“The cover, use, and land classification conundrum in the Philippines – how remote sensing may contribute towards a reasonable classification system” Enrico C. Paringit, Dr. Eng. Department of Geodetic Engineering University of the Philippines Diliman

Delivered during the Land Cover/ Changes (LC/LUC) and Impacts on Environment in South/Southeast Asia - International Regional Science Meeting, 28-30th May, 2018, Philippines Outline of this Presentation

• Forest change in the Philippines – Forest cover vs. Forest use vs. Forest Tenure • The Phil-LIDAR FREXLS – Canopy Cover Estimation using Airborne LIDAR – Average DBH Estimation and Biomass – Forest Cover Classification – Assessing the NGP sites – Bamboo Identification • From tree cover to forest cover • Concluding Remarks Definitions • - Either natural or plantations of forest crops such as trees, or both, occupying a definable, uninterrupted or contiguous area not less than one hectare in size with the tree crowns covering at least ten percent (10%) of the area (DENR DAO 96-29) • FOREST -- Lands of the public domain which have been classified as such under the land classification program of the DENR and all unclassified lands of the public domain.

DENR – Department of Environment and Natural Resources Forest cover

FOREST • Land spanning more than 0.5 hectares with trees higher than 5 meters and a canopy cover of more than 10% Approximate 68,400 sq.km. Area: Boracay as a landmark issue on Background land classification • The Philippines total land area is ~290,000km2 • Under the Regalian doctrine, all lands of the public domain belong to the State (1987 Philippine Constitution, Art XII, Sec 3) • Under the Public Land Act (Act 2874, ca. 1919), Lands of the public domain are classified into agricultural (A&D), forest or timber, mineral lands and national parks – Forest lands include the public forest, permanent forest or forest reserves, and forest reservations, which the government seeks to protect, develop, and rehabilitate so as to ensure their continuity in productive condition. • Presidential Decree 705 – >18% slope not(A&D) • SC (GR 173775 & 167707): Classification of lands of the public domain is descriptive of its legal nature or status, and does not have to conform to what the land actually looks like

Aspects of Forest Change Forest Cover (Physical)

Forest Change Forest Forest Land Land Use Classification (Actual) /Tenure (Legal)

Tenure in Forest Areas • Lands classified as public forest may be granted tenure for: – Timber License Agreements (TLAs) – no longer issued – Industrial Forest Management Agreements (IFMA) – Pasture Lease Agreements – Community Based Forest Management Agreement (CBFMA) - 25-year term – Certificate of Stewardship Contract (CSC) - 25-year term – Certificate of Ancestral Domain Claim-Community Based Forest Management Agreement (CADC-CBFMA) and Certificate of Ancestral Land Claim-Community Based Forest Management Agreement (CALC-CBFMA) The Philippine Land Classification System

Land Classification

Unclassified Classified Land Land

Alienable and Other Classifications Forest Land (mineral Land, Disposable National Parks ,etc.) Land Classification Map (as of 1989) National forest surveys in the Philippines using remotely sensed data

Year Source % Forest Cover Source of Data Method of Interpretation 1973 Lachowski et al. (1979) 38.0 Landsat Digital

1974 Bruce (1977) 29.8 Landsat Visual

1976 Bonita & Revilla (1977) 30.0 Landsat Visual

1980 Forestry Development Center 25.9 Landsat Visual (1985) 1987 Swedish Space Corporation 23.7 SPOT Visual (1988) 2003 National Mapping and Resource No data Landsat Visual Information Authority (2004) 2010 National Mapping and Resource No data Landsat, AVNIR Visual Information Authority (2014)

(extended by De Alban (2016) from Kummer et al, 1992) How are forest uses determined and governed? • Protected – NIPAS Act • Forest instruments – Timber License Agreement (TLA) – Community-Based Forest Management Land Use Distribution the Philippines Recent Nationwide Land Cover Maps

Recent national land cover maps of the Philippines: (a) 1987 SSC; (b) 2003 NAMRIA; (c) 2010 NAMRIA Forest Cover Classification

