Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji INTERNATIONAL CLIMATE INITIATIVE

Transcription

1 INTERNATIONAL CLIMATE INITIATIVE Regional project Climate Protection through Forest Conservation in Pacific Island Countries Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji

2

3 Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji March 2014

4 Prepared by: Johannes Eberenz Johannes Reiche On behalf of: SPC/GIZ Regional Project Climate Protection through Forest Conservation in Pacific Island Countries P.O. Box 14041, SUVA, Fiji 2 SPC/GIZ Regional REDD+ Project

5 Summary Data Ownership Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 3

6 Contents 4 SPC/GIZ Regional REDD+ Project

7 List of Figures List of Tables Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 5

8 Acronyms & Abbreviations 6 SPC/GIZ Regional REDD+ Project

9 1 Introduction 1.1 Context and Background 1.2 Problem Definition Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 7

10 1.3 Study Objectives and Questions 2 Study Site and Material 2.1 Study Site Fiji 8 SPC/GIZ Regional REDD+ Project

11 km ± Lololo study site Plantation Leases Fiji Pine Ltd Fiji Hardwood Co. Ltd. Kilometers ± Bacground mage: DMC 2011 Projection: UTM 60S Datum: WGS84 Figure 1: False color DMC mosaic of Viti Levu with plantation lease areas and Lololo study site extent Lololo Study Site REDD Program Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 9

12 2.2 Forest Cover Change Data 2.3 Remote Sensed Imagery Landsat Data Figure 2: Valid observations per non-water pixel for Landsat WRS Path 75 Row 72, SPC/GIZ Regional REDD+ Project

13 Table 1: Landsat imagery: number of scene per sensors Path/Row Sensor No. Of Scenes Dates 075 / 72 TM ETM+ SLC-on ETM+ SLC-on High and Very High Resolution Imagery 2.4 Remote Sensing Based Change Products 2.5 Additional Data 3 Methods 3.1 Forest Cover Change Data Inventory and Quality Control Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 11

14 3.2 Database Design 3.3 Data Integration and Harmonization 12 SPC/GIZ Regional REDD+ Project

15 3.4 Validation Concept 4 Results 4.1 Forest Change Data Inventory and Quality Control Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 13

16 Table 2: FCC datasets inventory overview. Orange text indicates shortcomings of the datasets for the purpose of FCC validation. Until now, no change data derived from digitized VHR imagery exists. Dataset: Temporal resolution (sampling interval): Temporal coverage: Temporal info. quality Spatial resolution / sampling density: Spatial coverage: Permanent sampling plots Indigenous forest harvesting GIS data Pine / Hardwood Plantation GIS Data (Potential Digitized VHR imagery) Source: FFD MSD FFD MSD FPL, FHCL Presently none Spatial information quality: Thematically quality: Sampling design: Probability sampling Medium (2 years) , continued Fine (1 year) , continued Very fine (1 year or sub year) Ca , continued Coarse (approx. 5 years) Not yet available (2013?) High High High High Low (100 plots for main islands) Sample of main islands Location accuracy medium (GPS errors) High Systematic (regular grids) Fiji Pine Limited data a) Landslides: High (GPS data) Harvested areas of indigenous forest of main islands, but incomplete Unknown, heterogeneous plots, GPS errors Unknown Based on management activity High (GPS data) Plantation area Medium (offsets, heterogeneous plots) Medium (some years missing) Based on management activity Very high (pixels < 10 x 10 m) Potentially complete main islands, some cloud cover Very high High, (depending on digitizing method) Complete survey (apart from clouds) Yes No (No) Yes (Chaudhary 2012) b) Logged Areas 14 SPC/GIZ Regional REDD+ Project

17 Table 3: Results of visual assessment of temporal information in FPL harvest data Type of temporal information in dataset Valid entries (plots) No. of plots where dataset period overlap with logging period derived from visual interpretation Logging Quarter of year (93.7%) Logging Year (99.5%) ± Digitized VHR FPL logged area Background image: LS ETM Scene Row/Path 75/72 Aquired Projection: UTM 60S Datum: WGS Figure 3: FPL logging data vs. Landsat imagery c) Planting Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 15

18 4.1.2 Fiji Hardwood Ltd.: Logging Data Indigenous Forest Harvesting Data National Forest Inventory Plots Permant Sampling Plots 16 SPC/GIZ Regional REDD+ Project

19 Figure 4: FFD permanent sampling plots grid Digitized VHR 4.2 Database design User Requirements Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 17

20 o o o o Conceptual Database Model a) Definition of Database Themes: Forest Cover Change Types Table 4: FCC theme working definitions, with corresponding FFD definitions and datasets that contain them. FCC theme Full harvest of plantation Selective harvest of plantation Selective harvest of indigenous forest Replanting of plantation Working definition Areas marked as logged in FPL data Areas marked as logged in FHCL data Areas marked ads harvested in MSD harvesting data, PSP plots with change in forest cover class inside harvest plots Areas marked as replanted in FHCL and FPL data Corresponding FFD definition (based on FAO 2004) Potentially degradation (if logging is not sustainable) Forest improvement Datasets FPL harvesting data FHCL harvesting data FFD harvesting data, PSP FPL and FHCL replanting data Uncontrolled Deforestation of indigenous forest Uncontrolled degradation of indigenous forest Landslides affected plantation areas (No-change) b) Representation of time and change Change of land cover from any forest class to non-forest in PSP data outside harvesting areas Change of land cover from closed to open forest in PSP data outside harvesting areas Areas marked as landslide affected in FPL data Without activities in plantation data, PSP with unchanged forest cover class Deforestation Degradation Degradation, potentially deforestation Forest or Non-Forest land PSP PSP FPL Landslide data, PSP All datasets 18 SPC/GIZ Regional REDD+ Project

