From field observations to AI-based phenology: an overview of advances in monitoring ecosystem responses to climate change

Ongoing climate change, together with land-use transformation, biological invasions, and increasing human pressure, is intensifying alterations in ecosystem dynamics and phenological patterns.  Plant phenology has gained a central role as a key biological indicator of ecosystem responses to climate variability, since long-term shifts in phenological timing influence ecosystem functioning, productivity, and trophic interactions. Historically, phenological monitoring has relied on labor-intensive field observations, often constrained by limited spatial coverage and temporal continuity. In recent years, advances in near-surface sensing technologies—including unmanned aerial vehicles, phenocams, and spectral reflectance sensors—together with satellite remote sensing have substantially transformed phenological studies. These new applications are particularly important for monitoring vulnerable hotspots, such ecosystems in Mediterranean regions, whose complex functioning result from the interaction of climatic gradients, geomorphological processes, disturbance regimes, and species-specific functional traits. Moreover, recently, new AI-based approaches have gained importance as powerful framework for continuous, multi-scale phenological monitoring.

This contribution is aimed to provide an overview of major methods and techniques adopted for phenological research supported by artificial intelligence (AI). Among the most widespread applications, developments in machine learning (ML) and deep learning (DL) enable the automated analysis of large and complex image datasets, facilitating the extraction of robust phenological signals and supporting the development of standardized monitoring frameworks for assessing ecosystem responses to climate change. Also, Computer Vision techniques applied to phenocam imagery exemplify these advances: deep learning models, particularly convolutional neural networks, can automatically classify phenological stages from high-frequency image data. Extracted visual features can be further analyzed using temporal modelling approaches, such as neural networks, temporal convolutional networks, or Transformer-based architectures, to characterize seasonal dynamics. In addition, chromatic indices derived from vegetation pixels can be combined with AI-based correction methods to reduce the effects of illumination variability, weather conditions, and sensor-specific biases, improving the reliability of phenological metrics. Despite these advances, challenges remain, including limited training data availability, spectral similarity among plant species, and strong local environmental heterogeneity, which require context-specific calibration.

Overall, the integration of satellite observations, phenocam networks, and AI-driven tools offers a powerful framework for continuous, multi-scale phenological monitoring. As climate change accelerates shifts in plant development and ecosystem seasonality, such integrated approaches will be essential for improving ecological forecasting and supporting resilient biodiversity monitoring strategies.