Scaling biodiversity recovery monitoring: linking satellite-derived structure with multi-taxa acoustic and camera-trap data to inform robust indicators for Amazonian forest restoration

Global commitments under the Kunming-Montreal Global Biodiversity Framework demand scalable, robust methods for monitoring ecosystem restoration. Remote sensing (RS) technologies offer unprecedented spatial and temporal coverage, yet their capacity to track the complex dynamics of faunal reassembly remains a critical uncertainty. Tropical forest restoration presents a particularly urgent testbed: recovery trajectories vary widely among taxa, traits and landscapes, and reliable metrics must discriminate true ecological recovery from transient colonisation. Relying solely on RS-derived vegetation structure as a proxy for biodiversity risks overestimating ecological recovery if structural gains outpace the return of sensitive animal communities. Accurately calibrating RS signals with on-the-ground biodiversity data is therefore essential for developing effective monitoring frameworks.

We address this challenge through an integrated, multi-sensor monitoring approach across three clusters of large-scale restoration projects in Pará, Brazilian Amazon. We combine multi-scale RS data (LiDAR & RGB) with extensive in-situ data from high-throughput passive sensors. Acoustic monitors and camera traps were deployed at over 160 sampling points in a repeated-measures design (i.e., time-series), benchmarking restoration trajectories against old-growth forest (positive) and degraded pasture (negative) controls. To achieve scalability, we utilise cutting-edge Artificial Intelligence (AI) pipelines (e.g., BirdNET, SpeciesNET) for automated classification of birds and mammals, implementing rigorous validation protocols.

Our analysis evaluates the congruence between remotely sensed structural metrics and ground-based faunal recovery indicators, including community composition,  functional diversity, and phylogenetic diversity. Demonstrating the sensitivity of this approach, our findings show that integrated acoustic and camera trap data are sensitive to early ecological changes, detecting measurable shifts in community composition towards reference conditions within the first years of restoration. We explore emerging patterns using hierarchical trajectory models and machine learning (random forest), uncovering how landscape context and RS-derived structural complexity predict multi-taxa recovery rates.

This research uncovers mechanisms of succession in early stages of recovery, while also demonstrating a scalable monitoring system that leverages the convergence of AI, in-situ sensors, and RS. By linking ground-truthed biodiversity data with advanced RS platforms, we provide an essential blueprint for aligning remote sensing technology with the ecological realities of biodiversity recovery and tracking progress towards global targets.