21 Time Baseline temporal extent X 2005 Baseline spatial extent Change feature with change year Y 2005 No-change Figure 5: Concept of baseline and change features. 3 Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 19

22 Baseline Feature ID Time Extent Data Source Geometry 1 N Change Event ID Location ID Baseline Feature ID Type Time Data Source N 1 Change Location ID Geometry Figure 6: E/R model of conceptual FCC database design 4.3 Data Harmonization and Integration GIS Forest change features VHR imagery derived change data PSP and Other Survey Plots 20 SPC/GIZ Regional REDD+ Project

23 4.4 Validation Concept Concept of Accuracy Assessment Accuracy Measures Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 21

24 Table 5: Proposed accuracy measures with interpretation and calculation. Measure Interpretation Calculation Confusion Matrix Sums up correctly classified and Matrix of counts of samples mapped as confused samples per class. class i with reference class j. Note that for stratified samples area corrected consistent estimators have to be used 4. Overall Accuracy Ratio of correct detected samples ith m the no. of classes (OA) Confidence interval (CI) for Overall Accuracy Producer s accuracy (per class) User s accuracy (per class) Sampling uncertainty for OA. Not applicable for complete survey sampling. Share of the of correctly classified samples in a reference class Share of the of correctly classified samples in a mapped class Calculation depends on sampling design. For a sufficiently large simple random sample with sample size n, the CI at confidence level α can be approximated as ( ) Card (1982) provides variance estimators for stratified random sampling. ; with being the reference class total sample count ; with being the mapped class total sample count Mean time lag of change detection How average time between RS-based detection of change and reference data time of change. Not for bi-temporal change detection. Mean of difference between detected and reference data time of change of all sample with change in both reference and TRS-based product. 4.5 Demo Implementation Selection of Study Site and Data: 22 SPC/GIZ Regional REDD+ Project

25 4.5.2 Demo Database a) Software and Hardware Choice b) Data standards and metadata c) Database Schema Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 23

26 Table 6: Schema of change feature table. The key attributes (FID, foreign key: Baseline) are underlined. Attribute Data format Domain/Format/Unit Content Required FID Integer 1 to number of change Unique feature identification number, key features attribute OrgFID Integer Feature Identification number in original dataset Other Identification attributes from in OrgID String Various original dataset, e.g. Stand/Coupe ID for Fiji - Pine data Type String See Table 4 Forest cover change type or no-change Baseline Integer Baseline feature FID Foreign key to match with baseline feature Year Integer YYYY Year of change - Period String Sub year period - Start Date YYYY-MM-DD Starting date of change/no-change period End Date YYYY-MM-DD End date of change/no-change period VolumeRm Float m 3 Volume removed, where applicable - AreaDiff Float ha Area harvested or replanted, where applicable - DataSrc String FFD, FHCL, FPL, or SOPAC Data Source (agency name) Location String island, province, mataqali where available - Geometry Multipart Projection: WGS 84 UTM Polygon 60 South The spatial information Table 7: Schema of baseline feature table. The key attribute (FID) is underlined. Attribute Data format Domain/Format/Unit Content Required FID Integer 1 to umber of baseline Unique feature identification number, key features attribute OrgFID Integer Feature Identification number in original OrgID String Sample Type String Various Complete-survey random, stratified, systematic dataset Other Identification attributes from in original dataset, e.g. Forest attribute for Fiji Pine Sampling type Start Date YYYY-MM-DD Starting date of sample period End Date YYYY-MM-DD End date of sampling period DataSrc String FFD, FHCL, FPL, or SOPAC Data Source (agency name) Geometry Multipart Polygon Projection: WGS 84 UTM 60 South The spatial information - d) Interfaces 24 SPC/GIZ Regional REDD+ Project

27 4.5.3 Data Integration and Harmonization Demo Validation Table 8: Overall accuracy per case Case a,i): yearly change, unfiltered reference data 0,7828 b,i) bi-temporal change, unfiltered reference data 0,8318 a,ii) yearly change, filtered reference data 0,7938 b,ii) bi-temporal change, quality filtered reference data 0,8398 Overall accuracy Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 25

28 Figure 7: Demo validation RS-detected change vs. reference data (detail) Figure 8: Results of the demo validation (detail). 26 SPC/GIZ Regional REDD+ Project

29 Table 9: Confusion matrix (pixel count) for yearly change and unfiltered reference data (case a,i) Reference Class Mapped Class no-change Row sums no-change Col. sums Table 10: : Produce s and User s accuracy for yearly change and unfiltered reference data (case a,i) Producer's Accuracy User's Accuracy no-change Table 11: Confusion matrix (pixel count) and per-class accuracy measures for case b,i): bi-temporal change with unfiltered reference data Reference Class Producer's Accuracy User's Accuracy Mapped Class no-change change Row sums no-change no-change change change Col. sums Discussion 5.1 Reference Data Availability, Suitability and Limitations Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 27

30 5.2 Data Harmonizing and Integration 5.3 Database Design 5.4 Demo Implementation 28 SPC/GIZ Regional REDD+ Project

31 6 Recommendations 6.1 Clear Definitions (2011) 6.2 Additional Permanent Sampling Plots 6.3 Further Investment in Very High Resolution Imagery 6.4 Central and Transparent Data management Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 29

32 6.5 Capacity Building 30 SPC/GIZ Regional REDD+ Project

33 Literature Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 31

34 32 SPC/GIZ Regional REDD+ Project

35 Reference Database and Validation Concept for Remote Sensing based Forest Cover Change Products in Fiji 33

36