1987 map (SSC, 1987) 2003/2010 NAMRIA Land Cover Closed canopy, mature > 50% Closed forest, broadleaved Open canopy, mature <50% Closed forest, coniferous Mossy forest Closed forest, mixed Pine forest Open forest, broadleaved Mangrove forest Open forest, coniferous Marshy area and swamp Open forest, mixed Submarginal forest Forest plantation, broadleaved Forest plantation, coniferous Forest plantation, mangrove (2003 only) Mangrove forest NAMRIA Land cover classification scheme (as of 2003) Classification Code 1 Closed forest, broadleaved NF4F 2 Closed forest, mixed NF4M 3 Closed forest, coniferous NF4C 4 Open forest, broadleaved NF2B 5 Open forest, mixed NF2M 6 Open forest, coniferous NF2C 7 Mangrove forest NFM 8 Forest plantation, broadleaved FPB 9 Forest plantation, coniferous FPC 10 Other wooded land, shrubs Sh 11 Other wooded land, fallow Fa 12 Other wooded land, wooded WGL 13 Other land, natural, barren land BL 14 Other land, natural, grassland GL 15 Other land, natural, marshland ML 16 Other land, cultivated, annual crop AC 17 Other land, cultivated, perennial crop PC 18 Other land, cultivated, pastures Pa 19 Other land, fishpond Fs 20 Other land, built-up area BUA 21 Inland IW Forest Cover Change

• Forest Cover Change estimated from time series ALOS/PALSAR data 2010 to 2015 of the REDD+ Replication Site in Davao Oriental (From GIZ (Estomata), et al. 2016) THE NATIONWIDE DISASTER RISK AND EXPOSURE ASSESSMENT FOR MITIGATION (DREAM) LIDAR PROGRAM (2012 TO 2016)

PHIL-LIDAR 1 PROGRAM The National LIDAR Mapping Program of the Philippines • Data Acquisition Phil- • Data Validation • Data Processing DREAM LIDAR1 • Training & IEC • Data Archiving

• Agriculture • Forest Phil- • Coastal LIDAR2 • Energy • Hydrology Airborne LiDAR Coverage • For nearly five years of operation, covered over 150,000km2 of LIDAR at 2pts/m2 • Started 22 November 2012 with inaugural flight • All flights conducted until 2017 • Covered mostly flood plains • Covered almost principal and major river basins – 18 principal river basins – 262 major river LIDAR Returns FOREST AREAS WITH LIDAR ± Introduction

Philippine Forest (NAMRIA 102,574.49 Land Cover 2010 Forest sq.km. Cover + Other wooded land) Forest with LiDAR (May 17,210.87 2017) sq.km.

Percentage of LiDAR 17.18% Coverage in Forest Areas

0 0.250.5 1 1.5 2 Kilometers ± Calibration Sites • 2 hectare plots per site

1. Natural Mangroves (NMF): San Juan, Batangas 2. Mangrove Plantation (MPF): Masinloc, Zambales 3. Nipa Plantation (NPF): Lingayen, Pangasinan 4. Mahogany Plantation (MahPF): Gerona, Tarlac 5. Gmelina Plantation (GmePF): Maramag, Bukidnon 6. Rubber Plantation (RubPF): Maramag, Bukidnon 7. Open Broadleaf (OBF): Los Baños, Laguna 8. Closed Broadleaf (CBF): Gattaran, Cagayan 9. Coniferous Plantation (CPF): Malitbog, Bukidnon 10. Natural Coniferous (NCF): Baguio City, Benguet Puerto Princesa

0 0.250.5 1 1.5 2 Kilometers LIDAR Processing Techniques Developed

1. Canopy Height Model, Canopy Cover Model and Digital Terrain Model from LiDAR Point Cloud 2. Forest Classification Using LiDAR Point Cloud Distribution 3. Detection of Tree Top Using OBIA 4. Use of Digital Hemispheric Photography in Creating Forest Cover Canopy Height Model (CHM) Canopy Cover Map (CCM) Digital Terrain Model (DTM) Forest cover classification using LIDAR data Involves the collection of LiDAR “signatures” • 50 training areas were selected from each calibration sites NMF MPF NPF  5m x 5m pixel size

band statistical values for each • MahPF training areas were used for the GPF RPF Support Vector Machine (SVM) classification

OBF CBF  Right image: selected ROIs displayed in LiDAR composite RGB image (average height, cover, density)

CPF NCF From Faelga et al. (2017) Process Workflow From Faelga et al. (2017)

Preliminary classification (DTM and CHM bands) Average height Parameter selection Cover Density • Out of the 27 LiDAR parameters, Maximum 16 were selected 10th non- cumulative bincentage (NC10)  parameters that are not absolute values NC20

• Inclusion of DTM & CHM NC30  DTM: contextualization of forest types NC40

 CHM: removal of undergrowth and non- NC50

tree species NC60 NC • Total number of parameters: 18 70 NC • Parameters were treated as bands to 80 NC create a multi-band raster image 90 NC99 (composite) Quadratic Average Standard Deviation Digital Terrain Model From Faelga et al. (2017) Canopy Height Model Results (Calibration) Overall Accuracy: 83.08% Kappa Coefficient: 0.81

