Climate change is deeply altering the structure and composition of forest ecosystems, particularly in mountain landscapes. Ecological Niche Modelling (ENM) provides a valuable tool to estimate the potential distribution of species under current and future climate scenarios. Although many studies have assessed tree-species distributions at continental or national extents with coarse spatial resolution, few have addressed regional, high-resolution modelling – despite its clear relevance for local forest planning and conservation. Key challenges for fine-scale modelling include limited field data availability and high computational and methodological complexity. This thesis implements Ecological Niche Modelling at 50-m resolution to estimate the potential distribution of 23 dominant forest tree species in the Autonomous Province of Trento (Italian Alps). The modelling was performed using the R package SSDM which combines an ensemble of statistical and machine-learning algorithms to produce individual species models and community-level outputs via Stacked Species Distribution Modelling (SSDM). Models were calibrated using downscaled, high-resolution climate projections (ECLIPS2.0), terrain metrics from a Digital Terrain Model (DTM), and detailed occurrence data from local forest inventories, remote sensing, and a citizen-science platform. Model performance was evaluated using metrics including Area Under the receiver-operating-characteristic Curve (AUC), Cohen’s Kappa, sensitivity, and specificity. Species richness maps were produced using the SSDM framework, providing a replicable protocol for modelling both individual species distributions and community composition. Additionally, future projections for the 23 species under two IPCC emission scenarios (RCP4.5 and RCP8.5) were generated to assess potential vegetation shifts across three time periods (2041–2060, 2061–2080, and 2081–2100). The analysis also included the assessment of feature importance and modelling uncertainty, which improves the interpretability and robustness of the projections. Species were classified according to projected changes in their distributional ranges. Taxa such as Fagus sylvatica L., Fraxinus ornus L. and Quercus ilex L. are predicted to exhibit range expansion (“gainers”), whereas Picea abies (L.) H.Karst., Pinus cembra L. and Fraxinus excelsior L. are anticipated to contract (“losers”). Other taxa, including Betula pendula Roth, Larix decidua Mill., and Populus tremula L., display non-linear temporal trend, with initial gains followed by subsequent losses of suitable areas. All the studied species will experience an upward elevation shift of their climate optimum. 6 Species richness is projected to remain largely stable under the RCP4.5 scenario, whereas a decline is expected under RCP8.5. Results can support adaptive forest planning and biodiversity conservation in the Alpine contexts. The approach adopted in this study contributes to bridging the gap between broad-scale modelling and local forest management needs. The work provides spatial insights into biodiversity patterns and could help to identify areas of high ecological significance or conservation concern.
Il cambiamento climatico sta modificando sostanzialmente la struttura e la composizione degli ecosistemi forestali, con effetti particolarmente rilevanti nelle aree montane. La modellizzazione della nicchia ecologica (Ecological Niche Modelling, ENM) rappresenta uno strumento utile per stimare la distribuzione potenziale delle specie in scenari climatici presenti e futuri. Sebbene numerosi studi abbiano analizzato la distribuzione delle specie arboree su scale continentali o nazionali utilizzando dati a bassa risoluzione spaziale, poche ricerche hanno affrontato la modellizzazione regionale ad alta risoluzione, nonostante la sua evidente rilevanza per la pianificazione forestale locale e le politiche di conservazione. Le principali sfide per la modellizzazione ad alta risoluzione comprendono la limitata disponibilità di dati di campo e l’elevata complessità computazionale e metodologica. La presente tesi applica l’ENM a una risoluzione di 50 m per stimare la distribuzione potenziale di 23 specie arboree dominanti nella Provincia Autonoma di Trento (Alpi italiane). La modellizzazione è stata eseguita con il pacchetto R SSDM, che integra un insieme ensemble di algoritmi statistici e di machine learning per produrre modelli di specie individuali e output a livello di comunità mediante l’approccio di Stacked Species Distribution Modelling (SSDM). I modelli sono stati calibrati utilizzando proiezioni climatiche ottenute tramite downscaling ad alta risoluzione (ECLIPS2.0), metriche topografiche derivate dal Modello Digitale del Terreno (DTM) e dati di presenza dettagliati provenienti da inventari forestali locali, telerilevamento e una piattaforma di citizen science. La validazione dei modelli è stata effettuata mediante l’uso di diverse metriche di valutazione, tra cui l’area sotto la curva ROC (AUC), il coefficiente Kappa di Cohen, la sensibilità e la specificità. Sono state inoltre prodotte mappe di ricchezza