Land cover mapping of wetland areas in an agricultural landscape using SAR and Landsat imagery

Abstract

9Saline wetlands in the Monegros Desert, NE Spain, are situated in an 10 agricultural landscape which is undergoing significant changes. Agricultural 11 intensification in recent decades and current installation of new irrigation systems 12 threaten these valuable habitats, set to be included in the Natura2000 network. Their 13 preservation and successful management depend on the information available regarding 14 the transformation of surrounding areas. When soil and vegetation maps at adequate 15 scale are not available, remote sensing is an alternative means to obtain needed data. We 16 have used SAR data, taking advantage of the soil surface characteristics perceived in 17 SAR images. The objective of this work is to explore the capability of multitemporal 18 SAR data to characterize the land covers of these wetlands and their environment. We 19 have developed specific contextual classifications which take into account the statistical 20 properties of the radar distribution. Moreover, we tested the contribution of radar in 21Landsat classification. 22 2

Influence of Multi-Source and Multi-Temporal Remotely Sensed and Ancillary Data on the Accuracy of Random Forest Classification of Wetlands in Northern Minnesota

“…These active sensors are not as sensitive to atmospheric effects, penetrate clouds, and are operational at night, thereby increasing the temporal coverage of wetland mapping. Research has shown that data from multiple sources and over multiple seasons capture greater variation in hydroperiod and vegetative condition and thus have the potential to increase both classification accuracy and confidence [29][30][31].…”

Land cover classification of Lago Grande de Curuai floodplain (Amazon, Brazil) using multi-sensor and image fusion techniques

“…Wetland classification results from either multi-sensor or image fusion methods have been evaluated in the literature for tropical forested wetlands (Bwangoy et al 2010;Silva et al 2010;Rodrigues et al 2011) and grass-dominated temperate wetlands (Li and Chen 2005;Castañeda and Ducrot 2009). For tropical environments, Silva et al (2010) and Bwangoy et al (2010) reported good results using the multi-sensor method, the first studying a wetland adjacent to our study site, in the Amazon várzea, and the second mapping forested wetlands in the Congo River basin.…”

“…The success of each method can likely be attributed to differences in vegetation structure and hydrological patterns among study sites, with an emphasis on achieving good results from multi-sensor classification of river floodplains. For grass-dominated temperate wetlands, however, fusionbased classification achieved the most accurate classifications, while the combination of SAR and optical data using a multi-sensor approach worsened classification accuracies (Li and Chen 2005;Castañeda and Ducrot 2009). Again, these diverging results may be linked to the different vegetation physiognomies, which will interact differently with the electromagnetic spectrum (Henderson and Lewis 2008).…”

“…The second method combines n images in n different layers algorithmically, without creating a new set of images (Pohl and Genderen 1998;Luo et al 2002). Several studies have assessed the effectiveness of integrating multi-sensor imagery to land-cover mapping (Castañeda and Ducrot 2009;Li and Chen 2005;Zhu and Tateishi 2006). Data from different sensors can be integrated using Object Based Image Analysis (OBIA) (Benz et al 2004;Blaschke 2010), which has been shown to be an effective method for wetland classification of combined optical and SAR imagery Santiago et al 2013).…”

The Effects of Point or Polygon Based Training Data on RandomForest Classification Accuracy of Wetlands

“…The additional information provided by including an area surrounding a training point, rather than using the data from a single pixel corresponding to a training point, has been reported to increase the representativeness of the training data and improve the accuracy of the classification [34][35][36]. After trial-and-error revealed minimal sensitivity to buffer size, we used a 5 × 5 cell window at 5 m spatial resolution (approximately 12.5 m buffer radius).…”