Jun Chen1,2
1Chief Scientist, National Geomatics Center
2President, ISPRS 25 May, 2016 , Rotterdam, The Netherlands
“Sensors”, GSW 2016
Contents
Methodology
Outlook
Background
Organization
Validation of Land Cover Maps
Documentation of the accuracy of land cover maps allows users to evaluate the utility of these maps for their particular applications
Image pre-processing
classification Verification
LC product Validation
1 2 3
GlobeLand30 - the World’s first 30m Earth Land Cover Map
Chen , et.al., 2014, China: Open access to Earth land-cover map, Nature, 514:434, 23
Oct. 2014
www.globeland30.org
POK based Operational Mapping
5
Chen et.al., 2015. Global Land Cover Mapping at 30m Resolution: a POK-based Operational Approach,
ISPRS Journal of Photogramme try and Remote Sensing , 103 (2015): 7-27
Per-class
extraction
Water
Wetland
Snow & Ice
Artificial cover
Cropland
Forest
Shrubland
Grassland
Barren land
Tundra
Pixel-based classification
Spectral
features
30m imagery NDVI time-series Reference data
Textural
features
Multi-temporal
information
Object-based identification
Multi-scale
segmentation
Automatic
labeling
Object-based
result
Supervised classifiers/Decision trees
Knowledge-based verification
Knowledge
integration
platform
Manual
interaction &
verification
Quality control
and product
integration
Global land cover mapping at 30m resolution
Pixel-based classification
Object-based Identification
Knowledge-based verification
Per-class extraction
Geographical Distribution of Users
111 countries and 6000 users
Geographical Distribution of Users
Download Frequency
90
88
87
85
82 84 86 88 90 92
Hybrid map
GlobeLand30
Hansen's TC
JAXA forest map
Overall accuracy,%
Preliminary Validation of GlobeLand30
Sample points
accuracy
Investigators and
sources
Full class 154,070 83% 5 Chinese institutions
Water ,N. Europe
91% (Prof. Y-F Ban)
GIM, Dec.,
2014
Forest IIASA
Report
Validation at individual countries
Country (Investigator)
Areas Accur. Sources
Germany (Dr. J. J. Arsanjan)
Whole Germeny (CORINE, UA, OSM, and ATKIS)
92% Int. J Digital Earth, 2016(on line)
Greece (Dr. Ioannis Manakos)
Thessaly Region
91% Land, 2015, 4,1-18
Iran (Dr. J. J. Arsanjan)
6 study sites
77.9% Habitat International ,2016,1-7
Italy (Prof Maria Broveli)
8 areas >80% Remote Sensing, 2015(7), 2107-2122
Towards a Comprehensive Validation
Separate and independent accuracy assessments of some 30-m products being made
A comprehensive validation of multiple 30-m GLC maps would allow comparative assessments of uncertainty among different maps
Output from an GEO/ISPRS Workshop (6-7 June 2015, Shanghai)
The workshop "Trust in Spatial Data and Validation of GLC
Products" discussed the issues related to the validation of
30m GLC products with suggestions:
this international validation activity be led by GEO and
ISPRS, co-organized by some other organs (i.e., UN-
GGIM,…).
all the existing 30m GLC products (including GlobeLand30)
be validated with the same approach.
be conducted from Aug. 2015 to the end of 2016.
