Integration of multi-sensor analysis and decision tree for evaluation of dual and quad-Pol SAR in L- and C-bands applied for marsh delineation

Abstract

Marsh is a wetland type characterized by hydromorphic soils, herbaceous vegetation, aquatic and emergent vegetation; usually, the apparent water surface does not exceed 25% of the area. Multi-polarized active remote sensors with different frequencies have characteristics that make them ideal for mapping and delineating marsh areas since they provide information on canopy roughness, vegetation moisture and amount of biomass. Therefore, the main objective of this study is to develop a method based on multi-frequency radar satellites images to delineate marsh areas using decision tree classification. In order to reach this objective, we sought to answer the following questions: (1) Are L-band SAR images more efficient for marshes delineation than C-band SAR images? (2) Is multi-sensor (L and C-band) integration more accurate for marsh areas delineation than a single sensor? and (3) What are the most efficient channels for marshes delineation? Our findings showed that L-band images present greater proportion correct (PC) for marshes delineation compared to C-band images. However, the greatest PC was found using integration of Alos Palsar 1 and Sentinel 1 satellites images, reaching more than 72% of correctness. Regarding the polarization importance to Alos Palsar 1 image, HVVH presented the highest importance, with 29%, followed by VH and HV polarizations, both with 28%. For Sentinel 1 image, the most important polarization was VH, with 22%, followed by VV + VH that presented 20%. HVVH polarization was the most important in Alos and Sentinel images integration, with 35%, followed by Alos Palsar HV and VH, with 34 and 33%, respectively. Thus, we concluded that the method based on SAR multi-frequency data integration used in this study can be easily applied by other researchers interested in marsh delineation since the radar images used are freely available and can be processed and manipulated in free GIS software.

SAR analysis of wetland ecosystems: Effects of band frequency, polarization mode and acquisition dates

“…We acquired five pairs of RADARSAT-2 and TerraSAR-X images with time-shifts of 2-11 days because it had been identified as acceptable (Amani et al 2019;Simioni et al 2019). After the correlation analysis, however, we excluded two pairs (timesteps 2 and 4) that, despite having short time-shifts (3 and 11 days, respectively), showed large differences in flooded area and thus in the values of SAR features (Fig.…”

“…band frequency, polarization mode, number of acquisition dates) on discrimination of wetland vegetation has been studied extensively in recent years (Mahdavi, Salehi Sr, et al 2017;Wohlfart et al 2018). Regarding band frequency, the advantage of the X-band over the C-band (Maleki et al 2020;S-band (van Beijma, Comber, andLamb 2014) for discriminating wetland vegetation types has been described, as has the advantage of the Lband over the C-band for mapping emergent (Simioni et al 2019) or woody (Amani et al 2019;Mahdianpari, Salehi, Mohammadimanesh, and Motagh 2017) vegetations. The literature also describes the advantage of quad-polarized data over dual-polarized data to discriminate woody (de Almeida Furtado, Silva, and de Moraes Novo 2016; Mahdianpari, Salehi, Mohammadimanesh, and Motagh 2017) (de Almeida Furtado et al, 2016, Mahdianpari et al, 2017a or boreal (Mahdianpari, Salehi, and Mohammadimanesh 2017) wet-land vegetation types.…”

“…However, the findings show that C-band is slightly more relevant than X-band to discriminate woody vegetations and ponds. In this study, L-band frequency SAR images, such as ALOS-2 data, were unfortunately unavailable during the period investigated, while several studies have highlighted the added value of the L-band over the C-band in discriminating woody vegetation such as swamps (Amani et al 2019;Mahdianpari, Salehi, Mohammadimanesh, and Motagh 2017) or emergent vegetation such as marshes (Simioni et al 2019). In summary, X-band frequency is the most discriminating for herbaceous vegetation, L-band or alternatively C-band for woody vegetation, and C-band for ponds.…”

“…Elsewhere, two time series, TerraSAR-X (dual-pol/X-band) and RADARSAT-2 (quad-pol/Cband), were acquired with 4-month time-shifts to map boreal wetlands . In another study, one ALOS-1 (quad-pol/L-band) and one Sentinel-1 (dualpol/C-band) image with a 7-year time-shift were compared to delineate a wetland marsh (Simioni et al 2019). Thus, new comparative analyses of band fre-quency, polarization mode and number of acquisition dates with SAR data acquired under similar hydrological and phenological conditions are needed.…”

Classificação Baseada em Objeto de Tipologias de Cobertura Vegetal em Área Úmida Integrando Imagens Ópticas e SAR

“…A maior importância observada para distinguir tipologias vegetais ocorreu para a polarização cruzada Sentinel-1 VH (9,7%), frente à polarização VV (8,9%). De forma semelhante, Simioni et al (2020) A resolução espectral influenciou na classificação, mas a resolução espacial não se mostrou relevante. As maiores importâncias ocorreram tanto em bandas com 10 m quanto com 20 m. As bandas ópticas apresentaram, no total, as maiores importâncias na classificação, mesmo com diferentes resoluções espaciais.…”