15arspc submission 82
TRANSCRIPT
1
REMOTE SENSING VICTORIA’S PUBLIC LAND FORESTS – A TWO TIERED SYNOPTIC APPROACH
Andrew Mellor, Andrew Haywood
Department of Sustainability and Environment (DSE) 8 Nicholson St, East Melbourne, 3002 VIC
Ph 03 9637 9818, Fax 03 9637 8589 [email protected]
Abstract
This paper describes the development of a new land cover classification system, for land cover mapping and change detection across Victoria’s public land forests (State forests and Parks and Reserves), using aerial photography and time-series medium resolution satellite imagery. Remote sensing derived land cover data and information is one component of the Victorian Department of Sustainability and Environment’s Forest and Parks Monitoring and Reporting Information System (FPMRIS).
The FPMRIS has been implemented to provide the framework and services necessary to assess and monitor the extent, state, condition and sustainable development of Victoria’s public land forests, in a timely and accurate manner and help Victoria meet its commitment to State and National reporting requirements.
The FPMRIS land cover classification system was required to take a synoptic view of public land forests land cover and map the fundamental characteristics of the dominant overstorey of vegetation, irrespective of ecological values or its biogeographic environment. Furthermore, the classification needed to include basic forest structure attributes and broad forest type information and be implemented using new passive and active remote sensing systems in the future.
The main elements of the work undertaken were a review of International, National, State and regional land cover classification systems and their respective advantages and disadvantages in the context of FPMRIS requirements; and, an evaluation of a proposed land cover classification system using two tiers of remote sensing data (high resolution aerial photography and moderate resolution Landsat TM imagery).
The final land cover classification system is a hybrid of the Australian National Vegetation Information System (NVIS) and National Forest Inventory forest types and UN FAO Land Cover Classification System (LCCS) – land cover categories and classifications from NVIS are used to map height, structure and broad floristic formations of primarily vegetated land cover, while FAO LCCS classes are used for mapping cultivated and non-vegetated land covers.
2
Introduction
Increasing concern over the status of global and regional forest resources has led to the implementation of many projects to establish long term operational systems for land cover monitoring (Rogan et al. 2003) and driving the need for more detailed, timely and accurate information on land cover and land cover change at all geographic scales (Ahlqvist 2008). Such information is important, both from an ecological and political perspective. Ecologically, land cover mapping is important to assess environmental health, monitor change, and to use as input to scientific models. Politically, land cover mapping is necessary to meet the commitments to agreements such as the Kyoto Protocol, the international Convention on Biological Diversity, and the Framework Convention on Climate Change.
The Victorian Department of Sustainability and Environment (DSE) is responsible for the sustainable management of public land in Victoria (including the public land forests) and engages a number of processes to monitor the sustainability of the Victoria’s forests. These include the five-yearly Victorian State of the Forests Report, the Sustainability Charter for Victoria’s State forests (Department of Sustainability and Environment 2006), Victoria’s State of the Parks Report (Parks Victoria 2007) and associated Criteria and Indicators for Sustainable Forest Management in Victorian Forests (Department of Sustainability and Environment 2007a). Victoria’s State forests, Parks and Reserves (hereafter referred to as public land forests) are managed for wood production and the provision of non-wood production values including recreation, biological and landscape diversity. Given that they provide many multi-value functions there is a necessity to monitor the sustainable management of Victoria’s public land forests and to determine the causes of change in extent, state and condition of the forests. A refined list of 7 criteria and 45 associated indicators have been developed through the Australian Montreal Process Implementation Group (MIG), against each of which DSE collates information under the Sustainable Forests (Timber) Act 2004 through Victoria’s 5-yearly State of the Forest Report –to the Minister for Water, Environment and Climate Change from the Secretary, DSE. The seven criteria are biodiversity, productivity, health and vitality, soils and water, carbon, socioeconomic and legal.
These reporting mechanisms are designed to enable Victoria to critically assess and evaluate progress towards achieving its sustainable forest management objectives and targets. To meet its reporting requirements, a DSE Forest and Parks Monitoring and Reporting Information System (FPMRIS) has been established, which utilizes a systematic ground-based sampling grid, stratified by Interim Biogeographic Regionalisation of Australia (IBRA) Regions (Version 4.1) and located across the state, and high and moderate spatial resolution remote sensing data. These three sources (or tiers) of data, are used to capture a set of forest attributes and monitor and report on the extent, state and condition of Victoria’s public land forests.
3
FPMRIS Sampling and Data Sources
Table 1 shows the three primary data sources (or ‘tiers’) in the FPMRIS (hereafter referred to as the ‘monitoring system’), together with data primitives and their associated monitoring themes, that will be used to report against monitoring and reporting criteria and indicators described above.
