validated on the field due to the travel restrictions by the COVID-19
Pandemic. A refined and corrected version of these maps and statistics are discussed
in the next section.
41
Figure 21. Results of stage 1 (coarse resolution) mapping of ITPs in Caraga Region using Sentinel-2
images.
Table 8. Preliminary statistics of ITPs in Caraga Region based on classification and analysis
of Sentinel-2 images.
Province
Falcata
Gmelina
Mangium
Bagras
Agusan del Norte
18,617.80
18,585.45
2,465.01
994.55
Agusan del Sur
73,365.96
64,607.14
12,542.21
3,320.53
Dinagat Islands
1,448.61
3,394.89
346.74
89.24
Surigao del Norte
11,488.38
10,756.13
1,190.15
558.58
Surigao del Sur
42,523.19
32,742.66
17,046.29
2,966.16
Total Area
147,443.94
130,086.27
33,590.40
7,929.06
42
Refined and corrected the ITP Maps Using High Resolution Google Earth
Images
Figure 22 shows the results of the refinement and correction of the Sentinel-2
image-derived ITP map using high resolution Google Earth images. Only the portions
were there is high confidence of determining whether a mapped location is an ITP are
included in the map. From this, a total of 93,636.50 has. of Falcata, 7.07 has. of
Bagras, 5.65 has. of Gmelina, and 1,668.32 has of Mangium were confirmed. It
should be noted that these areas may still increase after conduct of more field
validations.
Since the ITP Maps are important pieces of reference information for the ITP
industry, providing them the Sentinel-2-derived ITP Map may be inappropriate
considering that it was not thoroughly validated. For this reason, the ITP map
containing only the portions that were validated, refined and corrected will be released
to the end user.
Table 9 provides a summary of these statistics. It should be noted again that
these statistics are only for those Sentinel-2 mapped plantations that have been refined
and validated. Values indicated by asterisks (*) do not necessarily mean that a
particular ITP species is not present in those localities.
Table 9. Updated statistics of ITPs in Caraga Region based on refinement and correction o the
results of the classification and analysis of year 2019-2020 Sentinel-2 images. Values are in
hectares.
Province
Falcata
Bagras
Gmelina
Mangium
Agusan del Norte
14,593.16
*
1.63
20.15
Agusan del Sur
55,337.99
2.40
1.07
1,102.43
Dinagat Islands
20.32
*
*
16.84
Surigao del Norte
2,376.39
4.67
2.82
197.37
Surigao del Sur
21,308.64
*
0.13
331.53
Total Area
93,636.50
7.07
5.65
1,668.32
*Refinement and validation were not conducted in these localities.
43
Figure 22. Result of refinement and correction of the Sentinel-2 ITP Map using high resolution Google
Earth images.
44
Illustrated Statistics of ITPs in Caraga Region based on Refinement and
Correction
The following figures show the maps and statistics of ITPs in Caraga Region
based on the result of refinement and correction.
FALCATA
Agusan del Sur (ADS) was found to have largest area planted with Falcata at
55,337.99 hectares (Figure 23). Among the top 10 municipalities with the largest area
of Falcata plantations (Figure 24) are Prosperidad (ADS), Bayugan City (ADS), Butuan
City (Agusan del Norte), Esperanza (ADS), Tagbina (Surigao del Sur or SDS), Sibagat
(ADS), Loreto (ADS), Buenavista (Agusan del Norte), San Luis (ADS), and Veruela
(ADS).
Figure 23. Province-level aggregated area statistics of Falcata plantations.
45
Figure 24. Municipal-level aggregated area statistics of Falcata plantations.
GMELINA
Surigao del Norte has the largest area of Gmelina plantations or tree farms at 2.82
hectares (Figure 25). The municipal-level aggregated statistics are illustrated in Figure
26.
46
Figure 25. Province-level aggregated area statistics of Gmelina plantations.
47
Figure 26. Municipal-level aggregated statistics of Gmelina plantations in Caraga Region. Areas in white
were not covered in the refinement and correction.
