Francesco Valerio

The transferability of single or joint species distribution models ((j)SDMs) depends on their ability to predict beyond the observed environmental range and to remain consistent despite shifts in biotic interactions. Transfer accuracy may be improved by recent advances in the application of deep learning that provide greater flexibility and potentially superior predictive accuracy than traditional approaches. We implemented jSDMs with deep and machine learning algorithms and measured the transfer accuracy from continental to regional areas in communities with different species composition. We ran jSDMs with deep neural networks (DNN), elastic net (EN), and stacked SDMs (sSDM) with random forests (RF). We used 134 689 occurrence records representing 1776 species of six taxonomic groups (beetles, birds, bryophytes, fungi, lichens and plants) from 2387 forest plots in Europe. We employed an agnostic modelling approach that covered most of the environmental conditions by including more than 100 satellite‐derived variables and 98 climatic variables. The predictive power of the models within the training continental area was evaluated using AUC, whereas the transfer accuracy in the regional area was evaluated with the Boyce index calculated with independent presence records. We found that the DNN–jSDMs outperformed other models at continental scale, but model transfer from continental to regional extent was less accurate. We found that the accuracy of regional predictions was higher for taxonomic groups with better representation in the continental data, such as birds, bryophytes and plants. Depending on the algorithm and the taxonomic group, we achieved acceptable (Boyce > 0) to accurate (Boyce > 0.5) transferability for 32–78% of the species. Our findings underscored the need of considering trade‐offs among hyperparameter tuning, spatial scales and model complexity. Our findings also suggest that the varying biotic interaction structures and, particularly, the different species compositions of the transfer areas, may affect model transferability more than previously considered.

Accurate geolocation is essential for reliable use of GEDI (Global Ecosystem Dynamics Investigation) LiDAR data in footprint-scale applications such as biomass modeling, data fusion, and ecosystem monitoring. Residual geolocation errors from systematic biases and random ISS-induced jitter can degrade vegetation and terrain metrics accuracy. This study presents GEDICorrect, an open-source Python framework for correcting GEDI geolocation at orbit, beam, and footprint levels using multiple approaches, including waveform matching, terrain matching, and relative height (RH) profile matching. The framework integrates key components of gediSimulator (gediRat and gediMetrics) and extends them with flexible correction logic, multiple similarity metrics, footprint clustering, and parallelized processing. We applied GEDICorrect to a heterogeneous Mediterranean woodland in Portugal to assess accuracy and computational performance. Using Kullback–Leibler divergence for waveform similarity, GEDICorrect improved canopy height (RH95) accuracy from R2 = 0.61 (uncorrected) to 0.74 with orbit-level correction and to 0.78 with footprint-level correction. RMSE decreased from 2.62 m (rRMSE = 43.13%) to 2.12 m (34.97%) and 2.01 m (33.05%), respectively. Terrain elevation RMSE was reduced by 0.34 m relative to uncorrected data and by 0.37 m compared to the gediSimulator baseline. GEDICorrect also enabled evaluation of pointing accuracy improvements implemented since July 2024, providing timely insights for the growing GEDI user community. Computationally, it achieved ~2.4x speedup over gediSimulator in single-process mode and scaled to 24 cores, reducing runtime to ~4.3 h (~19.5x improvement) while delivering multilevel offset corrections. GEDICorrect provides a robust, scalable framework for improving GEDI geolocation accuracy enabling functionalities not available in existing tools.

Extensive farmland in Europe hosts high biodiversity levels but is threatened by land use changes associated with market and agricultural policy drivers. We show that a nationwide replacement of rainfed cereals in favor of beef production in Portugal has coincided with a nearly 80% decline in the population of a ground‐nesting raptor, the Montagu's harrier Circus pygargus , since 2002. Reduced cereal crop cover led to nesting attempts occurring mostly in fodder crops, the predominant habitat with suitable sward structure at the onset of breeding. This results in a twofold ecological trap: early hay harvesting destroys first nesting attempts, whereas late cereal harvesting destroys replacement clutches. This double phenological mismatch prevents any breeding success and likely explains the country's observed population decline of Montagu's harriers and other key farmland birds. Reforming current agroenvironmental schemes to promote practices compatible with wild population phenology is paramount to revert these changes.

Inland freshwater resources in semiarid environments play a key role in maintaining ecological systems and supporting human development. Space-based remote sensing spatiotemporal data have emerged as a new paradigm for understanding ecohydrological processes and trends, particularly in water-stressed areas. However, comprehensive cataloging is still lacking, especially in semi-arid regions and for small-sized water bodies (i.e., ponds), which are often overlooked despite their ecological relevance. In this study, high-resolution optical and radar Sentinel data (Sentinel-1 and Sentinel-2) were used to construct Sentinel-1&2-based local surface water (SLSW) models, to infer surface water occurrence and extent. To assess the reliability of this model, the results were compared with verification data, and separately with Landsat-based global water (LGSW) models. Three distinct semiarid regions were selected in SW Iberia, within a Mediterranean climate, each encompassing special protection areas for conservation and subjected to marked seasonality and bioclimatic changes. Surface water attributes were modeled using Random Forests for SLSW time series forecasting, which included the period from January 1, 2020, to December 31, 2021. During this period, the completeness of the archived information was compared between SLSW and LGSW, considering both intra-annual and inter-annual variations. The predictive performance of these models was then compared for specific periods (dry and wet), and each was independently validated with verification data. The results showed that SLWM achieved satisfactory predictive performances in detecting surface water occurrence ( μ ≈72%), with far greater completeness and reconstructed seasonality patterns compared to LGSW. The relatedness between SLSW and LGSW was stronger during wet periods (R ² =0.38) than dry periods (R ² =0.05), and SLSW related much better with the verification data (R ² =0.66) than when compared to LGSW (R ² =0.24). The proposed SLSW approach may therefore provide advantages in the delineation of dynamic surface water characteristics (occurrence and extent) in very small-sized water bodies (i.e., <0.5 ha), allowing for uninterrupted surface water time series forecasting at high spatiotemporal detail, and over extensive areas. Given the water constraints in semiarid regions and water resources vulnerability to climate change, our results show high potential for supporting a variety of activities underlying rural development and biodiversity conservation. Additionally, the socio-ecological applications of this research may help identify surface water anomalies (e.g., drought events) and enhance sustainable water supply governance, a particular priority in climate change hotspots. Highlights: Surface water occurrence and extent was modeled across three semiarid regions Sentinel-1&2 data was compared with Landsat for characterizing very small water bodies Models based on Sentinel-1&2 resulted in a satisfactory classification precision Very high series completeness across seasons was found using Sentinel-1&2 data Sentinel-1&2 data was more reliable than Landsat when compared to verification data