Table 1: Forests and Parks Monitoring and Reporting Information System - Monitoring Theme and Data Sources
Data Sources Data Primitives Monitoring Themes
Tree DBH
Tree Height
Tree Species
Diameter Distribution
Tree Mortality
Crown Class
Crown Health Class
Coarse Woody Debris
Ground: Large Tree Plot (Tier 1)
Slash Piles
Ground: Small Tree Plot (Tier 1)
Stumps and small trees
Dead plants frequency/height Ground: Vegetation Quadrants (Tier 1)
Living plants frequency/height
Bird abundance Ground: Bird Survey (Tier 1)
Bird diversity
Soil type
Soil Nutrient Status
Duff Ground: Soil Pits and Sampling (Tier 1)
Fine woody debris
Forest Type
Forest Structure
Land cover High Resolution
1 Remote Sensing (Tier 2)
Disturbance
Moderate Resolution Remote Sensing (Tier 3)
Forest extent and cover
Tree Growth
Tree Mortality
Site Productivity
Above-ground Biomass
Below-ground Biomass
Flora Diversity
Canopy Health
DWD
Below-ground Carbon
Nutrient Status
Fauna Diversity
Growth Stage
Old Growth
Canopy Disturbance
Fragmentation
Forest Area by Forest type
Remote sensing components of the monitoring system represent tiers 1 and 2 of the system – respectively, high resolution data (sample coverage) and moderate spatial resolution data (‘wall-to-wall’ State-wide). Approximately 800
1 Unless otherwise specified, high resolution and moderate resolution refer specifically to spatial resolution
(the measure of the smallest area identifiable on an image as a discrete separate unit).
4
on-ground permanent measurement plots (tier 1) distributed across the entire public land forest estate will supply land cover classification “training” and verification data for remote sensing data analysis components of the system – namely land cover mapping and land cover change detection. Remote sensing data will be used to derive information about forest extent, forest type, structure, land cover and fragmentation metrics at landscape and local scales and to generate spatial and aspatial data over time. Figure 1 shows the locations of ground plots across the Victoria’s public land forests.
Accurate and up-to-date land cover information is necessary for a successful monitoring system. This paper describes the development of a land cover classification system, designed to be used for mapping land cover (including forest types and their broad structures) across Victoria’s public land forests for the DSE Forests and Parks monitoring system. The paper includes a description of the monitoring system’s specifications and conditions, a review and suitability assessment of existing land cover classification and mapping approaches used in Australia and proposes a system to be evaluated for land cover mapping in the monitoring system.
Figure 1: Location of ground plots across the project area
5
Definitions
The following definitions will be adopted by the monitoring system and are conditions of the land cover classification system against which the suitability of existing land cover classification systems for the monitoring system suitability will be reviewed.
Forest
Concepts and definitions of forest are typically diverse and sometimes conflicting. As Victoria reports forest at a state level for national reporting, the National Forest Inventory (NFI) definition of forest has been adopted for use in the monitoring system:
“A land area, incorporating all living and non-living components, dominated by trees having usually a single stem and a mature or potentially mature stand height exceeding two metres and with existing or potential crown cover of overstorey strata about equal to or greater than 20 per cent. This definition includes native forests and plantations and areas of trees that are sometimes described as woodlands”
(Wilson and Talbot 2003)
Minimum Mapping Unit
In classifying and mapping land cover, a minimum mapping unit, is required to standardise data and because the concept of a particular land cover type or unit (such as a forest or other vegetation community) is at a larger scale than components that make up the land cover unit. In general, a minimum mapping unit refers to the smallest area entity mapped as a distinct area and determines the extent of detail conveyed by an interpretation (Saura 2002). At the time of publication, as the organisation responsible for collating forest data and area statements from Australian State and Territory forest management agencies and organisations, NFI does not impose a minimum unit area on forests, but generally resolves forest areas in the range of one to five hectares (Howell 2009). In some circumstances, it may be appropriate to apply more than one minimum mapping unit in a land classification system. Given that ‘forest’ is the dominant land cover (by unit area) in the monitoring system project area, a minimum area of 0.5 hectares was incorporated into the land cover classification system conditions, based on the Food and Agriculture Organization of the United Nations Forestry Department definition of forest:
“Forest Land spanning more than 0.5 hectares with trees higher than 5 meters and a canopy cover of more than 10 percent, or trees able to reach these thresholds in situ. It does not include land that is predominantly under agricultural or urban land use”
(FAO Forestry 2005)
It is recognised that the implementation of a single minimum mapping unit may effect the classification and reporting of some land cover types and
6
extents, this is an issue will be further investigated through the land cover classification mapping and evaluation phases of the project.
Land Cover Classification
Land cover results from a complex mixture of natural and anthropogenic influences and is the composition and characteristics of land surface elements (Comber et al. 2005). Land cover refers to what is physically on the earth’s surface. Typically, land cover is used to describe both land cover (e.g. vegetation and built up areas), as well as the land itself (e.g. bare soil, sand - desert etc.). Land use differs from land cover, in that it is a description of the primary socio-economic role of a given location, be it industrial, agricultural, forestry, recreational or conservancy (Castilla and Hay 2006). The intimate relationship between the two terms, mean that ‘land cover’ and ‘land use’ are frequently used interchangeably in many maps and classification systems.
‘Classification’ is an abstract representation of the situation in the field using well-defined diagnostic criteria and the ordering of objects into groups on the basis of their relationships using a systematic framework with names of classes and criteria to distinguish them (Di Gregorio and Jansen 2000); (Sokal 1974). A classification requires the definition of class boundaries which should be clear, precise, and possibly quantitative and based on objective criteria (Di Gregorio and Jansen 2000). A classification system should be both scale independent (classes at all levels of the system should be applicable at any scale or level of detail); and, source independent – meaning independent of the method by which information is collected (e.g. field survey, aerial photograph, satellite image etc.).