48
MANGIUM
Agusan del Sur has the largest area of Mangium plantations or tree farms at 1,102.43
hectares (Figure 25). The municipal-level aggregated statistics are illustrated in Figure
26.
Figure 27. Province-level aggregated area statistics of Mangium plantations.
49
Figure 28. Municipal-level aggregated statistics of Mangium plantations in Caraga Region. Areas in
white were not covered in the refinement and correction.
50
BAGRAS
Agusan del Norte has the largest area of Bagras plantations or tree farms at 4.67
hectares (Figure 29). The municipal-level aggregated statistics are illustrated in Figure
30.
Figure 29. Province-level aggregated area statistics of Mangium plantations.
51
Figure 30. Municipal-level aggregated statistics of Bagras plantations in Caraga Region. Areas in white
were not covered in the refinement and correction.
52
Detailed Maps and Statistics of ITPs in Caraga Region
This section presents the detailed maps and statistics of ITPs in Caraga Region
based on the result of refinements and corrections.
Figure 31. Refined mapped ITPs in Caraga Region.
53
Figure 32. Refined mapped ITPs in the province of Agusan del Norte.
54
Figure 33. Refined mapped ITPs in the province of Agusan del Sur.
55
Figure 34. Refined mapped ITPs in the province of Dinagat Islands.
56
Figure 35. Refined mapped ITPs in the province of Surigao del Norte.
57
Figure 36. Refined mapped ITPs in the province of Surigao del Sur.
58
Table 10. City or municipal-level aggregated area statistics of mapped ITPs in Caraga Region
(in hectares).
Province
City or Municipality
Falcata
Bagras
Gmelina
Mangiu
m
AGUSAN DEL NORTE
BUENAVISTA
3,468.03
*
0.88
1.51
AGUSAN DEL NORTE
BUTUAN CITY (Capital)
5,811.23
*
0.62
9.52
AGUSAN DEL NORTE
CITY OF CABADBARAN
673.82
*
*
*
AGUSAN DEL NORTE
CARMEN
221.97
*
*
*
AGUSAN DEL NORTE
JABONGA
217.48
*
0.07
*
AGUSAN DEL NORTE
KITCHARAO
152.58
*
*
0.81
AGUSAN DEL NORTE
LAS NIEVES
2,485.16
*
*
8.31
AGUSAN DEL NORTE
MAGALLANES
25.06
*
*
*
AGUSAN DEL NORTE
NASIPIT
480.50
*
*
*
AGUSAN DEL NORTE
SANTIAGO
565.11
*
*
*
AGUSAN DEL NORTE
TUBAY
260.00
*
0.06
*
AGUSAN DEL NORTE
REMEDIOS T. ROMUALDEZ
232.21
*
*
*
AGUSAN DEL SUR
CITY OF BAYUGAN
7,004.90
*
*
*
AGUSAN DEL SUR
BUNAWAN
2,862.04
1.61
0.10
180.65
AGUSAN DEL SUR
ESPERANZA
5,521.35
*
*
40.39
AGUSAN DEL SUR
LA PAZ
2,757.30
*
*
25.31
AGUSAN DEL SUR
LORETO
3,531.03
*
*
0.48
AGUSAN DEL SUR
PROSPERIDAD (Capital)
11,852.95
*
*
38.98
AGUSAN DEL SUR
ROSARIO
2,565.67
*
*
56.35
