Title

Rainforest Mapping for state-wide Victoria 

Alternate title

RAINFOR 

Purpose

All rainforest communities must be protected from threatening processes such as timber harvesting and fire management activities. Forest Planners across Government and non-Government organisations are encouraged to utilise this new data as part of their risk management plans to protect rainforest communities. For areas of modelled rainforest that have been mapped outside of the formal Comprehensive, Adequate and Representative (CAR) reserve system; forest management planners will need to confirm in the field the presence and extent of the rainforest community prior to any forest or fire management activity commencing. The data provided within RAINFOR is a modelled extent of rainforest based upon the spectral characteristics extractable from the Sentinel satellite imagery. The data has been Quality Assured through a supervised assessment of the classification result to ensure that communities have identified and mapped accurately 

Supplemental Information

Current Design Issues: The layer identifies five classes of rainforest community: 1. Cool Temperate Rainforest 2. Dry Rainforest 3. Gallery Rainforest 4. Littoral Rainforest 5. Warm Termperate Rainforest Related Documents: None 

Status

Completed  

Code

4283 

Maintenance and update frequency

Not planned  

Topic category

• Biota

Use limitation

. 

Classification

Unclassified  

Statement

Dataset Source: Sentinel imagery Dataset Originality: Primary 

Description

Collection Method: Model derived from Satellite imagery 

Description

TBA 

Classification

Unclassified  

Attribute Quality

Comments

In this project the modelling has essentially been used to highlight and identify places that may support rainforest and warrant further checking or `truthing' using high resolution aerial photography. As such, the success or otherwise of the modelling is partially qualitative as it relates to `efficient searching' and spatial precision in a context of and elevated sensitivity to false absence or missing extant rainforest stands. In general, the model over predicted the extent of rainforest, misclassifying significant areas of various vegetation types including but not exclusive to, non-Eucalypt dominated vegetation, particularly mesophyll shrublands, urban gardens and exotic broadleaf trees in rural areas, Acacia stands (particularly those dominated by Blackthorn Acacia silvestris), tree-fern dominated gullies and myrtaceous shrublands associated with streams and swamps. These were removed through the hand editing process, but significant regions of these types persist in the modelled rainforest extent in the Strzelecki Ranges. 

Positional Accuracy

Comments

10m 

Conceptual Consistency

Comments

Not Applicable 

Missing Data

Comments

State-wide Victoria Completeness Verification: The extent of rainforest in the Strzelecki Ranges is the only area of the state that hasnt been verified. No field validation of this dataset has been undertaken to date. 

Excess Data

Comments

The Sentinel rainforest products are based on spectral supervised classifications. The training data for rainforest presence were obtained using Aerial Photographic Interpretation (API) with locations selected using contemporary imagery, informed by previous mapping. Rainforest absence data were derived from both the allocation of random or background absence and the deliberate placement of a significant number (around 10,000 in Eastern Victoria and 3000 in the Otway Ranges) of absences in a range of forest types including plantations and regrowth and many within 50 m of the rainforest training polygons. The latter were included accurately render the rainforest, sclerophyll forest boundary. In all, approximately 247,000 training sites were supplied to the eastern Victorian rainforest model comprising approximately 55,000 rainforest sites. The Otway Ranges model was informed by approximately 17,000 training sites, 2,300 of which were Rainforest sites. These training data became `targets' in the respective models.