Predicting the ecological condition of grazed Australian landscapes using satellite-derived indices, patch metrics, and passive acoustic monitoring of birds

Livestock grazing is a dominant land use in Australia, with farmers managing large areas of intact and modified ecosystems. Effective monitoring and management are essential to maintain the ecological condition and biodiversity of these grazed landscapes under land use pressures. However, measuring biodiversity at scale is expensive and time-consuming, which makes automating the survey process highly desirable. Automated remote technologies can augment traditional observer-based biodiversity survey approaches without compromising quality. While satellite-derived remote sensing can provide vegetation metrics, other solutions such as passive acoustic monitoring (PAM) may be more responsive to fine-scale ecological variability. In this study we used a combination of remote sensing tools - PAM, a multi-species bird detection convolutional neural network, satellite vegetation indices, and landscape metrics - in an elastic net regression model to accurately predict farm-scale ecological condition across multiple vegetation types in Queensland, Australia. Our results reveal that a robust, remote assessment of ecological condition and biodiversity can be achieved using a short two-week sampling period. This approach provides landowners with an inexpensive on-farm biodiversity assessment tool for measuring ecological condition at multiple scales.