AGUSAN DEL SUR
SAN FRANCISCO
1,739.61
*
*
5.02
AGUSAN DEL SUR
SAN LUIS
3,397.46
*
*
105.46
AGUSAN DEL SUR
SANTA JOSEFA
468.59
*
*
-
AGUSAN DEL SUR
TALACOGON
2,647.00
*
0.97
177.38
AGUSAN DEL SUR
TRENTO
2,934.31
0.79
*
472.40
AGUSAN DEL SUR
VERUELA
2,917.65
*
*
*
AGUSAN DEL SUR
SIBAGAT
5,138.15
*
*
*
59
Province
City or Municipality
Falcata
Bagras
Gmelina
Mangiu
m
SURIGAO DEL
NORTE
ALEGRIA
131.59
*
*
0.85
SURIGAO DEL
NORTE
BACUAG
151.17
*
*
*
SURIGAO DEL
NORTE
BURGOS
1.44
*
*
*
SURIGAO DEL
NORTE
CLAVER
66.52
*
*
0.93
SURIGAO DEL
NORTE
DAPA
6.78
*
*
*
SURIGAO DEL
NORTE
DEL CARMEN
32.03
*
*
23.12
SURIGAO DEL
NORTE
GENERAL LUNA
6.54
*
*
*
SURIGAO DEL
NORTE
GIGAQUIT
178.28
*
*
*
SURIGAO DEL
NORTE
MAINIT
437.76
*
0.57
*
SURIGAO DEL
NORTE
MALIMONO
26.40
*
*
*
SURIGAO DEL
NORTE
PILAR
13.58
*
*
2.12
SURIGAO DEL
NORTE
PLACER
266.82
*
*
33.51
SURIGAO DEL
NORTE
SAN BENITO
12.66
*
*
*
SURIGAO DEL
NORTE
SAN FRANCISCO (ANAO-
AON)
86.28
*
*
*
SURIGAO DEL
NORTE
SAN ISIDRO
52.79
*
*
0.57
SURIGAO DEL
NORTE
SANTA MONICA (SAPAO)
7.34
*
*
*
SURIGAO DEL
NORTE
SISON
197.43
1.75
0.21
13.07
SURIGAO DEL
NORTE
SOCORRO
24.06
*
*
107.91
SURIGAO DEL
NORTE
SURIGAO CITY (Capital)
403.44
2.92
2.04
8.95
SURIGAO DEL
NORTE
TAGANA-AN
174.90
*
*
6.35
SURIGAO DEL
NORTE
TUBOD
98.59
*
*
*
SURIGAO DEL SUR
BAROBO
2,268.65
*
*
11.29
SURIGAO DEL SUR
BAYABAS
149.36
*
*
*
SURIGAO DEL SUR
CITY OF BISLIG
3,113.48
*
*
114.77
SURIGAO DEL SUR
CAGWAIT
120.23
*
*
*
60
Province
City or Municipality
Falcata
Bagras
Gmelina
Mangiu
m
SURIGAO DEL SUR
CANTILAN
98.74
*
*
*
SURIGAO DEL SUR
CARMEN
372.96
*
*
9.20
SURIGAO DEL SUR
CARRASCAL
99.51
*
*
16.90
SURIGAO DEL SUR
CORTES
80.82
*
*
*
SURIGAO DEL SUR
HINATUAN
1,306.70
*
0.13
0.47
SURIGAO DEL SUR
LANUZA
373.78
*
*
*
SURIGAO DEL SUR
LIANGA
2,283.88
*
*
6.13
SURIGAO DEL SUR
LINGIG
1,867.66
*
*
158.71
SURIGAO DEL SUR
MADRID
191.08
*
*
*
SURIGAO DEL SUR
MARIHATAG
633.67
*
*
*
SURIGAO DEL SUR
SAN AGUSTIN
366.03
*
*
*
SURIGAO DEL SUR
SAN MIGUEL
1,104.14
*
*
3.43
SURIGAO DEL SUR
TAGBINA
5,448.05
*
*
2.41
SURIGAO DEL SUR
TAGO
1,072.61
*
*
8.22
SURIGAO DEL SUR
CITY OF TANDAG (Capital)
357.30
*
*
*
DINAGAT ISLANDS
BASILISA (RIZAL)
6.13
*
*
5.07
DINAGAT ISLANDS
CAGDIANAO
6.03
*
*
*
DINAGAT ISLANDS
DINAGAT
3.39
*
*
*
DINAGAT ISLANDS
LIBJO (ALBOR)
2.05
*
*
9.51
DINAGAT ISLANDS
LORETO
0.42
*
*
*
DINAGAT ISLANDS
SAN JOSE (Capital)
0.74
*
*
1.26
DINAGAT ISLANDS
TUBAJON
1.56
*
*
1.00
*Refinement and validation were not conducted in these localities.