NMF MPF NPF

From Faelga et al. (2017) MahPF GPF RPF Ave. point cloud Forest Type Test Site Sensor density per sqm (all returns) Natural Mangroves (NMF) Batangas Pegasus 2.05 Mangrove Plantation (MPF) Zambales Pegasus 3.05 Nipa Plantation (NPF) Pangasinan Pegasus 3.35 OBF CBF Mahogany Plantation (MahPF) Tarlac Gemini/ Pegasus 2.1011.23 Gmelina Plantation (GPF) Bukidnon Pegasus 21.911.23 Rubber Plantation (RPF) Bukidnon Pegasus 11.44 Open Broadleaf (OBF) Los Banos Pegasus 1.31 Closed Broadleaf (CBF) Cagayan Pegasus 3.04 Coniferous Plantation (CPF) Bukidnon Pegasus 2.17 CPF NCF Natural Coniferous (NCF) Baguio Leica 11.75 Results (Calibration & Validation) CALIBRATION (1ST HA PLOTS) VALIDATION (2ND HA PLOTS/ ADJACENT SITES) Overall Accuracy: 83.08% Overall Accuracy: 74.58% Kappa Coefficient: 0.81 Kappa Coefficient: 0.72 From Faelga et al. (2017)

NMF MPF NPF NMF MPFMPF NPF

MahPF GPF RPF MahPF GPF RPF

Natural Mangroves (NMF) OBF CBF Mangrove Plantation (MPF) OBF CBF Nipa Plantation (NPF) Mahogany Plantation (MahPF) Gmelina Plantation (GPF) Rubber Plantation (RPF) Open Broadleaf (OBF) Closed Broadleaf (CBF) CPF NCF Coniferous Plantation (CPF) CPF NCF Natural Coniferous (NCF) ASSESSING THE NATIONAL GREENING PROGRAM (NGP): RESULTS OF THE CALIBRATION AND VALIDATION

Brgy. San Pascual, Ubay, Bohol November 7 to 16, 2016 UP Diliman and USC FRExLS Team From Phil-LIDAR 1 FREXLS: A DOST- funded Project Ground Invento vs. LIDAR-detected TREES • Total Trees: 2292 DBH (cm) Height (m) MIN 0.10 0.20 MAX 21.37 11.00 MEAN 3.54 3.05 STDV 2.81 1.70 • Trees with height 1 m and up: 2193

DBH (cm) Height (m)

MIN 0.16 1.00

MAX 21.37 11.00

MEAN 3.66 3.16

STDV 2.80 1.65 ANALYSIS

Actual vs Estimate

Per 20x20m, 10x10m, 5x5m subplots

with 1m Threshold Eliminate below Cogon level

Estimat Actual e 1236 1295 Comparison: Tree top count

Actual Estimate Diff % Diff All Inventories 2292 2154 -138 -6.21 (NGP extent) 39 Plots 1317 1387 70 5.18 >1m threshold 1236 1295 59 4.66

NOTE: If acceptable mortality rate of 5% for NGP sites per year is considered as error, then result is acceptable. FOREST PERCENT COVER 4 Question: How open is “OPEN” is open forest? 6 4 3

Using digital hemispherical photograph

1

Site: Gmelina 0.00% 30.61% 47.18% Plantation (Maramag,Bukidn DHP DHP 1 DHP on) 46 43 II. OVERVIEW OF PCA ANALYSIS RESULT

No. of CASE FOREST TYPE PCs R2 RMSE Samples 1 ALL 230 18 0.71 10.82 2 ALL Plantation 150 19 0.68 12.21 3 ALL Broadleaf Plantation 100 3 0.68 10.94 4 CBF + All BP 130 10 0.72 11.22

5 50 15 0.51 21.16 Coniferous Forest - Baguio 6 ALL + CF-Baguio 280 15 0.54 14.58 7 ALL + CF-Baguio+ Rubber 330 7 0.40 15.67 ACCURACY ASSESSMENT OF FOREST PERCENT COVER MAPPING

100 100 Model y = 0.6921x + 20.453 80 Score:FOREST R² = 0.7066 80 CASE n PCs R2 RMSE TYPE 60 60 23 1 18 0.71 10.82 ALL 0 40 40 Predicted Cover (%) Cover Derived Intercept and 20 20 Coefficients: 0 0 1 5 9 131721252933374145495357616569 0 20 40 60 80 100 PLOTS Field Cover(%) Field predicted

MAE: 8.10 Pearson Correlation: 0.84

40

20

0 20 40 60 80 100 -20 Residuals

-40

-60 IV. RESULTS OF FOREST COVER MAPPING USING DHP CANOPY COVER RESULTS (RASTERS)

Site: Baguio City NAMRIA: Open-Coniferous Forest CHM DHP (40- ADD (40- LESS (40- ADD (40- 100) 100) 100) >100)

