Maquis, which provides numerous ecosystem services and constitutes an integral part of the Mediterranean ecosystem, is highly heterogeneous. However, despite its importance and heterogeneity, maquis is generally mapped as a single class, while forests are mapped for management purposes. Detailed mapping of the maquis formations is necessary to understand their ecology and manage them sustainably. This study presents a method that generates alliance-level maps of the maquis ecosystems through satellite images using various machine learning techniques with different feature combinations and evaluates the proposed approach in the Mediterranean region of Southern Turkey, which has an area of 95,000 km2. Multi-temporal images extract information from vegetation phenology, while topographic and meteorological data are used to improve the classification. Cross-validation is performed using a ground-truth data set of approximately 7500 polygons. Results show that cost-effective and accurate maquis classification at the alliance level is possible using a combination of environmental features, multi-spectral, and multi-temporal satellite images. Adding environmental features to remotely sensed classification has improved the accuracy by 18%. The Random Forest (RF) algorithm improves classification accuracy by 7.3% and 14.6% relative to Support Vector Machine and Quadratic Discriminant Analysis algorithms, respectively. With the help of newly introduced features, we have succeeded in mapping 11 alliances with 64.2–82.7% overall accuracy. We believe the proposed classification approach will help improve the mapping of the shrubland ecosystems, which will significantly affect natural resource management, conservation, and adaptation to climate change.

Mapping Mediterranean maquis formations using Sentinel-2 time-series

Listiani, Indira Aprilia;
2022

Abstract

Maquis, which provides numerous ecosystem services and constitutes an integral part of the Mediterranean ecosystem, is highly heterogeneous. However, despite its importance and heterogeneity, maquis is generally mapped as a single class, while forests are mapped for management purposes. Detailed mapping of the maquis formations is necessary to understand their ecology and manage them sustainably. This study presents a method that generates alliance-level maps of the maquis ecosystems through satellite images using various machine learning techniques with different feature combinations and evaluates the proposed approach in the Mediterranean region of Southern Turkey, which has an area of 95,000 km2. Multi-temporal images extract information from vegetation phenology, while topographic and meteorological data are used to improve the classification. Cross-validation is performed using a ground-truth data set of approximately 7500 polygons. Results show that cost-effective and accurate maquis classification at the alliance level is possible using a combination of environmental features, multi-spectral, and multi-temporal satellite images. Adding environmental features to remotely sensed classification has improved the accuracy by 18%. The Random Forest (RF) algorithm improves classification accuracy by 7.3% and 14.6% relative to Support Vector Machine and Quadratic Discriminant Analysis algorithms, respectively. With the help of newly introduced features, we have succeeded in mapping 11 alliances with 64.2–82.7% overall accuracy. We believe the proposed classification approach will help improve the mapping of the shrubland ecosystems, which will significantly affect natural resource management, conservation, and adaptation to climate change.
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S1574954122002643-main.pdf

non disponibili

Licenza: NON PUBBLICO - Accesso privato/ristretto
Dimensione 8.49 MB
Formato Adobe PDF
8.49 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11582/334107
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact