A machine learning approach to the assessment of the vulnerability of Posidonia oceanica meadows

Catucci, Elena; Scardi, Michele

In this study, we adopted a modelling approach to assess the vulnerability of Posidonia oceanica meadows, the most widespread seagrass in Mediterranean Sea. P. oceanica has a crucial ecological role all over the basin. In fact, this seagrass is a habitat-forming species that can extend from the surface to 45 m depth, forming meadows. These meadows rank among the most valuable ecosystem in the Mediterranean Sea, in term of the services they provide. However, in areas where alterations of environmental conditions happened, regression of the meadows may occur. Despite it is one of the main targets of conservation actions all over the basin, P. oceanica is declining at alarming rate, especially due to the anthropogenic impacts. Thereby, there is a urgent need to study the effects of environmental factors that could affect its ecological status. We used a Random Forest for developing a Habitat Suitability Model (HSM) for P. oceanica in the Italian seas. The use of HSMs has been especially promoted to support ecosystem assessment and conservation planning, since they allow to better understand both the habitat requirements and the potential distribution of species. Since the spatial distribution of meadows in Italian seas is already known, we used the HSM predictions to evaluate the suitability of the habitat for P. oceanica at large spatial scale and, consequently, we assessed the vulnerability of the meadows. Particularly, our occurrence data included both areas were P. oceanica was known as living and regressed meadows. After the RF training, we validated the model using an independent test set and we evaluated the performance using both ROC curve and K statistic. The results showed that the HSM presented a quite good level of accuracy. Thus, we carried out a spatial analysis of the HSM predictions in relation to the actual ecological status of P. oceanica. The results showed that in areas where living meadows actually occurred, high habitat suitability predictions were significantly more frequent. On the contrary, where regressed meadows were actually observed, predictions indicated low habitat suitability for P. oceanica. This study stressed that modeling can effectively support the assessment of ecosystem status as well as conservation actions.

Zitieren

Zitierform:

Catucci, Elena / Scardi, Michele: A machine learning approach to the assessment of the vulnerability of Posidonia oceanica meadows. Jena 2018.

Zugriffsstatistik

Gesamt:
Volltextzugriffe:
Metadatenansicht:
12 Monate:
Volltextzugriffe:
Metadatenansicht:

Grafik öffnen

Rechte

Nutzung und Vervielfältigung:

Export