Land Cover Classification System Requirements and Conditions
To meet its reporting requirements, the land cover classification system to be adopted by the FPMRIS, needed to meet the conditions outlined in Table 2.
7
Table 2: Conditions of the FPMRIS Land cover classification System
Condition (or requirement) Description/definition
Synoptic land cover mapping of vegetation
Primarily vegetated land cover types need to be mapped based upon the fundamental characteristics of the dominant overstorey of vegetation, irrespective of ecological values or its biogeographic environment. Furthermore, as remote sensing data is to be the primary information source for mapping using the FMRIS Land Cover Classification System. The system should focus on a synoptic view of land cover – with the identification and classification of land cover types recorded from what can be “seen” from above.
Classification of broad forest structure attributes
Given available data, forest stands may be attributed with information pertaining to forest type and structure, including broad species, height and canopy cover.
Scale-independent mapping
Land cover types should be classifiable and mapped using the classification system, independent of scale of remote sensing data used (i.e. high resolution, high detail, small scale, local mapping vs. moderate resolution, less detailed, medium scale, large area mapping). At different scales (e.g. using high vs moderate remote sensing data sources, some classes may be aggregated or disaggregated within the classification system.
Remote Sensing (technology) independent
Land cover classification system can be applied regardless of remote sensing data/sensor or mapping tools and techniques. Land cover classes need to be mapped consistently overtime, irrespective of new remote sensing data, mapping processes (e.g. new automated/semi-automated image classification techniques) and technology (e.g. optical, ranging data, software).
Include all potential land covers in the project area The land cover classification system should be capable of classifying and mapping all land covers across Victoria’s public land forests.
In the context and requirements of the FPMRIS, alternative ways to classify, map and register land cover across predominantly vegetated landscapes are assessed. A land cover classification system for the FPMRIS is proposed to be tested and employed to map land cover across Victoria’s public land forest using high and moderate resolution remote sensing data.
United National Food and Agriculture Organisation (UN FAO) Land Cover Classification System (LCCS)
The UN FAO LCCS is intended to meet specific user requirements for any land-cover classification initiative anywhere in the world. Land cover classes in the LCCS are defined by a set of feature description terms or ‘classifiers’. By using criteria based classes rather than classes based on nomenclature, these
8
classifiers are standard diagnostic criteria that describe local categories and allow for comparison with other existing land cover classifications.
The main goals of the UN FAO LCCS program include responding to a need for land cover data for a variety of end-users; applying a methodology in mapping exercises, independent of the means used, which may range from high resolution satellite imagery to aerial photography; linking existing classifications and legends, allowing comparison and correlation; and, harmonizing principles and methodology for land cover mapping (Di Gregorio and Jansen 2000).
The UN FAO LCCS is intended to be both scale and source independent (Wulder and Nelson 2003). The LCCS comprises two main phases – a Three-level Dichotomous Phase (shown in Table 3) and a Modular Hierarchical Phase, whereby each of the eight dichotomous phase level 3 classes – such as bare areas, natural terrestrial vegetation etc. – may be combined with environmental and technical aspects (e.g. soils, floristic characteristics, forest structure, salinity etc.) and provide additional land cover information (Thackway and Atyeo 2006), thereby offering a classification system capable of capturing any land cover identified in the world (Di Gregorio and Jansen 2000).
9
Table 3: Dichotomous Levels of the UN FAO LCCS from (Thackway and Atyeo 2006)
Level 1 Level 2 Level 3
Managed terrestrial Area
Terrestrial Natural and Semi-natural terrestrial vegetation
Cultivated aquatic areas
Primarily Vegetated
Aquatic or regularly flooded
Natural and Semi-natural aquatic vegetation
Artificial surfaces
Terrestrial
Bare areas
Artificial water bodies, snow and ice
Primarily Non-vegetated
Aquatic or regularly flooded
Natural water bodies, snow and ice
Modular Hierarchical Phase elements that may be defined include land cover height, leaf type, phenology and life form and environmental land form, lithology and floristics. The design of the UN FAO LCCS classifiers followed a requirement to define boundaries so that they are clear, precise, quantitative and based on objective criteria. (Ahlqvist 2008) considers a major limitation of this approach is that classifiers (or parameters) are a rigid and “compartmentalised” view on land cover categories – which possibly results in a failure to truly represent many real-world situations, where classes are often found to be vague and partly overlapping (Ahlqvist 2008).
The National Forest Inventory (NFI)
The Australian National Forest Inventory (NFI) collects and communicates information on Australia's forests. It aims to provide a single authoritative source of data at the national level. The NFI is a partnership between the Commonwealth and all State and Territory governments. The Commonwealth provides funding for the NFI's three staff and for core activities, including the collection of data and the dissemination of value-added information.
NFI has developed agreed standards and protocols for compiling, analysing and reporting forest types over the last decade, which has formed one of the inputs in developing the guidelines for the National Vegetation Information System (NVIS).
Guided by a Commonwealth, State and Territory represented steering committee, NFI collects information about Australia’s forests from various land management agencies and companies – which varies in scale, data source, purpose and method of data collection and land tenure. NFI aims to add value to these disparate datasets, by standardizing them so that they can be combined – thereby developing an overview of Australia’s forests and for comparison between States and Territories. The NFI is also involved in the
10
development of protocols to promote nationally consistent data collection and data management standards.