61
UAS Mapping
UAS image acquisition was carried out on 31 sites in various barangays of
Butuan City (Figure 37). These sites have 41 aggregated Falcata plantations at
different growth stages. The number of plantations per growth stage wherein the
images were acquired are tabulated in Table 11. The example outputs that were
generated from UAS images are shown in Figure 38, Figure 39, Figure 40.
Table 11. Number of Falcata plantations in Butuan City with acquired images per growth stage.
Growth Stage
Number of Plantations
1-year
5
2-years
5
3-years
5
4-years
5
5-years
5
6-years
5
7-years
4
8-years
4
10-years
3
Figure 37. Map of selected Falcata plantations with UAS images acquired.
62
Figure 38. Example UAS mapping outputs for a 2-year-old Falcata plantation in Butuan City.
63
Figure 39. Example UAS mapping outputs for a 5-year-old Falcata plantation in Butuan City.
64
Figure 40. Example UAS mapping outputs for a 7-year-old Falcata plantation in Butuan City.
65
Falcata Stand Age Estimation Using Sentinel-2 Image
Relationship between stand age and Sentinel-2 Bands and Vegetation
Indices
Table 12 presents the Pearson’s correlation coefficients of stand age versus
the Sentinel-2 bands and vegetation indices (Vis). The results indicated that the
Sentinel-2 visible bands are negatively correlated with stand age, while positive
correlations were found for Band 8 and VIs. Among the four bands, a moderately
strong correlation was found for Band 4 (Red), with the highest coefficient of -0.53; for
VIs, the highest correlation was found for ARVI and RVI (r = 0.50).
The negative correlations found for the visible bands may be explained by the
differences in canopy structures as a Falcata plantation becomes older. Like rubber
trees, Falcata grow fast (Krisnawati et al., 2011), such that young canopy stands will
have fewer gaps resulting in high reflectance (Chen et al., 2012). However, these gaps
increase as the trees continue to grow; compounded by its alternate, bipinnately
compound leaves, this would result to lower reflectance. Another plausible explanation
is the process of regular thinning. This is a type of Falcata plantation maintenance that
removes diseased or pest-infested trees, deformed or poorly shaped trees, and
suppressed trees within a plantation starting 2 years after planting and then every year
up to 10 years (Krisnawati et al., 2011). Falcata stands that are 4-5 years old can be
thinned to a density of 250 trees per hectare (Krisnawati et al., 2011).
Estimated Stand Age of Falcata Plantations Using Sentinel-2 Image
Table 13 summarizes the stand age estimation models derived through
exponential regression of observed stand age and Sentinel-2 band reflectance and
VIs. Based on the R2 values, majority of the models do not show good fitness, with R2
ranging from 0.11 (Band 8) to 0.54 (Band 4). On the other hand, the RMSE values are
relatively low, with values ranging from 1.08 to 1.56 years.
From these results, the exponential model developed using Band 4 is the best
among the 11 models. Using only Band 4, this model (Figure 41) could provide more
accuracy than the other models. and yielded the highest R2 of 0.53 and RMSE of 1.08
years. However, the model has a limitation in predicting age. When the model was
used to predict stand age using the validation data, it fails to correctly estimate ages
of Falcata plantations that are more than 5 years, indicating underestimation for this
age group. This can be seen from the scatter plot of observed age versus predicted
age for validation data (Figure 42), and further confirmed by the plot of observed age
versus the residuals of estimation (Figure 43).
An example Falcata stand age map using the Band 4 stand age estimation
model is shown in Figure 44.
