Using digital photography to monitor changes in biocrusts and ground cover in a savanna rangeland

Biocrusts form a living soil cover in Australia's northern savannas, delivering essential ecosystem services. More accessible tools are needed to quantify and monitor ground cover, including biocrusts, as current methodologies are time-consuming, expensive, or specialised. At Victoria River Research Station (Northern Territory, Australia), long-term fire research plots were used to monitor the response of low vegetative ground and soil covers for different burning intervals and seasons. Mobile phone photographs were analysed using machine-learning software and a derived decision tree-based segmentation model (DTSM). The resulting data were compared to visual in-field assessment by trained researchers. Visual assessments and photographs were taken at two time points during the post-fire recovery period, mid-wet and dry seasons, at three burning intervals (2, 4, and 6 years) and for two different burning times, early or late dry season. DTSM-derived grass and litter cover were statistically similar to field observations in the burnt and unburnt plots. Biocrust cover derived from DTSM also matched field observations in fire treatments and unburnt control plots in the dry season, except when obscured by grass or litter. In the wet season, DTSM underestimated biocrust cover in some treatments, and DTSM did not detect biocrust obscured under dense grass cover. Nevertheless, biocrust pigment analysis confirmed a significant presence of biocrusts both on seemingly bare soil and under the grass canopy. We concluded that mobile phone photographs are suitable for monitoring dry-season ground cover. When similar colours of grass and litter cover were combined, the modelled accuracy reached 95-97%. With some refinements, DTSM analysis of photographs could accurately quantify the impact of fire disturbance on biocrusts and grass cover. However, it would be advantageous to improve the model by additional field records to determine how much biocrust occurs under the grass. This study provides land managers with an efficient method of recording ground cover over time to aid land-condition assessments.

Automated summary

Biocrusts are important for ecosystem services in Australia's northern savannas, but current monitoring methods are time-consuming and expensive. This study used machine learning software and a segmentation model to analyze mobile phone photographs and compared the results to field observations. The study found that mobile phone photographs can effectively monitor dry-season ground cover, providing an efficient method for land managers to track ground cover over time.