Data produced by the NFI are used by many different organisations and individuals, including non-government organisations, forest growers, wood processors, consultants, State and Commonwealth agencies, and investors.
The NFI data dictionary has no land use or economic classes and is based on current forest cover. Each forest mapping unit has 3 core attributes, forest type (eg. Acacia, Eucalypt, Mangroves), crown cover (woodland, open, closed or unknown), and height (low, medium, tall or unknown). These three attributes are then combined to generate a forest formation. There are 71 different forest formations (64 native forest classes, 4 plantation classes and 3 non forest classes) (NLWRA 2001). The 11 NFI forest types are:
1. Acacia
2. Banksia
3. Callitris
4. Casuarina
5. Eucalypt (including Eucalypt Mallee)
6. Leptospermum
7. Mangroves
8. Melaleuca
9. Mixed Species
10. Other
11. Rainforest
For national reporting purposes, forests are grouped into eight native forest types defined by dominant species (acacia, callitris, casuarina, eucalypt, mangrove, melaleuca and rainforest) and other (comprising banksia, Leptospermum, mixed species and other) (National Forest Inventory 2003). The crown cover and forest canopy height definitions and thresholds are consistent with the National Vegetation Information, described in the following section.
National Vegetation Information System (NVIS)
NVIS is a collaborative program between Australia’s Commonwealth States and Territory Governments. The NVIS framework was developed following concerns over the need to manage vegetation across all land tenures and compile information about the extent, use and condition of all vegetation (including forests), across Australia (Thackway et al. 2007). The NVIS framework is a means to describe and represent vegetation types, by establishing a link between structural and floristic information in a relational database management system.
An NVIS information hierarchy (Table 4) is used to define and standardize structural and floristic information, and provide frameworks for quality control and assurance of vegetation description and generating map products at the various levels. The information hierarchy is based firstly on structural information and secondly on dominant genus and growth form collected at the sub-stratum level (ESCAVI 2003), described as sub-association.
11
Table 4: NVIS information hierarchy, reprinted from (Thackway et al. 2007)
Hierarchy Level Description NVIS structural/floristic components required
I Class Dominant growth form for the ecologically or structurally dominant stratum
II Structural Formation Dominant growth form, cover and height for the ecologically or structurally dominant stratum
III Broad Floristic Information
Dominant growth form, cover, height and dominant land cover genus for the upper most or the ecologically or structurally dominant stratum
IV Sub-formation Dominant growth form, cover, height and dominant genus for each of the three traditional strata (i.e., upper, mid and ground)
V Association Dominant growth form, height, cover and species (3 species) for the three traditional strata (i.e., upper, mid and ground)
VI Sub-association Dominant growth form, height, cover and species (5 species) for all layers/sub-strata
Source: (ESCAVI 2003)
The NVIS database contains over 9000 distinct vegetation types, from which 23 Major Vegetation Groups (MVG) have been derived to summarise the type and distribution of Australia’s native vegetation.
The classification used to develop the MVGs is, in broad terms, based on typical aggregations of the structure (especially height and cover), growth form and floristic composition (vascular plant species) in the dominant stratum of each vegetation type in the NVIS database (Department of the Environment and Water Resources 2007). MVGs contain different mixtures of plant species within the canopy, shrub and ground layers – but vegetation within each group is structurally similar and often dominated by a single genus. 67 Major vegetation subgroups (MVSs) have been developed from the MVGs – which are based on the typical understorey characteristics as well as the woody stratum and further identification of floristic affinities (Department of the Environment and Water Resources 2007).
The NVIS information hierarchy includes eight stratum and sub-stratum (including tallest tree sub-stratum, tallest shrub layer and tallest ground species), eight height classes assigned to growth form (single and multi-stemmed woody plants and lower storey plant forms), cover (assigned to stratum or sub-stratum) and stratum/sub-stratum dominance (indicated by relative biomass across a vegetation type). Tables 5 and 6, show NVIS height and cover classification thresholds.
12
Table 5: NVIS height classes - based on (Walker and Hopkins 1990)
Height Growth Form
Height Class
Height Range (m)
Tree, vine (M & U), palm (single-stemmed)
Shrub, heath shrub, chenopod shrub, ferns,samphire shrub, cycad, tree-fern,grass-tree, palm (multi-stemmed)
Tree mallee, mallee shrub
Tussock grass, hummock grass, other grass, sedge, rush, forbs, vine (G)
Bryophyte, lichen, seagrass, aquatic
8 >30 Tall
7 10-30 Mid Tall
6 <10 Low Mid
5 <3
Low
4 >2 Tall Tall
3 1-2 Mid Tall
2 0.5-1 Low Mid Tall
1 >0.5
Low
Low Low
Table 6: NVIS cover classes – based on (ESCAVI 2003)
Cover type Cover Ranges (%)
Foliage cover† 70-100 30-70 10-30 <10 » 0 0-5
Crown cover* >80 50-80 20-50 0.25-20 <0.25 0-5
% Cover** >80 50-80 20-50 0.25-20 <0.25 0-5
† defined from each stratum as the ‘proportion of the ground, which would be shaded if sunshine
came from directly overhead'. * Crown Cover (canopy cover) as per Walker & Hopkins (1990).