A Falcata tree count map is also included (Figure 45) wherein the number of
trees is calculated based on the area Falcata plantation and using a density of 250
trees per hectare as per Krisnawati et al., 2011.
66
Table 12. Pearson’s correlation coefficient of Falcata stand age versus Sentinel-2 bands and vegetation
indices (n=81).
Band 2 (Blue)
-0.52
ARVI
0.50
Band 3 (Green)
-0.49
GNVDI
0.47
Band 4 (Red)
-0.53
RVI
0.50
Band 8 (NIR)
0.24
DVI
0.32
NDVI
0.49
PVI
0.32
SAVI
0.39
Table 13. Summary of age estimation models derived to exponential regression. ME
(Mean Error) and RMSE (in years) were computed when the models were applied to
the validation dataset.
Variables
Stand Age Estimation
Model
R2
ME
RMSE
Band 2
33.259e-74.32x
0.48
0.07
1.09
Band 3
73.944e-56.25x
0.44
-0.01
1.23
Band 4
12.013e-38.13x
0.53
0.03
1.08
Band 8
0.8539e4.1233x
0.11
0.01
1.23
ARVI
0.0836e4.518x
0.49
0.03
1.10
DVI
0.7453e4.7893x
0.19
-0.10
1.45
GNVDI
0.0089e8.0267x
0.41
0.10
1.17
NDVI
0.0232e6.0068x
0.46
-0.16
1.40
PVI
0.7453e6.773x
0.19
-0.17
1.56
RVI
1.0696e0.0876x
0.40
0.15
1.26
SAVI
0.2504e4.7789x
0.29
-0.13
1.53
Figure 41. Exponential regression plot for stand age estimation model using Sentinel-2
Band 4.
y = 12.013e-38.13x
R² = 0.3235
0
1
2
3
4
5
6
7
8
9
10
0.000 0.020 0.040 0.060 0.080
Stand Age (yr)
Band 4 Reflectance
Model Using Sentinel-2 Band 4
67
Figure 42. Scatter plot of observed age versus predicted age for validation data.
Figure 43. Observed age versus the residuals of estimation (predicted ageobserved age) for validation
data.
0
1
2
3
4
5
6
7
8
9
10
0 1 2 3 4 5 6 7 8 9 10
Predicted Age (yr)
Observed Age (yr)
Model Using Band 4
-6
-4
-2
0
2
4
6
012345678910
Difference between Predicted
and Observed Age (yr)
Observed Age (yr)
Model Using Band 4
68
Figure 44. Example Falcata Plantation age map derived using the Sentinel-2 image-derived stand age
estimation model.
Figure 45. Example Falcata tree count map, assuming a density of 250 trees per hectare as per Krisnawati
et al., 2011.
69
Concluding Remarks on Falcata Stand Age Estimation Using Sentinel-2
Image
In this case study, we investigated the use of a single-date relatively-cloud free
Sentinel-2 image for estimating the age of Falcata plantations by making use of
reflectance characteristics and vegetation indices (Vis) values of these plantations that
can be extracted from the image. Pearson’s correlation analysis revealed that the
Sentinel-2 visible bands are negatively correlated with stand age, while positive
correlations were found for Band 8 and VIs. Among the four bands, a moderately
strong correlation was found for Band 4 (Red), the highest among the 11 variables
investigated.
Eleven (11) exponential regression models were also developed for stand age
estimation. The model using Band 4 was found to be the best by yielding the highest
R2 and lowest RMSE among the models. However, the model’s relatively good
prediction accuracy is limited to Falcata plantations that are five years and younger;
the model underestimates stand age for Falcata plantations that are more than 5 years
old. The model’s prediction accuracy maybe improved using a greater number of
samples per age. Use of multiple regression analysis may also develop more accurate
models than the ones derived in this study.
As the study was limited to 4 bands, future studies should also focus on the
investigation of the relationship of Falcata stands are to the 9 bands that are available
for the 20-m resolution Sentinel-2 dataset.