295.8ha 266.24ha 285.12ha 311.48ha IV. DHP CANOPY COVER RESULTS (RASTERSSite: Baguio City) NAMRIA: Open-Coniferous Forest

CHM DH AD LE ADD > P D SS 100

68.36 77.53 58.66 78.69 % % % % V. Correlation Analysis for CCM and DHP(PCA)

Site : Baguio City (Coniferous CHM Forest)

CCM

DHP V. Correlation Analysis for CCM and DHP(PCA)

100 100 y =95 -0.0889x + 87.353 95 RMSE 9.12 R² = 0.0331 90 90 MAE 7.20 85 Pearson -0.182 85 80

correlation 80 Cover (%) 75 DHP cover (%) 70 75 65 70 60 60 70 80 90 1 5 9 13 17 21 25 29 33 37 41 45 49 CCM cover (%) PLOT

CCM DHP

20 15 10 5 0 65 75 85 95 Residuals -5 -10 -15 -20 Tree crown delineation using Airborne From Zaragosa et al. 2017 LIDAR Natural Coniferous Forest

Delineated 661 Tree Crowns

CHM Height Tree Count based (meters) from the intersected crown and actual trees OBIA GIS 0 18.8 18.44 1 19.41 18.84 2 19.65 19.01 3 19.49 19.34 4 17.47 20.25 5 17.45 13.28 BAMBOO STAND DETECTION USING AIRBORNE LIDAR - Bamboo - Detected From Maralit et al. 2017 validation site bamboo

MODEL ACCURA CY Calibration of GLCF Tree Maps for Philippine Conditions: Some Early Results and Examples

Global Tree Cover Data 2010 Global Tree Cover RMSE GLCF, Sexton (2009) Forest cover at 10% threshold

Forest Cover at 10% threshold Forest Cover Probability at 10% Forest cover: 10% vs. 30% threshold

Forest Cover at 10% threshold Forest Cover at 30% threshold Calibration of GLCF Tree Canopy Cover (TCC) maps using LIDAR-based Canopy Cover Maps (CCMs)

Uncalibrated TCC Correlation result after Calibration

LIDAR-derived CCM Calibrated GLCF TCC Map for Ph Uncalibrated vs. Calibrated Tree Cover

Forest Cover at 30% threshold Forest Cover at 30% threshold - Calibrated Image Classification Result: Landsat 8 (January 22, 2015) 2003 and 2010 NAMRIA Land Cover Summary of Land Cover Statistics for 2015, 2050, and 2100 Batiano, Baganga Parameters Distribution of Change 2015 RIDF Discharge Results

Batiano Bridge Total Precipitation Peak rainfall Peak outflow RRP Time to Peak (mm) (mm) (cms) 4 hours, 10 5-Year 275.6 31.7 576.8 minutes 25- 3 hours, 50 386.5 42.7 1123.2 Year minutes 100- 3 hours, 30 478 51.7 1634.5 Year minutes 2050 RIDF Discharge Results

Taytayan Bridge

RRP Total Precipitation (mm) Peak rainfall (mm) Peak outflow (cms) Time to Peak

5-Year 308.86 33.41 3747.6 4 hours, 50 minutes

25-Year 433.15 46.85 5026.5 4 hours, 40 minutes

100-Year 535.69 57.94 6083.6 4 hours, 40 minutes Some thoughts…

• “What is official?” vs. “What is accurate?” vs. “What is available?” • Gaps in understanding relationship between Forest LC, LU and LT open possibilities for further research and development • Current efforts to generate forest cover data products will be critical to the crafting of policies such as the National Land Use Act. Policy Interventions Conclusions

• Various land cover and forest classification systems being espoused by government agencies depending on the context of use. • Accurate information is needed in management and of forest • Techniques to assess and monitoring forest cover, forest uses and associated changes using remote sensing digital image analysis recognizing the differences between their definitions, requirements and applications. • Progress that has been made with the analysis and mapping of different forest structure using airborne LIDAR data from acquisition missions that have been conducted in recent years • These recent technological developments and historical perspectives play important roles in formulating key policies not only for generation of information on management, rehabilitation and conservation in the future. ACKNOWLEGEMENTS

• The results presented in this research is funded by the Department of Science and Technology (DOST) under the Nationwide Detailed Resource Assessment Using LiDAR (Phil-LiDAR 2), Project 3: Forest Resource Extraction from LiDAR Surveys (FRExLS). • Disaster Risk and Exposure Assessment for Mitigation (UP DREAM) and Phil-LiDAR 1 for the LiDAR data of the study site. • Project Climate Twin Phoenix • terraPulse Group for their cooperation