** The percentage cover is defined as the percentage of a strictly defined plot area, covered by
vegetation.
NVIS classifications of tree growth form height (m) and crown cover (%) are consistent with Australian National Forest Inventory (NFI) standards, whereby tree (or forest stand canopy) heights are classed as Low, Medium and High (2-10 m, 11-30 m and > 30 m respectively) and crown cover as Woodland, Open and Closed (21-50%, 51-80% and > 80% respectively).
Ecological Vegetation Classes (Victoria)
Ecological Vegetation Classes (EVCs) are groupings of vegetation communities based on floristic, structural and ecological features. They are the basic mapping units used for biodiversity planning and conservation assessment at landscape, regional and broader scales in Victoria (Oates and Taranto 2001).
13
EVCs are derived from large-scale vegetation type and plant community mapping and are based on the following information:
• Plant communities and vegetation types (including species and structural information);
• Ecological information relevant to the species that comprise the communities (including life-form and reproductive strategies); and
• Information that describes variation in the physical environment (including aspect, elevation, geology and soils, landform, rainfall, salinity and climatic zones).
EVCs are defined at a qualitative level by their floristic and structure attributes and include in their description the ecological processes which characterise them. Furthermore, the EVC allow the link to be made between vegetation patterns and broad landscape features, such as coasts, lakes, plains, mountains etc. EVCs are normally floristically and structurally distinct from each other.
Beginning in 1992, EVC datasets have been developed as part of various studies. Updates to EVC datasets are infrequent and typically associated with projects and flora surveys (Abuzar and Morris 2000). To date, 1:100 000 scale mapping of EVCs on public land has been undertaken throughout Victoria except for the northern half of the Wimmera and the Riverine. 1:25 000 scale EVC mapping has been undertaken in the Gold-fields Ironbark, Greater Grampians Gippsland and Westernport-Port Phillip (Davis et al. 2002).
A Victoria-wide modelled 2005 EVC dataset has been prepared at 1:100,000 scale, through an intersection of IBRA bioregions, pre-1750 EVC data and the current extent of native vegetation data. Bioregional Conservation Status (BCS) and geographic occurrence has been applied to each Bioregion EVC unit and in November 2007 the BSC was updated following revised native vegetation mapping. The BSC indicates the degree of EVC alteration since European settlement in Australia and takes into account the EVCs current level of depletion and remaining stand degradation levels. BSC categories include presumed extinct, endangered, depleted, rare and least concern.
The common name given to each EVC describes its key elements and typically includes a description of broad environmental or structural characteristics. For example, “Plains (A) Grassy (B) Woodland (C)”, where A is an adjective specific to the vegetation described; B refers to the general aspect of the vegetation; and, C is a noun which is the usual structure of the vegetation (Department of Sustainability and Environment 2007b). Table 7 shows examples of EVC descriptive attributes.
14
Table 7: Examples of EVC Descriptive Attributes
Type (broad) EVC Descriptive Part Example(s)
Environmental (broad) A or B (adjectives)
Altitude (e.g. lowland, alpine etc.) Climatic (e.g. Warm, temperate, semi-arid etc.) Temporal (e.g. Ephemeral, intermittent, permanent etc.) Geological (e.g. Granite, Basaltic, Limestone etc.)
Environmental (non-specific) A or B (adjectives or pronouns)
Alkaline Aquatic Bird colony Chenopod Flood Halophytic
Structure/life form C (nouns)
Forest Grassland Heathland Herbland Mallee Mixed Forest Rainforest
Statewide Forest Resource Inventory (SFRI)
Statewide Forest Resource Inventory (SFRI) was a one-off systematic inventory of the native forest on public land in Victoria, designed to create a complete inventory of Victorian State forest resources to enable timber volume estimation. Detailed aerial photograph interpretation (API) and model-based field sampling were the main techniques used to create the inventory. Initiated in 1994, the project’s stated mission was “to provide forest managers with a reliable, timely and complete set of forest resource information for making informed and consistent sustainable yield forecasts, and decisions on forest land-use planning and resource allocation” (Department of Natural Resources and Environment 1997). The original strategic goals of the project included publishing a complete and consistent inventory of Victoria’s public native forest resources, providing updated resource information for the review of sustainable yields for State forest in each Forest Management Area and integrating the collection and management of forest resource information with other related natural resource information projects.
The method involved stereo aerial photo interpretation and mapping of eucalypt forest stands (at an average of 10 ha) and attributing each stand with the following information:
• Eucalypt species present;
• Crown cover projection;
• Crown form class (reflecting stages of maturity and development);
• Height class;
• Stand disturbance.
15
A forest stand classification coding system (comprising some or all of the information described above) was mapped into measurable forest stands. These stand boundaries were determined mainly be interpretation of distinct eucalypt single species or ‘species alliance boundaries’ (Department of Natural Resources and Environment 2000). Interpretation and mapping of stands used the following order of mapping to determine the stand boundary: 1) Species, 2) Crown form, 3) Crown cover, and 4) Stand height. Forest stand outlines were required to be well defined and without intricate linework that would define individual trees. Most mapped stands were between 10-20 hectares. Table 8 shows a summary of information attributed to each forest stand in SFRI.
