Does GEOBIA enhance vegetation classificationperformance in heterogeneous and dynamic landscapes?
DOI:
https://doi.org/10.17398/3101-7177.2.279Palabras clave:
Landsat, Continuous Change Detection, Superpixels SegmentationResumen
Mapping vegetation types in heterogeneous and dynamic landscapes using long satellite image time series remains challenging due to uncertainties associated with spectral similarity among classes, phenological variability, and sensor differences. This study evaluates different classification approaches using Landsat imagery (1984-2024) processed in Google Earth Engine to map landscape dynamics in Asturias (NW Spain), a region characterized by steep terrain and high environmental heterogeneity. Image preprocessing included cloud masking, topographic correction, and biseasonal compositing to capture phenological differences. Four analysis years were selected (1984, 1998, 2008, and 2023), and 21 predictor variables were compiled, including 12 spectral bands, six texture metrics, and three biophysical traits. To perform the classifications, temporally invariant training polygons were defined using a change detection algorithm, allowing their reuse across all analysis years. Two Random Forest modelling approaches were implemented, pixel-based and object-based, exploring the performance of the three predictor groups both independently and additively. Objects were defined using object-based image analysis techniques (GEOBIA), grouping spectrally and spatially homogeneous pixels. Pixel-based models using spectral bands showed the highest performance (OA ≈ 0.88) and low interannual variability, indicating that invariant polygons enable the use of a single training dataset without loss of performance and with reduced data preparation effort. Although GEOBIA did not improve overall accuracy, it did enhance the classification of fast-growing forest plantations (FSCORE +0.13), likely due to their monospecific composition and well-defined spatial boundaries. We conclude that GEOBIA approaches may be more suitable for clearly
delineated vegetation types, whereas pixel-based classifications may be more appropriate for monitoring heterogeneous landscapes.
Descargas
Referencias
Achanta, R., & Susstrunk, S. (2017). Superpixels and Polygons Using Simple Non-iterative Clustering. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 4895–4904. https://doi.org/10.1109/CVPR.2017.520
Arévalo, P., Bullock, E. L., Woodcock, C. E., & Olofsson, P. (2020). A Suite of Tools for Continuous Land Change Monitoring in Google Earth Engine. Frontiers in Climate, 2. https://doi.org/10.3389/fclim.2020.576740
Fernández-Guisuraga, J. M., González-Pérez, I., Reguero-Vaquero, A., & Marcos, E. (2024). Estimating Grassland Biophysical Parameters in the Cantabrian Mountains Using Radiative Transfer Models in Combination with Multiple Endmember Spectral Mixture Analysis. Remote Sensing, 16(23), 4547. https://doi.org/10.3390/rs16234547
López Trullén, D., Álvarez-Martínez, J. M., Sánchez Labrador, J. D., Jiménez-Alfaro, B., Pérez-Silos, I., Hernández-Romero, G., & Barquín, J. (2022). Espectrofenología con datos Sentinel 2: definición de curvas de referencia para la caracterización de ecosistemas forestales. Ecosistemas, 2411. https://doi.org/10.7818/ECOS.2411
Martin-Gallego, P., Marston, C. G., Altamirano, A., Pauchard, A., & Aplin, P. (2024). Mapping alien and native forest dynamics in Chile using Earth observation time series analysis. Forest Ecology and Management, 560, 121847. https://doi.org/10.1016/j.foreco.2024.121847
Pfitzer-López, D., Ramírez-Rodríguez, R., Roces-Díaz, J. V., Álvarez-Martínez, J. M., Quevedo, M., García, D., Jiménez-Alfaro, B., & Suárez-Seoane, S. (2026). A new index for monitoring non-linear trends in passive forest recovery at regional scale. Ecological Indicators, 182, 114532. https://doi.org/10.1016/j.ecolind.2025.114532
Verhulst, M., Heremans, S., Blaschko, M. B., & Somers, B. (2024). Temporal Transferability of Tree Species Classification in Temperate Forests with Sentinel-2 Time Series. Remote Sensing, 16(14), 2653. https://doi.org/10.3390/rs16142653
Yang, L., Wang, L., Abubakar, G. A., & Huang, J. (2021). High-Resolution Rice Mapping Based on SNIC Segmentation and Multi-Source Remote Sensing Images. Remote Sensing, 13(6), 1148. https://doi.org/10.3390/rs13061148
Zhu, Z., & Woodcock, C. E. (2014). Continuous change detection and classification of land cover using all available Landsat data. Remote Sensing of Environment, 144, 152–171. https://doi.org/10.1016/j.rse.2014.01.011
Descargas
Publicado
Número
Sección
Licencia
Derechos de autor 2026 Daniel Pfitzer-López, José V. Roces-Díaz, José M. Fernández-Guisuraga, Lucía García-Candanedo, Susana Suárez-Seoane (Autor/a)
Esta obra está bajo una licencia internacional Creative Commons Atribución-NoComercial-CompartirIgual 4.0.
Todo el contenido disponible en el Portal de Revistas-UEx se distribuye bajo una licencia Creative Commons: Atribución 4.0 Internacional (CC BY 4.0)
