Remote sensing and mapping tool development of NFA Project in Vietnam
1 like932 views
The document summarizes remote sensing and mapping tool development by the National Forest Administration (NFA) project in Vietnam to support national forest inventories, statistics programs, and REDD+ monitoring requirements. Key activities included: 1) Developing land use and forest type classification methods using SPOT-5 and DMC satellite imagery to map resources on national and local levels. 2) Creating forest distribution maps and a centralized forestry database to provide statistical data. 3) Developing change detection tools to monitor land use and forest type changes over time to support reporting under NFI Cycle 5 and REDD+.
Remote sensing and mapping tool development of NFA Project in Vietnam
- 1. Remote Sensing and Mapping Tool Development by NFA Project in Vietnam
- 2. 2 NFI Cycle 2 1996-2000 NFI Cycle 1 1990-1995 NFI Cycle 3 2001-2005 NFI Cycle 4 2006-2010 NFI & Statistics 2011-2015 NFA Project 2011-2014 NFI Cycle 5 2016-2020 • Learns from past experiences in Vietnam and best practices from abroad • Develops methodology for NFI Cycle 5 to provide data on forest resources on national and provincial level • Supports NFI & Statistics Programme • Develops forest distribution maps and statistical data on forest resources on local levels (province, district, community, village) • Map production and data analyses supported by NFA UN REDD Phase II 2013-2017 FORMIS Phase II 2013-2017 • Development of change detection and carbon monitoring systems (supported by NFA) and benefit distribution mechanism • Development of centralized forestry database and information sharing system NFI Cycle 5 2014-2016 ?
- 3. RS AND MAPPING TOOL DEVELOPMENT Land Use and Forest Type mapping and change detection: • For NFA and future NFI cycles • To support National Forest Inventory and Statistics Programme • To support REDD+ MRV reporting requirements With the following satellite imagery data: • SPOT-5, high resolution, 10 meter resolution, 2.5 pan-chromatic for local level forest type and land use mapping • DMCI, medium resolution 22 m for large scale mapping and change detection With following software: • eCognition (land use and forest type mapping) • FAO OpenFORIS RS toolkit (land use and forest type mapping) • METLA developed Open Source tools (volume map creation and sampling simulations) 3
- 4. WHAT HAS BEEN AND WILL BE DONE • Developed rule sets for land use and forest type classification for SPOT-5 and DMCI imageries using eCognition software • NFA project plans to convert SPOT 5 and DMC satellite imagery based land use and forest type classification tools developed with eCognition to be part of Open Foris RS package to be used by any parties, work is on-going • NFA project has acquired two sets of DMC imageries from North- East part of Vietnam from years 2010 and 2012. This dataset will be used to develop land use and forest type change detection tools to monitor nationwide changes though time, work on-going • NFA plans to integrate annual / biannual change detection from DMCI imagery to be part of NFI cycle 5 • Land use and forest type classification and change detection tools could be utilized by REDD+ initiative 4
- 5. LAND USE AND FOREST TYPE MAPPING, MAIN STEPS 14.11.2011 5 1. Dem preparation: to calculate the slope and aspect 2. Satellite image pre- processing 3. Field data collection 4. Field data division into training and test sets 5. Rule set development and image segmentation 6. Segments classification 7. Accuracy assesment
- 6. EXAMPLE HA TINH PROVINCE, LAND USE CLASSIFICATION USING SPOT-5 6 Land cover class Number of plots Training set Test set Agriculture 38 19 19 Bare 18 9 9 Forest 125 63 62 Residential 47 24 23 Water 14 7 7 Total 242 122 120 • 242 plots, combined in 5 general land cover classes, checked in the field • 35525 segments were classified using 122 sample plots
- 7. HA TINH LAND USE MAP (ECOGNITION) 7
- 8. ACCURACY OF HA TINH LAND USE MAP 8 User Reference Class Forest Agriculture Water Bare Sum Forest 126833 3267 0 0 130100 Agriculture 2944 15240 0 0 18184 Water 0 265 3638 0 3903 Bare 2414 0 0 15919 18333 unclassified 2491 0 0 0 2491 Sum 134682 18772 3638 15919 Producer 0,94 0,81 1 1 User 0,97 0,84 0,93 0,87 Overall Accuracy 0,93 • Object level accuracy using eCognition software • Iterative process based on conflict / confusion matrix
- 9. EXAMPLE HA TINH PROVINCE, FOREST TYPE CLASSIFICATION 9
- 10. ACCURACY OF HA TINH FOREST TYPE CLASSIFICATION 10 Class Total Correct Mixed % of correct Forest garden 5 4 1 80.00 Mixed timber and bamboo 5 4 1 80.00 Natural evergreen broad leaved forest (<100m3/ha) 6 5 1 83.33 Natural evergreen broad leaved forest (>200m3/ha) 5 5 0 100.00 Natural evergreen broad leaved forest (100-200m3/* 8 7 1 87.50 Natural regrowth 10 8 2 80.00 Plantation 23 19 4 82.61 Total 62 52 10 83.87 • Object level accuracy is higher compared to pixel level accuracy • The use of DEM in the process doubles the accuracy
- 11. LAND USE CLASSIFICATION USING DMCI IMAGERY IN BAC KAN PROVINCE 11 • Overall accuracy was 0.89 • Accuracy for forest cover class: 0.81 • The accuracy of forest type mapping was lower than 0.1 • DMCI imagery is suitable for land use classification and large scale change detection, not for forest type classification
- 12. CONCLUSIONS OF WORK DONE 12 • It is possible to produce accurate land cover and forest types maps using object oriented image processing of SPOT 5 data • The slope and aspect calculated from Digital Elevation Model significantly improves the accuracy of segmentation and classification (doubles the accuracy). • Image classification should be implemented in 2 steps approach: «forest/non-forest», «forest types» due to the different groups of features used in classification • Key features for classification: 1. Topography 2. Texture 3. Spectral values • Alternative data source for forest cover monitoring is DMCI
- 13. VOLUME MAP CREATION USING METLA TOOLS, KNN APPROACH, MAIN STEPS 13 Field sample plots SPOT 5 images Detecting clouds and shadows Merging and homogenizing images Selecting bands for making volume map creation Checking raw data Volume calculation for field sample plots Test set: Plots for accuracy assessment Training set: Plots for volume map development Volume Map Accuracy assessment
- 14. FIELD SAMPLE PLOT DATA USED IN BACKAN VOLUME MAP CREATION 14 621 sample plots 2011-2012 from: - NFI & Statistics programme - NFA collected during Bac Kan pilot inventory
- 15. SPOT-5 IMAGE PREPROCESSING 15 Calibrated image: 4 multispectral bands 3 natural color bands 1 panchromatic band
- 16. BAC KAN PROVINCE VOLUME MAP USING 7 NEAREST NEIGHBOURS 14.11.2011 16 - 621 plots: 500 for making volume map, 121 plots for accuracy assessment - Iterations: 10 times RMSE: +/- 65 m3 in pixel level
- 17. POTENTIAL USE OF VOLUME MAP • In sampling simulation and accuracy assessment using tools developed by METLA • Number of plots and clusters needed for desired accuracy • Shape of cluster, to identify statistically most efficient cluster design • Distance between plots inside cluster • Distance between clusters • Used together with land use / forest type map to predict • To be used by NFI & Statistics and UNREDD programmes to predict volumes for smaller geographical units and identify required sampling intensity 17
- 18. THANK YOU! 14.11.2011 18 Contact info: tani.hoyhtya@fao.org, tel. +84 (0)982428147