Table 8 SFRI forest stand classification attribution adapted from Department of Natural Resources and Environment (2000).
Classification Attribute Attribute Description Attribute Examples
Eucalypt Species
Species groups - recorded according to their relative proportions within the stand crown cover they occupied
Single Species Eucalypt Stand (where a single species eucalypt occupied 76% or more of the stand crown cover); Mixed Species Eucalypt Stand (two or more species, each of which occupied 26% or more of the stand crown cover).
Crown Cover Percentage of crown area projection to land area
1: crowns occupy 0–9% of the forest stand area, 4: crowns occupy 50–69% of the forest stand area, 6: crowns occupy 85–100% of the forest stand area
Crown Form
Eucalypt crown and stem characteristics (to classification growth stages of a eucalypt over time).
Irregular, moderately regular, regrowth and associated abuance of each in a stand (e.g. t: crowns occupy 1–9% of the forest stand crown cover, e.g d: crowns occupy 50% or more of the forest stand crown cover)
Crown Size Average crown size of the most abundant crown class within a forest stand
0-5 m, 5-9.9 m, 10-14.9 m….20-24.9 m and > 25 m
Stand Height Average height of the most abundant crown form class, stratified into 5m height intervals
<5.0 m, 5.0-14.9 m, 15-27.9 m…..40.0-51.0 m….> 51.0 m
Stand Disturbance Evidence and severity of disturbance recorded primarily as an indication of stand condition.
Negligible, low, moderate and severe Forest utilisation, recent fire events and Psyllid insect attack.
Proposed Monitoring System - Land Cover Classification System
Table 9 is a summary of the reviewed land cover and vegetation classification, mapping and information systems against the conditions of a FPMRIS land cover classification system, for classifying and mapping Victoria’s public land forest land cover. Table 9 includes a summary of criteria used to classify vegetation and other land cover types.
16
Table 9 Land cover/vegetation classification/mapping systems suitability assessment
FPMRIS LCCS Conditions
Landcover/Vegetation Classification/mapping System
Land Cover Classification System focus
Forest stand species and structural attribution
Synoptic land cover mapping of vegetation
Scale-independent mapping
Remote Sensing (technology) independent
Potential to include all likely land cover types
Comment
UN FAO Land Cover Classification System
Multi-use land cover mapping, including land use elements.
In part YES YES YES YES
Suitable for classifying broad land cover types, but standard LCCS vegetation class height and cover thresholds are inconsistent with Australian NFI forest height and forest canopy cover classes. Flexibility to incorporate new (and revise existing) non-forest land covers in the future. Includes both land cover and land use classes.
National Vegetation Information System (NVIS) and National Forest Inventory
Nationally consistent Mapping of Physiognomy, floristics and vegetation structure information
YES YES YES YES NO
Suitable for classifying and mapping broad forest type, height and canopy cover, consistent with the Australian NFI. Suitable for applying a synoptic approach to mapping, using the top 3 levels of the NVIS information hierarchy. Only applicable for primarily vegetated land covers.
Ecological Vegetation Classes (EVC)
‘Biodiversity planning and conservation assessment’
YES
In part (native vegetation extent)
In part YES NO
Includes interpretive information which relies on environmental and biogeographical ancillary information, such as climate and salinity, as well as the conservation status of the vegetation.
State-wide Forest Resource Inventory (SFRI)
Mapping Commercial timber species and structures
YES primarily commercial species, structure in State Forest
In part YES NO NO
Primary focus of system is commercial (predominantly eucalypt) forest species. High level of detail around stand species and structure information – unsuited to change detection and landcover classification using most types of remote sensing data (i.e. non-stereo aerial photography).
17
Based on the outcomes of the land cover classification systems evaluation against the FPMRIS LCCS conditions (Table 9), a land cover classification system combining forest stand classification (type and structure) attribution of NVIS and NFI together with non-forest land cover categories from the UN FAO LCCS, is recommended for classifying and mapping land cover across Victoria’s public land forests, using high and moderate resolution remote sensing data. Figure 2 shows the proposed land cover classification system.
Figure 2: Proposed FPMRIS land cover classification system
Application of the Proposed FPMRIS Land Cover Classification System
Two-tiers of remote sensing data will be classified to map land cover over Victoria’s public land forests, using the proposed land cover classification system (Figure 1). These are 800 (2 x 2 km) ‘photoplots’ distributed across the State (stratified by bioregion) – representing a sample of approximately 4% of the total State forest and Parks and Reserves land area (about 7.2 million hectares). The 800 photoplots representing high resolution ‘tier 1’ data, will be mapped to generate accurate aspatial statistics and accuracy statements, on forest type, area and structure (height and canopy cover) and
Land area (mapping polygon)
≥0.5 ha
Predominantly Vegetated Predominantly Non-vegetated
Forest
“A land area, incorporating all living and
non-living components, dominated by
trees having usually a single stem and
a mature or potentially mature stand
height exceeding two metres and with
existing or potential crown cover of
overstorey strata about equal to or
greater than 20 per cent"
National Forest Inventory
This definition includes native forests
and plantations and areas of trees that
are sometimes described as
woodlands.
