Annual woody vegetation and canopy cover grids for Tasmania

This collection provides annual woody vegetation (> 10 % canopy cover, > 2 m height) and canopy cover (0 – 100%) grids for Tasmania with a spatial resolution of 10 m. This dataset was developed to improve the availability of information suitable for farm-scale analyses of tree cover using publicly available, non-commercial remote sensing data. It enables fine-scale analyses of woody vegetation and canopy cover trends in natural and modified ecosystems across Tasmania between 2017 and 2024.\n\nNote: In response to request, canopy cover grids were developed for Riveaux Road using imagery acquired between November 2018 and mid-January 2019. This enables comparisons of canopy cover before and after the Riveaux Road fire.\n\nLineage: All modelling was performed in Google Earth Engine using the random forest algorithm in classification and regression mode for woody vegetation and canopy cover, respectively. The model was trained on 44,009 points derived from airborne lidar data acquired between 24/Jan/2019 and 20/Apr/2019 across Tasmania. Woody vegetation and canopy cover models used Sentinel 1 Synthetic Aperture Radar (SAR) and Sentinel 2 MultiSpectral Instrument (MSI) imagery and derived features (e.g., vegetation indices, temporal variability, spatial texture) as model covariates. All imagery was acquired during late summer (01/Jan/2019 to 31/Mar/2019) to enhance the separability of trees from crops and grasses. This same period (01/Jan to 31/Mar) is used for modelling all subsequent years.\n\nIndependent validation on 18,867 points yielded the following results:\nWoody vegetation: Overall accuracy = 0.94, Kappa = 0.87, Sensitivity = 0.94, Specificity = 0.94.\nCanopy cover: R² = 0.83, Lin's Concordance Coefficient (CCC) = 0.90, MAE = 0.09, Bias = 0.00, Mean = 0.27. \n\nNote that in a small number of cases, valid SAR pixels were not available (e.g., regions of active shadow and layover) and therefore a backup algorithm was used to fill no data gaps where possible. This backup algorithm uses only Sentinel 2 MSI imagery. The specific model used for each pixel is given in the 'source' image folder provided in this data collection (S1S2 = Sentinel 1 SAR and Sentinel 2 MSI; S2 = Sentinel 2 MSI only). \n\nThe backup algorithm yielded the following results during independent validation:\nWoody vegetation: Overall accuracy = 0.93, Kappa = 0.86, Sensitivity = 0.93, Specificity = 0.93.\nCanopy cover: R² = 0.81, Lin's Concordance Coefficient (CCC) = 0.90, MAE = 0.09, Bias = 0.00, Mean = 0.27.