Mapping land cover in complex Mediterranean landscapes using Landsat: Improved classification accuracies from integrating multi-seasonal and synthetic imagery

TitleMapping land cover in complex Mediterranean landscapes using Landsat: Improved classification accuracies from integrating multi-seasonal and synthetic imagery
Publication TypeJournal Article
Year of Publication2015
AuthorsSenf, C., Leitão P. J., Pflugmacher D., van der Linden S., & Hostert P.
JournalRemote Sensing of Environment
Volume156
Pagination527-536
Keywordsimage classification, Landsat, Mediterranean, Phenology, Pseudo-steppe, STARFM
Abstract

Low-intensity farming systems are of great importance for biodiversity in Europe, but they are often affected by soil degradation or economic pressure, leading to either land abandonment or intensification of agriculture. These changes in land use influence local biodiversity patterns and require annual monitoring of land cover. To accuratelymap landcover in such spatio-temporal complex landscapes, it isimportant to capture their phenolog- ical dynamics andfine spatial heterogeneity.Multi-seasonal analyses using optical sensorswith amediumspatial resolution from10 to 60m(e.g. Landsat) have been used for this task, but data availability can be scarce due to cloud cover, sub-optimal acquisition schedules and data archive access restrictions. Combining coarse spatial res- olution data fromtheMODerate-resolution Imaging Spectroradiometer(MODIS) and Landsat provides opportu- nities to close these gaps by simulating Landsat-like images atMODIS temporal resolution. In this study,we test whether and by what degree land cover maps of complex Mediterranean landscapes improve by integrating multi-seasonal Landsat imagery, as well aswhether STARFM-simulated imagery can be usedwhenever original multi-seasonal Landsat observations are unavailable. Therefore, we develop different classification scenarios based on seasonally varying data availability and based on original and simulated Landsat data. Results show that multi-seasonal Landsat data from spring and early autumn are crucial for achieving satisfying mapping accuracies (overall accuracy 74.5%). Using synthetic Landsat imagery increases classification accuracy compared to using single-date Landsat data, but accuracieswere never as good as a classification based on original data.We conclude thatmulti-seasonal data is essential for mapping complex Mediterranean landscapes and that STARFM can be used to compensate for missing Landsat observations. However, if Landsat data availability is sufficient to cover all phenologically important dates, we suggest relying solely on Landsat.