specifica tramite l’approccio SSDM, fornendo un protocollo replicabile per la modellizzazione sia delle singole specie sia della composizione comunitaria. Le proiezioni future sono state generate per tre orizzonti temporali (2041–2060, 2061–2080, 2081–2100) e per due scenari di emissione dell’IPCC (RCP4.5 e RCP8.5). L’analisi include infine una valutazione dell’incertezza modellistica, volta a migliorare l’interpretabilità e la robustezza delle proiezioni. Le specie sono state classificate in base alle proiezioni di variazione dei rispettivi areali di distribuzione. Specie come Fagus sylvatica L., Fraxinus ornus L. e Quercus ilex L. sono 8 previste in espansione del loro areale («gainers»), mentre Picea abies (L.) H.Karst., Pinus cembra L. e Fraxinus excelsior L. sono attese in contrazione («losers»). Altre specie, tra cui Betula pendula Roth, Larix decidua Mill. e Populus tremula L., mostrano andamenti temporali non lineari, con guadagni iniziali seguiti da successive perdite dell’areale ottimale. La ricchezza specifica è prevista rimanere sostanzialmente stabile nello scenario RCP4.5, mentre ci si aspetta una diminuzione nello scenario RCP8.5. I risultati possono supportare la pianificazione forestale adattiva e la conservazione della biodiversità in contesti alpini. L’approccio adottato contribuisce a colmare il divario tra modellizzazioni su larga scala e le esigenze gestionali locali, offrendo informazioni spaziali dettagliate sui pattern di biodiversità e strumenti utili per l’individuazione di aree di elevato valore ecologico o di interesse conservazionistico.
Ecological Niche Modelling of Forest Tree Species in the Alpine Space: a Stacked SDM Approach at Regional Scale
OBEROSLER, DAMIANO
2024/2025
Abstract
Climate change is deeply altering the structure and composition of forest ecosystems, particularly in mountain landscapes. Ecological Niche Modelling (ENM) provides a valuable tool to estimate the potential distribution of species under current and future climate scenarios. Although many studies have assessed tree-species distributions at continental or national extents with coarse spatial resolution, few have addressed regional, high-resolution modelling – despite its clear relevance for local forest planning and conservation. Key challenges for fine-scale modelling include limited field data availability and high computational and methodological complexity. This thesis implements Ecological Niche Modelling at 50-m resolution to estimate the potential distribution of 23 dominant forest tree species in the Autonomous Province of Trento (Italian Alps). The modelling was performed using the R package SSDM which combines an ensemble of statistical and machine-learning algorithms to produce individual species models and community-level outputs via Stacked Species Distribution Modelling (SSDM). Models were calibrated using downscaled, high-resolution climate projections (ECLIPS2.0), terrain metrics from a Digital Terrain Model (DTM), and detailed occurrence data from local forest inventories, remote sensing, and a citizen-science platform. Model performance was evaluated using metrics including Area Under the receiver-operating-characteristic Curve (AUC), Cohen’s Kappa, sensitivity, and specificity. Species richness maps were produced using the SSDM framework, providing a replicable protocol for modelling both individual species distributions and community composition. Additionally, future projections for the 23 species under two IPCC emission scenarios (RCP4.5 and RCP8.5) were generated to assess potential vegetation shifts across three time periods (2041–2060, 2061–2080, and 2081–2100). The analysis also included the assessment of feature importance and modelling uncertainty, which improves the interpretability and robustness of the projections. Species were classified according to projected changes in their distributional ranges. Taxa such as Fagus sylvatica L., Fraxinus ornus L. and Quercus ilex L. are predicted to exhibit range expansion (“gainers”), whereas Picea abies (L.) H.Karst., Pinus cembra L. and Fraxinus excelsior L. are anticipated to contract (“losers”). Other taxa, including Betula pendula Roth, Larix decidua Mill., and Populus tremula L., display non-linear temporal trend, with initial gains followed by subsequent losses of suitable areas. All the studied species will experience an upward elevation shift of their climate optimum. 6 Species richness is projected to remain largely stable under the RCP4.5 scenario, whereas a decline is expected under RCP8.5. Results can support adaptive forest planning and biodiversity conservation in the Alpine contexts. The approach adopted in this study contributes to bridging the gap between broad-scale modelling and local forest management needs. The work provides spatial insights into biodiversity patterns and could help to identify areas of high ecological significance or conservation concern.| File | Dimensione | Formato | |
|---|---|---|---|
|
Oberosler_Damiano.pdf
accesso aperto
Dimensione
13.9 MB
Formato
Adobe PDF
|
13.9 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/91305