Contents
Methodology
Outlook
Background
Organization
Major tasks
• Develop advanced tools for facilitating
the validation processes
• Formulate a technical specification suitable for the validation of
30-m GLC data , including the sampling design, response design and analysis protocols
• Organize collaborative validation and report to
GEO plenary
• Mobilize GEO members and participating organizations,
including Integrating all available validation resources and expertise
The Existing Verification Technology
The Top-level Designing of Global Land Cover products
in 30 m Verification
Estimation of Sample Size· Two-ranking sampling· Multiple stratified
sampling
Layout of Sample· Regular grid· Landscape index
· Spatial correlation
Integration of Resources· Semantic mapping· Scale conversion
· Phase integration
Online Check· Interactive
interpretation
· Credibility
Accuracy Assessment· Confusion matrix· Area-weighted overall
accuracy
· Kappa
Technical Specification· Sampling scheme· Sample check· Accuracy assessment
· Online geo-tagging
Global Overall Accuracy
Assessment
Regional Verification
R&D and Testing Online Verification System
Major Methodology
Identify or develop suitable methods for 5 key items
Conceptual design
Develop online tools
Formulate a technical specification (guidelines)
A Landscape-Index Based sampling approach
More reasonable sample densities and sample sizes derived using the LSI approach
Country rLSI Sample size Sample density
SRS
Mauritania 2978 29 Spain 1461 29
Bangladesh 426 29
SWS Mauritania 1910 19
Spain 2115 42 Bangladesh 460 31
LSI Mauritania 1.29 1579 15
Spain 2.68 1627 32 Bangladesh 3.46 617 42
Countries with different spatial heterogeneities
SWS-Shannon index-based cluster sampling
SRS-stratified random sampling
On line Validation System
Collaborative Validation
Region based validation
Global Validation
Continent Based validation
Feature based validation
Country based validation
Coordination
Technical Specification Validation support tools
Sample plan
· Sample plan for region
· Sample plan for country
· Sample plan for continent
· Sample plan for feature
Validation Resources
· GLC products· HR images· Aerial photographs· Geo-tagged pictures· Geo-wiki/DCP· In-situ measures
Assessment
· Assessment report for region
· Assessment report for country
· Assessment report for continent
· Assessment report for feature
· Global assessment report
Sampling
· Collaborative collection of samples from world
· Distribution of samples to world
Contents
Methodology
Outlook
Background
Organization
Key Players
NGCC
GOFC/GOLD
IIASA
Tongji Uni.
KTH
CO-DATA
ISPRS
…
Collaborative Validation
Initiativing Validation
A GEO-led Validation
Establish a GEO-led collaborative framework for the validation of GLC data products
Complete the validation (assessment) of all available multiple 30-m GLC maps
40 countries and PO joined
25个GEO members and 15个others(UN-GGIM、CO-DATA)
序号 国家/组织 序号 国家/组织 序号 国家/组织
1 Bulgaria 14 UAE 27 Mongolia
2 Canada 15 Zimbabwe 28 Thailand
3 China 16 Sweden 29 EU- JRC
4 Czech Republic 17 Rwanda 30 FAO
5 Ecuador 18 Uganda 31 ISPRS
6 Laos 19 Indonesia 32 IIASA
7 Greece 20 Brazil 33 CODATA
8 Italy 21 Mexico 34 GOFC-GOLD
9 Japan 22 Argentina 35 PAIGH
10 Portugal 23 Shirilanka 36 RCMRD
11 Romania 24 Tanzania 37 UNGGIM
12 South Africa 25 Columbia 38 UNESCAP
13 Ukraine 26 Jordan 39 ECA
Countries Participated
Workshop on Validation of GLC Datasets June 28 – 30, Tongji University, Shanghai
http://celiang.tongji.edu.cn/GLCVal2016/Home.html
, Generic approaches for validation of land cover datasets Validation schemes for GLC datasets at various spatial resolutions Allocation of global samples in the case of heterogeneity and
correlation Accuracy assessment in remote sensing-derived GLC datasets Development of online tools for validation of GLC datasets Identification of scientific/technological/application challenges related
to GLC validation Trust and degree of trust in validation of global land cover datasets Others
A Special Issue on ISPRS J Geoinformation
http://celiang.tongji.edu.cn/GLCVal2016/Home.html
present the new concepts and outputs of GLC datasets
validation, including algorithms and models of validation
schemes, sample collection and accuracy assessment, as well as
on-line validation and tagging tools.
not limited to the GEO led international validation campaign..
Validation on Global Land Cover Datasets
Contents
Methodology
Outlook
Background
Organization
Outlook
Establish a GEO-led collaborative framework for the validation of GLC data products
Report both the validation results and the collaborative validation framework to GEO community
Lead to a coordinated regular validation practices