70
Falcata Stand Age Estimation Using Landsat 8 OLI
Image
Multivariate Regression Models for Age Estimation
Table 14 presents the multivariate regression models that utilized combinations of
Landsat 8 OLI image bands and vegetation indices.
Table 14. Summary of multivariate regression models developed for a regression-based Falcata stand
age estimation model using Landsat 8 OLI image.
ID
Independent Variables
N
R2
Ra2
RMSE
(years)
1
B3, B4, B5, B7
4
0.30
0.28
2.74
2
B3, B4, B7, NDVI
4
0.30
0.28
2.76
3
B4, B7, NDVI, Elev, Slope
5
0.33
0.31
2.82
4
B3, B4, B7
3
0.24
0.23
2.22
5
B3, B4, B5, B7, NDVI
5
0.24
0.24
2.25
6
B3, B4, B7, NDVI, RVI, Elev,
Slope
7
0.28
0.27
2.21
7
B3, B4, B5, B7
4
0.30
0.28
2.60
8
B3, B4, B6, B7, NDVI, GNDVI
6
0.32
0.30
5.07
9
B3, B4, B5, B7, NDVI, GNDVI,
Elev
7
0.38
0.35
5.50
The multivariate regression models developed indicated that the inclusion of the
derived vegetation indices bands did not improve the fitness of the regression models.
The combination of the spectral bands of the Landsat-8 OLI imagery and the four
derived vegetation indices bands could not explain more variability in the Falcata stand
ages than the spectral bands. On the other hand, incorporating the elevation and slope
as additional bands to the spectral and vegetation indices bands improved the
prediction accuracy of the multivariate regression models. Among the nine models,
model 9 explained the most variability in Falcata stand ages (R2=0.38).
Accuracy of the Age Estimation Models
Using the equations of the models presented in Table 14, the performance of
the models was evaluated using the validation data. Based on the scatter plots
presented in Figure 46, all the multivariate regression models have tended to give
inaccurate results due to underestimation and overestimation of the actual stand ages
of Falcata. The estimated age for the old Falcata stands years 8 and 10 show
underestimation while results for the young Falcata stands years 1 and 2 show
overestimation.
This analysis was jointly conducted with John Carl Escasio and Rachelle Soliman, BS Geodetic
Engineering Students, who served as on-the-job trainees.
71
Figure 46. Scatter plots of the actual Falcata stand ages versus predicted age by the Landsat 8 OLI-
based age estimation models (n=125).
Among the nine models developed in this study, model 6 was used for the generation
of a map that shows the stand ages of Falcata plantations as estimated by a regression
model, having achieved the lowest RMSE value of 2.21 years. The equation of model
6 was applied in the Band Math functionality of Envi 5.1 to estimate the stand ages of
Falcata plantations in Butuan City, a city within Caraga where shapefiles of Falcata
plantations are available. Figure 47 presents the map showing the stand ages of
Falcata plantations as estimated by the multivariate regression model 6.
72
Figure 47. Map showing the stand ages of Falcata Plantation as estimated by Model 6.
Concluding Remarks on Falcata Age Estimation Using Landsat 8 OLI
In this study, multivariate regression models utilizing the spectral bands,
vegetation indices bands, and the elevation and slope as additional bands were
developed to estimate the stand ages of Falcata plantations. This study finds the
lowest RMSE value obtained at 2.21 years by model 6 among the nine models
developed. On the other hand, the fitness of the multivariate regression models to
explain the variability in the Falcata stand ages only ranged from 0.24 to 0.38.
73
Furthermore, the accuracy assessment of the nine multivariate regression models
shows their tendency to give inaccurate results due to overestimation and
underestimation of the actual stand ages of Falcata plantations. For the accurate
determination of the stand ages of Falcata plantations, these tendencies of the
regression approach must be considered into decisions.
