Global land cover validation activities Martin Herold (FSU Jena) & Mark Friedl (UBoston) Global Observations of Forest Cover and Land Dynamics With contributions from Curtis Woodcock (UBoston), Olivier Arino (ESA), Steve Stehman (SUNY) CEOS WGCV LPV Focus Group Lead Meeting, Missoula, Montana, June 15, 2009

Overview 1.Political initiatives driving observation progress and needs for validation –Observing land cover as ECV 2.Global land cover observations and accuracy assessments –MODIS and GLOBCOVER –Best use of existing reference datasets 3.Coordinated effort to develop LC validation database 4.Accuracy assessment for fine-scale land cover change and area estimates –Post-Kyoto agreement and best practices

International drivers 1.United Framework Convention on Climate Change: –Reduce uncertainties in monitoring the global climate system through observing essential climate variables –Capacity building needs to address stronger role of developing countries in post-2012 agreement Major REDD readiness funds are currently being allocated 2.Group on Earth Observation (GEO) task DA-07-02: –Provide a suite of global land cover datasets, initially based on improved and validated moderate resolution land cover maps and eventually including land-cover change at high resolution (task co-lead by USGS and GOFC-GOLD) 3.Global land cover monitoring and assessments: –GLOBCOVER, FAO-Forest Resources Assessm –Operational validation / Efforts for deriving “Best map”

Observing land cover as ECV

DA key activities Global level National level Strategies (IGOS): Integrated Global Observations for land (IGOL) Standards: LCCS land cover classifiers and validation procedures Harmonization: “best” available map New global products: GLOBCOVER (link to regional level) Continuity of observations: Mid-decadal global Landsat survey (MDGLS) Global Land Survey 2010 Specifications for fine-scale global land cover change dataset (incl. validation framework) Technical guidance for UNFCCCC/REDD (GOFC-GOLD sourcebook) Integration of IGOL into GEO Capacity building and support of global assessments: GLCN + GOFC-GOLD networks / FAO-FRA global remote sensing survey

GCOS implementation plan actions for land cover 1.Establish international standards (T22)  UN Land Cover Classification System (LCCS) classifiers 2.Methods for map accuracy assessment (T23)  CEOS WGCV/GOFC-GOLD best practices report 3.Continuity for fine-scale satellite observations (T24)  Commitments to operate Landsat 8 (US) and Sentinel 2 (EU) 4.Develop an in situ reference network (T25)  Global operational validation implementation plan 5.Annual global land-cover products (T26)  Release (and continuation) of GLOBCOVER 6.High-resolution global land cover change (T27)  FAO/FRA 2010 global sampling & GEO definition of specifications

International consensus on technical issues “Best Practices Document” Strahler et al., 2006

Integrated land cover observations IN-SITU (+ IKONOS type) periodically (usually 1-10 yrs) Detailed physionomy Floristics and species distribution Crop type and rotation etc. Effort for frequent update high Thematic detail Spatial detail high low LANDSAT/SPOT – type inter-annual (1-5 yrs) Vegetation physionomy Global archives In situ database Assuming observationcontinuity andconsistency Global daily observations Land type/ Phenology MODIS/MERIS (intra-)annual Completed and endorsed by IGOS partnership and GEO in 2007 From Herold et al 2008, IEEE Systems

GLOBCOVER (2005/6) Dataset release: September 2008

GLOBCOVER validation 10 More than 4300 validation points interpreted by int. experts Blue points: Globcover project (3835 points) + Gond’s set (n~80) (including 225 double interpretation by 2 experts) Brown points: IMWI data (403 points) Global area weighted accuracy: ~73 % based on 3167 reference points Validation report available ~ 15. October 2008

MODIS Collection 5 Land Cover (2001-present)

MODIS Land Cover Validation 1860 Training Sites – Cross Validation

Categories in existing global datasets Terminology: land cover classifiers (LCCS) Deciduous Built up Snow & Ice Bare Herbaceous Shrubs Trees Evergreen Broadleaved Needle-leaved Cultivated/ managed Aquatic/ flooded Translation Classifiers commonly used to characterize land cover worldwide i.e. life form & surface type, leaf type & phenology, terrestrial/aquatic Common classifiers (Terminology standard) Basic set of standardized classes based on combination of common classifiers and independent of any cartographic standard i.e. broadleaved evergreen trees, herbaceous crops, built up area Generic classes (Thematic standard) Application of cartographic generalization (MMU) to generic classes Definition of mixed categories or using density thresholds i.e. Closed to open (>15%) broadleaved evergreen forest (> 5m) Mapping Categories (Cartographic standard)

Thematic standards Reference database (GLC2000) Comparative validation & assessment

Thematic standards Reference database (GLC2000) Comparative validation & assessment Probability

Thematic standards Reference database (GLC2000) Comparative validation & assessment Probability maps For different datasets, classifiers and landscape heterogeneities Synthesis/improved maps Dominant life form Accuracy

Overview 1.Political initiatives driving observation progress and needs for validation 2.Global land cover observations and accuracy assessments 3.Coordinated effort to develop LC validation database (Curtis Woodcock) 4.Accuracy assessment for fine-scale land cover change and area estimates

Overview 1.As the land cover community matures, an increasing emphasis on validation and accuracy assessment - a difficult, somewhat unpleasant and somewhat surprisingly expensive activity 2.The GOFC-GOLD LC IT has decided to try to support the broader community through validation 3.Idea is to collect ground reference data independent of any single land cover product to support validation of many land cover datasets 4.Intent is to supplement and complement ongoing validation activities associated with individual land cover datasets