Non-Forest (predominantly
vegetated)
Attribute: vegetation land cover type
Grassland
Graminoid crops
Non-graminoid crops
Shrubland
etc.
Source: UNFAO Land Cover
Classification System (LCCS) derived
land covers
Attribute: Major Forest Type
1. Acacia
2. Banksia
3. Callitris
4. Casuarina
5. Eucalypt (including Mallee Eucalypt)
6. Leptospermum
7. Mangroves
8. Melaleuca
9. Mixed Species
10. Plantations*
11. Rainforests
Source: National Forest Inventory,
Australian Forest Profiles (Forest
Types)
Attribute: Forest (stand) Height
1. Low (2-10 m)
2. Medium (11-30 m)
3. Tall (>30 m)
Source: National Vegetation Information
System (NVIS) - Structural Formation
(Height classification) of the dominant
growth form
Attribute: Forest (stand) Crown Cover
1. Woodland (20 - 50%)
2. Medium (51 - 80%)
3. Tall (>80%)
Source: National Vegetation Information
System (NVIS) - Structural Formation
(Crown cover) of the dominant growth
form
(Predominantly) Non-vegetated
Attribute: non-vegetated land cover
type
Non-natural surfaces
Bare ground (Soil)
Base groun (Rock)
Water
Shrubland
etc.
Source: UNFAO Land Cover Classification
System (LCCS) derived land covers
18
other (non-forest) land cover types by area, stratified by IBRA bioregion and public land tenure.
The mapped high resolution plots will be used as training data for the classification of the second tier of remote sensing data (moderate spatial resolution remote sensing data) which covers all 7.2 million hectares of the public land forest estate. The classification of the second-tier remote sensing data will apply land cover types from the dichotomous stage of the proposed land cover classification (e.g. forest, non-forest/vegetation, primarily non-vegetated). Figure 2 shows the distribution of high resolution remote sensing data across the project area (moderate resolution remote sensing data covers all of the National Parks and Reserves and State forest across Victoria).
Figure 3 Location of ground plots and high-resolution 2 x 2 km photoplots across Victoria’s public land forests
Semi-Automated Aerial Photography Interpretation
The proposed land cover classification system is being applied to generate a 2008 Victorian public land forests land cover baseline, using a semi-automated digital aerial photographic interpretation of 30-50 cm resolution colour and near infra-red aerial photography acquired between 2006 and 2010. Aerial photographs are ‘segmented’ using a multi-resolution image segmentation routine, to delineate common land cover areas (or landscape elements) and attribution forests stand type and canopy height. The generated segmentation polygons (objects) are then verified and classified through traditional manual Aerial Photograph Interpretation (API) techniques using comprehensive protocols. Final segmentation objects are then attributed accordingly (e.g. non-forest or non-vegetated land cover type, or forest type, height class and canopy cover class).
19
The automated delineation of landscape elements (i.e. image segmentation) is advantageous, as it minimises subjectivity in the placement of boundaries between land cover types (as well as reducing the time taken to delineate boundaries). Furthermore, the use of segmentation parameters which are based on objective criteria ensures an approach to mapping land cover which is robust and repeatable.
Verification of the final classification product will be undertaken throughout the API work, designed to verify accuracy of the land cover classification product and identify any areas of increased error. Improvements to the API protocols will be made accordingly. An accuracy assessment, based on a probabilistic sample design, will be also undertaken and reported.
The initial land cover classification will be based on the actual date of the aerial photography acquisition. The mapped land cover will then be ‘modelled’ to a 2008 baseline date, chosen to be March 2008 – which is coincident with most of the Moderate Resolution remote sensing scene acquisition dates (or seasonal conditions, where the best available acquisition is late summer 2009). Logging history and fire history ancillary GIS data, will be used to model pre and post March 2008 images to the March 2008 baseline.
Modelling to the baseline date will be undertaken by ‘cutting out’ (removing) or attributing as disturbed, API mapped land covers in photography acquired between January 2006 and February 2008, which intersect with burnt or cleared after the aerial photography acquisition date and before march 2008. Areas in post-March 2008 acquired aerial photography, identified to have been burnt or cleared after March 2008 and before April 2010, will be ‘re-attributed’ using information from surrounding ‘undisturbed’ land cover polygons or using other ancillary data or imagery.
Forest Extent Mapping using Landsat TM
Land cover information derived from aerial photography, will be used to ‘train’ 19 Landsat TM scenes covering Victoria that have undergone relative radiometric, geometric and topographic correction. Two main approaches to land cover classification of moderate resolution remote sensing data will be tested, evaluated and applied to generate forest extent layers for years 2003/04 and 2008/09:
Machine Learning (e.g. Neural networks, Random forest, logistic regression) – (Figure 3) and established multi-spectral image classification techniques.
20
Landsat Bands 1-7
Raw – Input Data 2008 Land Cover
DTM
Slope
Aspect
Wetness
Gullies
Ridges
Climate data
Ground plots (~500)
API data (800 x 2km x km photo plots)
Veg. Indices
Topo. Data
Climate Data
e.g. Neural networks, Random forest, logistic regression etc.
Forest Type/Land cover 2008
Forest / Non-Forest Mask Layer
Land cover change analysis
2003
2008
2009
2010
? ??
