Woody Age Since Disturbance - Sentinel-2, DETSI Algorithm, QLD Coverage 

The woody vegetation extent for Queensland is attributed with an estimated age in years since the last significant disturbance. The method uses a sequential Conditional Random Fields classifier applied to Landsat time series starting 1988 to predict woody cover over the time period. A set of heuristic rules is used to detect and track regrowing woody vegetation in the time series of woody probabilities and record the approximate start and end dates of the most recent regrowth event. Regrowth detection is combined with the Statewide Land and Trees Study (SLATS) Landsat historic clearing data to provide a preliminary estimate of age since disturbance for each woody pixel in the woody extent. The 'last disturbance' may be due to a clearing event or other disturbance such as fire, flood, drought-related death etc. Note that not all recorded disturbances may result in complete loss of woody vegetation, so the estimated age since disturbance does not always represent the age of the ecosystem. The age since disturbance product is derived from multiple satellite image sources and derived products which represent different scales and resolutions: Landsat (30 m), Sentinel-2 (10 m) and Earth-i (1 m). 

We at TERN acknowledge the Traditional Owners and Custodians throughout Australia, New Zealand and all nations. We honour their profound connections to land, water, biodiversity and culture and pay our respects to their Elders past, present and emerging. 

To provide an estimation of woody vegetation age for ongoing monitoring and reporting. 

The long time series of Landsat imagery was used as the primary image data to derive woody vegetation age estimates. Landsat imagery from Landsat-7 and -8 is downloaded from the United States Geological Survey (USGS website) (earthexplorer.usgs.gov) and processed to surface reflectance as described in Flood (2013a). All Landsat imagery is geometrically corrected by the USGS. Cloud and cloud shadow masks are computed using the Fmask methods of Zhu and Woodcock (2012). Three monthly seasonal fractional cover composites are produced for Queensland from surface reflectance using the medoid method described in Flood (2013b).

Thirty-three years (1988 to 2021) of seasonal Landsat fractional cover components (Scarth, 2008) were used in the woody likelihood modelling and regrowth event detection. The fractional cover data consisted of bare, green, and non-green sub-pixel components.

Additional inputs include the historic Landsat clearing data (1988-2018) and the 2018 woody extent baseline data set.

The 2019 woody extent data set was used as sample strata to generate woody/non-woody training sequences for a Conditional Random Field (CRF) classifier. The 2019 woody extent data were resampled from 10 m to 30 m spatial resolution to align spatially with the Landsat time series. A temporal augmentation method of sampling was used to generate sequences of woody training labels. For each tile in a regular grid of 150X150 km tiles across Queensland, a model was fit between the fractional cover data and the training data sequences. The CRF model produces a sequence of annual woody probabilities (0 to 1). This release involved a refit of the CRF model as the previously published 2018 Age Since Disturbed data was based on the 2017 woody extent and fractional cover time series. A simple set of heuristic rules are then applied to detect patterns in the woody probability sequence that might characterise a typical woody regrowth response curve and track that regrowth event over time. Regrowth detection is combined with the SLATS Landsat clearing record to provide a preliminary estimation of age since last disturbance for all woody pixels recorded in the 2018 baseline woody extent.

Each data file includes data from the beginning of the data collection (1988) to the year when the data were processed and published in the Statewide Landcover and Trees Study (SLATS) report.