Uncertainty in ground layer indicators - arises from combining field observations with drone-derived estimates, leading to inaccuracies in cover classification and spatial variation. This issue was amplified at Ambalindum, where litter data were estimated from photographs, causing discrepancies between drone imagery and benchmark values. Future work may benefit from direct observation of drone imagery.

Cryptogam cover - was excluded from the account due to the difficulty of developing benchmarks without large regional datasets. While field observations of cryptogams were collected, they were not included in the final condition score calculations. These observations have been preserved and could be used if benchmarks become available in the future.

Native tree size structure - was excluded from the account due to significant uncertainty. The method involved converting DBH-based benchmarks to tree crown characteristics using drone data. However, errors in tree segmentation and uncertainties in the benchmarks led to unsatisfactory results. As more accurate benchmarks and segmentation methods become available, this indicator may be revisited in future accounts.

Classification accuracy - at certain sites reduced the model's overall reliability, due to high-resolution data noise, blurred indicators, and canopy height model inaccuracies. For sites with less than 50% accuracy, a manual approach was used to calculate cover proportions. These sites were excluded from validation, preserving original data for future re-processing with improved classification methods.

Coarse woody debris - the computer vision model for segmenting and classifying coarse woody debris lacked sufficient accuracy. Instead, coarse woody debris was manually digitised using a line draw tool in Drone-Deploy within 0.1-hectare sub-plots at each site. The total line lengths were summed and multiplied by 10 to estimate debris length per hectare.