Flood Monitoring in Vegetated Areas Using Multitemporal Sentinel-1 Data: Impact of Time Series Features

Abstract

Synthetic Aperture Radar (SAR) is particularly suitable for large-scale mapping of inundations, as this tool allows data acquisition regardless of illumination and weather conditions. Precise information about the flood extent is an essential foundation for local relief workers, decision-makers from crisis management authorities or insurance companies. In order to capture the full extent of the flood, open water and especially temporary flooded vegetation (TFV) areas have to be considered. The Sentinel-1 (S-1) satellite constellation enables the continuous monitoring of the earths surface with a short revisit time. In particular, the ability of S-1 data to penetrate the vegetation provides information about water areas underneath the vegetation. Different TFV types, such as high grassland/reed and forested areas, from independent study areas were analyzed to show both the potential and limitations of a developed SAR time series classification approach using S-1 data. In particular, the time series feature that would be most suitable for the extraction of the TFV for all study areas was investigated in order to demonstrate the potential of the time series approaches for transferability and thus for operational use. It is shown that the result is strongly influenced by the TFV type and by other environmental conditions. A quantitative evaluation of the generated inundation maps for the individual study areas is carried out by optical imagery. It shows that analyzed study areas have obtained Producer’s/User’s accuracy values for TFV between 28% and 90%/77% and 97% for pixel-based classification and between 6% and 91%/74% and 92% for object-based classification depending on the time series feature used. The analysis of the transferability for the time series approach showed that the time series feature based on VV (vertical/vertical) polarization is particularly suitable for deriving TFV types for different study areas and based on pixel elements is recommended for operational use.

Flooding Extent Mapping for Synthetic Aperture Radar Time Series Using River Gauge Observations

“…False positives may also occur for flat man-made objects, such as roads, parking lots, and airfields [12]; this was not the case in our case because of the low degree of urbanization in the study area. In contrast, the inclusion of VH data into flood monitoring analyses can be beneficial in comparison to using VV data alone; the effect is, however, study site specific [10], [45]. Moreover, multi-data-source methods can benefit from C-band VH by exploring other properties of this polarization, such as its lower sensitivity to surface water roughness than VV data [12].…”

“…A patchy pattern due to vegetation is often present in nonarid study areas. Several processing steps have been proposed to produce continuous flooding extents, e.g., region growing [7], [12], [17], [46], rule-based algorithms [6], multitemporal analysis [10], [14], [45], wavelet analysis [47], and others.…”

A Modeling-Based Approach for Soil Frost Detection in the Northern Boreal Forest Region With C-Band SAR

“…Floods in boreal forests were mapped by Voormansik, et al, (2014), who concluded that additional testing of the X-band sensor should be carried out with more detailed information regarding tree species, stand age, height and density. ESA's Sentinel-1 satellite pair offers a valuable SAR data source suitable for monitoring flood conditions (Twele, et al, 2016;Clement, et al, 2018;Ruzza, et al, 2019;Tsyganskaya, et al, 2019;Uddin, et al, 2019), especially in northern latitudes due to nearly daily temporal resolution. However, the land areas are regularly imaged only with VV-and VH-polarizations, which are less useful for detecting floods under forest canopy compared to HH-polarization (see section 2.2)…”

Construction of High Spatial-Temporal Water Body Dataset in China Based on Sentinel-1 Archives and GEE

“…With all this, on the VV polarization image, the pixels with a backscattering coefficient value not greater than −15 dB are regarded as a body of water, and on the VH polarization image, the pixels with a backscattering coefficient value not greater than −23 dB are regarded as a body of water. These characteristics of water bodies are what past research studies have used [16,[23][24][25][26]. These characteristics also allow us to detect water bodies with periodically frozen ice in winter using the above thresholds.…”

“…When the water level is high or the transmission is low, the radar signal is usually attenuated, while when the water level is low relative to the vegetation, double rebound scattering may occur [16]. Some studies have tried to detect water with obvious vegetation canopy on its surface [26,39,40], but most of these studies require the relative heights of vegetation and water surface and the distribution characteristics of vegetation leaves as input parameters of the model. These parameters need to be measured in the wild, so these existing studies are basically limited to small scales.…”