Existing Forest data E.g. SFRI, Tree25 etc.
Other Forest Data
Error matrix
Training data
Figure 4: Machine learning flow diagram for moderate resolution remote sensing land cover/forest classification
Conclusion
The research undertaken involved a review of existing land cover classification systems and an evaluation of their suitability with respect to several conditions for the FPMRIS. The proposed land cover classification system for mapping land cover across Victoria’s public land forest (through the FPMRIS) is a hybrid of the National Vegetation Information System (including National Forest Inventory forest formations) and the UN FAO Land Cover Classification System. The former will be used to classify broad forest types and their structural attributes (stand height and cover) and the latter, to classify non-forest primarily vegetated and non-vegetation land cover types. Mapping using the proposed classification system will be undertaken using high and moderate resolution remote sensing (digital aerial photography and Landsat TM). The proposed land cover classification system is currently being implemented through the generation of 2008 public land forest land cover area statistics, using semi-automated classification of 800 2 x 2 km aerial photo ‘plots’. The next phase of the research will focus on the generation of state-wide forest vegetation layers for years 2003 and 2008, using Landsat TM and aerial photography generated land cover ‘training’ data.
21
References
ABUZAR, M., and MORRIS, M., 2000, Land Use Mapping of East and West Gippsland Catchment Management Authority Regions, Department of Natural Resources and Environment.
AHLQVIST, O., 2008, In search of classification that supports the dynamics of science: the FAO Land Cover Classification System and proposed modifications. Environment and Planning B: Planning and Design, 35, 169-186.
CASTILLA, G., and HAY, G. J., 2006, Uncertainties in land use data. Hydrology and Earth System Sciences Discussions, 3, 3439 – 3472.
COMBER, A., FISHER, P., and WADSWORTH, R., 2005, What is land cover? Environment and Planning B: Planning and Design, 32, 199-209.
DAVIS, J. B., OATES, A. M., and TRUMBULL-WARD, A. V., 2002, Ecological Vegetation Class Mapping at 1:25,000 in Gippsland, Department of Natural Resources and Environment.
DEPARTMENT OF NATURAL RESOURCES AND ENVIRONMENT, 1997, SFRI Victoria's Statewide Forest Resource Inventory Program Overview, Department of Natural Resources and Environment.
DEPARTMENT OF NATURAL RESOURCES AND ENVIRONMENT, 2000, Statewide Forest Resource Inventory: Forest Stand Classification.
DEPARTMENT OF SUSTAINABILITY AND ENVIRONMENT, 2006, Sustainability Charter for Victoria’s State forests, Victorian Government Department of Sustainability and Environment.
DEPARTMENT OF SUSTAINABILITY AND ENVIRONMENT, 2007a, Criteria and Indicators for Sustainable Forest Management in Victoria, Department of Sustainability and Environment.
http://www.dse.vic.gov.au/CA256F310024B628/0/44DBCB04FD0233DBCA257418007692B1/$File/EVC+nomenclature.pdf
DEPARTMENT OF THE ENVIRONMENT AND WATER RESOURCES, 2007, Australia’s Native Vegetation: A summary of Australia’s Major Vegetation Groups, 2007.
DI GREGORIO, A., and JANSEN, L., 2000, Land Cover Classification System (LCCS): Classification Concepts and User Manual [LCCS version 1.0].
ESCAVI, E. S. C. F. A. V. I., 2003, National Vegetation Information System, Version 6.0.
FAO FORESTRY, 2005, Global forest resources assessment update 2005. HOWELL, C., 2009. NATIONAL FOREST INVENTORY, 2003, Australia's state of the forests report 2003. NLWRA, 2001, (National Land and Water Resources Audit), Australian Native
Vegetation Assessment. OATES, A., and TARANTO, M., 2001, Vegetation Mapping of the Port Phillip &
Westernport Region, Arthur Rylah Institute for Environmental Research, Department of Natural Resources and Environment.
PARKS VICTORIA, 2007, State of the Parks 2007, Parks Victoria. ROGAN, J., MILLER, J., STOW, D., FRANKLIN, J., LEVIEN, L., and FISCHER, C.,
2003, Land-cover change monitoring with classification trees using Landsat TM and ancillary data. Photogrammetric Engineering and Remote Sensing, 69, 793-804.
SAURA, S., 2002, Effects of minimum mapping unit on land cover data spatial configuration and composition. International Journal of Remote Sensing, 23, 4853 - 4880.
SOKAL, R. R., 1974, Classification: Purposes, Principles, Progress, Prospects. Science, 185, 1115-1123.
22
THACKWAY, R., and ATYEO, C., 2006, Classifying Australian Land Cover. THACKWAY, R., LEE, A., DONOHUE, R., KEENAN, R. J., and WOOD, M., 2007,
Vegetation information for improved natural resource management in Australia. Landscape and Urban Planning, 79, 127-136.
WALKER, J., and HOPKINS, M. S., 1990, Vegetation, Second Edition edn (Melbourne: Inkata Press).
WILSON, A., and TALBOT, J., 2003, Australia's State of the Forests Report 2003, Bureau of Rural Sciences.
WULDER, M., and NELSON, T., 2003, EOSD Land Cover Classification Legend Report, Version 1, Canadian Forest Service.