Operational lc validation framework Degree of usability and flexibility Updated valid./change Validation of new products Design based sample of reference sites In-situ global Primary validation LCCS-based Interpretation (Regional Networks) Reference database: statistically robust, consistent, harmonized, updated, and accessible Updated interpretations Time Comparative validation Existing global LC products Legend translations Product synergy Data reprocessing Link to regional datasets

Operational lc validation framework Effort serves purpose for estimating: –Individual map accuracy / best available map –Area of land-cover classes or land-cover change Sampling design: –10 km by 10 km block (Landsat – MODIS) –Flexible to increase sample size to provide precise country or region specific estimates –Stratification by geographic reporting regions, areas where maps differ, important rare land-cover classes Response design: –Reference data (i.e. SPOT) interpreted by regional experts (i.e. GOFC-GOLD networks) using LCCS classifiers Analysis design: –Error matrix for each map and region –Estimates of class area –Supplementary accuracy information on land-cover composition and landscape pattern

Project Overview (Woodcock Lead) Randomized stratified sampling Primary sampling units by blocks of 10 × 10 km World divided into seven continents Strata based on Köppen’s climate zones together with population density Strata independant of the land cover products Currently about 70 samples per continent planned Validation workshops planned – VHR satellite data (source not determined) interpreted by local experts Landsat data will be used to monitor change at sample sites First workshop will be held in Almaty, Kazakhstan in September 2009

Stratification and Sample Design Seven Continents: North Am./South Am./Europe/Africa/ Former USSR/South Asia/Oceania

Current Status – Early Stages of Stratification and Sample Design Workshop tentatively planned for late July in Boston

Observing land cover as ECV Towards standardization of satellite derived products: 1.Coordinated observations 2.Integrated and standardized mapping and monitoring 3.Independent quality assessment

Overview 1.Political initiatives driving observation progress and needs for validation 2.Global land cover observations and accuracy assessments 3.Coordinated effort to develop LC validation database 4.Accuracy assessment for fine-scale land cover change and area estimates

Fine-scale land cover change Suite of national / regional experiences: –National/regional monitoring programs –UNFCCC Kyoto reporting on LULUCF/AFOLU UNFCCC process on reducing emissions from deforestation in developing countries: –case studies, readiness activities, post 2012 Projects with global/large scale focus: –EU/JRC: TREES 3 (sampling approach) –UMD/SDSU: combined MODIS/Landsat approach FAO-Forest Resources Assessment 2010 Accuracy assessment of area changes?

Satellite observation system/program Technical observation challenges solved Access to information on quality of archived data worldwide Continuous observation program for global coverage Pre-processed global image datasets generated & accessible Image data available in mapping agencies for land change analysis Capacities to sustainably produce/use map products in developing countries OPTICALOPTICAL LANDSAT TM/ETM ASTEROn demand SPOT HRV (1-5)Commercially CBERS 1-3Regionally IRS / Indian programRegionally DMC programProbablyCommercially SARSAR ALOS/PALSAR + JERSRegionally ENVISAT ASAR, ERS 1/2Regionally TERRARSAR-XCommercially IKONOS, GEOEyeProbablyCommercially ICESAT/GLAS (LIDAR) Current availability of fine-scale satellite data sources and capacities for global land cover change observations (Note: dark gray=common or fully applicable, light gray=partially applicable/several examples, white=rare or no applications or examples)

FAO FRA 2010 –remote sensing survey ~ 13,500 monitoring sites

Building national capabilities for IPCC reporting Important componentsPractical considerations FOREST AREA CHANGE Primary source: Landsat-type satellite data DeforestationStarting point for historical assessment (3 time steps minimum) Build basic satellite data proc. capabilities Monitoring of forest degradation Forest fire and burned area Relevance and characteristics for human- induced carbon emissions Definition of suitable monitoring system Accuracy assessmentUsing best/transparent methods and efforts for continuous improvement Prepare for statistically robust approach CHANGE in CARBON STOCKS Primary source: ground/inventory data Existing stratifications and forest carbon estimates Inventory of available data Decide on carbon pool/TIER level to report Towards improved carbon stock change estimations New inventory including other carbon pools Stratification in relevant areas/forest types ACCOUNTING & REPORTING Provide conservative estimates

Accuracy assessment of area change estimates  Uncertainties should be quantified and reduced as far as practicable (IPCC reporting guidelines, LULUCF)  Need for standard/best practice methods (CEOS/GOFC-GOLD)  Capacity development in developing countries: Robust validation may not be achievable or practicable i.e. monitoring historical land changes in developing countries Good practice recommendation if no thorough accuracy assessment is possible or practicable:  apply the best suitable mapping method in a transparent manner  consistency assessment allow some estimation of the quality  Principle of conservativeness: estimates (i.e. C-emissions) should not be overestimated assumes reward for more robust and accurate monitoring system  GOFC-GOLD REDD sourcebook:

Final remarks Deforestation and land change prominent on the political agenda and driving observation progress: Reducing uncertainties in observing the global climate system National capacity development (i.e. climate change mitigation) Key issues for land cover validation: Need for continuous and consistent baseline observations Move towards operational global system (framework exists) Efforts for “best available” land cover map Best practices and good practices for accuracy assessment of land cover change and area estimates Communication with political/policy level (GEO, UNFCCC) Key role for CEOS WGCV for evolving ECV monitoring Sustained support for validation activities