74
Field Mapping of WPPs and Furniture Makers
A total of 48 WPPs were located on the field by the project (Table 15). Of these,
40 were included in the list of DTI-registered WPPs for the year 2019. The remaining
8 WPPs registration status were unknown at the time of field surveys as the project
only made use of the 2019 DTI list. On the other hand, only 24 furniture makers were
geolocated/validated by the project (Table 16).
Figure 48 shows some of the pictures taken during the WPP and furniture
maker field mapping. Figure 49 shows the map of WPPs and furniture makers that
have been mapped/geolocated in the field. Figure 50 provide an overview of the annual
capacities of the mapped WPPs.
From the field data and the maps, it can be noticed that the WPPs are not well
distributed in the region. The distribution of these mapped WPPs are as follows:
Agusan del Norte: 26
Agusan del Sur: 18
Surigao del Sur: 4
They are mainly concentrated in Agusan del Norte, particularly in Butuan City. The
project was not able to map/locate WPPs in Surigao del Norte. WPPs with larger
annual capacities (greater than 46,000 cubic meters) are located in Agusan del Norte,
and only a few are in Agusan del Sur.
Figure 48. Example of WPPs and furniture makers visited in Butuan City.
75
Figure 49. WPP and furniture maker locations in Caraga Region based on field mapping conducted by
the project. Also shown are the mapped Falcata plantations.
76
Table 15. List of mapped WPPs in Caraga Region. Annual log requirement in asterisk (*) means they were unknown during the time of data of gathering.
No
.
Name
Province
Municipality
Baranga
y
Type
Detailed
Daily Rated
Capacity
(DRC), in
cubic meters
DRC, in
cubic
meters
Annual
Capacity
, in
cubic
meters
Annual Log
Requiremen
t, in cubic
meters
Latitude
Longitude
Date of
Field
Mapping
1
A&V VENEER
MANUFACTURIN
G & MINI-
SAWMILL
AGUSAN
DEL
NORTE
BUTUAN
CITY
San
Vicente
Veneering with
Mini-sawmill
Component
5.89 Mini-
sawmill; 7.5
Veneering
13.390
4887.350
6856.00
8.900111
125.554000
25/03/202
1
2
A&V VENEER
MANUFACTURIN
G & MINI-
SAWMILL
AGUSAN
DEL
NORTE
BUTUAN
CITY
San
Vicente
Integrated
13.39
13.390
4887.350
535.60
8.900111
125.554000
25/03/202
1
3
AGUSAN
PLYWOOD
CORPORATION
AGUSAN
DEL
NORTE
BUENAVISTA
Manapa
Integrated
64.08 cu.m
Plywood;
58.96 cu.m
Blockboard
123.040
44909.60
0
68290.90
8.980239
125.448000
26/03/202
1
4
AMPARO MINI
SAWMILL
AGUSAN
DEL
NORTE
BUTUAN
CITY
Amparo
Integrated
28.26
28.260
10314.90
0
16008.00
8.866450
125.556000
29/03/202
1
5
BINGKILAN
WOODCRAFT
AGUSAN
DEL
NORTE
BUENAVISTA
Malpoc
Mini-sawmill
20.00
20.000
7300.000
8000.00
8.942850
125.400000
26/03/202
1
6
BUTUAN
ESPERANZA
VEENER
CORPORATION
AGUSAN
DEL
NORTE
BUTUAN
CITY
Rajah
Soliman
Pob.
(Bgy. 4)
Integrated
26.77
26.770
9771.050
18676.00
8.945425
125.543458
29/03/202
1
7
C. JAPITANA
WOODCRAFT
AGUSAN
DEL
NORTE
BUTUAN
CITY
Pangabu
gan
Mini-sawmill
4.72
4.720
1722.800
1888.00
8.931789
125.549000
29/03/202
1
8
CARLOABEL
MINI-SAWMILL
AGUSAN
DEL
NORTE
BUTUAN
CITY
San
Vicente
Mini-sawmill
11.79
11.790
4303.350
4716.00
8.919019
125.553000