**1. Introduction**

Near real-time and statistical information about flooded areas is essential for several public services, i.e., emergency, rescue, recovery, spatial planning, habitat monitoring, and adaption to climate change. Satellite remote sensing can provide timely and operational data as well as statistical spatial information about inundated areas covered with water. Two types of satellite imagery are available for monitoring surface flood dynamics: optical and synthetic aperture radar (SAR). Optical satellite remote sensing can only be applied in cloud-free situations. However, floods often occur during long-lasting periods of precipitation and persistent cloud cover. Therefore, SAR systems are usually a preferred tool for the monitoring of floods from space. A smooth open water surface is characterized by a low SAR backscatter, and this difference in backscatter response generally allows flood mapping [1]. Over the last decade, various methods for deriving the flood extent from SAR data have been proposed [2–18]. Based on summaries by Martinis et al. [18] and Liang and Liu [8], the most commonly applied methodology for flood mapping from a single image is histogram thresholding, which can be used in combination with different image processing approaches. Temporal change detection techniques [19,20] and coherence analysis [21] have also been used for open water mapping. However, temporal change detection approaches require two images and can therefore be limited by the temporal coverage of satellite imagery. To improve flood mapping accuracy, the advantages of ancillary data, such as the DEM (digital elevation model) derived HAND (height above the nearest drainage) index and the catchment derived DIST (distance from drainage) index as well as land use map, have been demonstrated in several studies [17,18,20,22]. Most of the proposed approaches

**Citation:** Sipelgas, L.; Aavaste, A.; Uiboupin, R. Mapping Flood Extent and Frequency from Sentinel-1 Imagery during the Extremely Warm Winter of 2020 in Boreal Floodplains and Forests. *Remote Sens.* **2021**, *13*, 4949. https://doi.org/10.3390/ rs13234949

Academic Editor: Alban Kuriqi

Received: 15 October 2021 Accepted: 2 December 2021 Published: 6 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

for flood mapping are semi-automatic. A fully automatic methodology that integrates split thresholding and fuzzy logic classification has been proposed and applied by Martinis et al. [18] for the processing of TerraSAR-X, and by Twele et al. [23] for the processing of Sentinel-1 (S1).

Recent studies by Grimaldi et al. [24] and Tsyganskaya et al. [25] have summarized the approaches of flood mapping under the forest canopy. The study by Grimaldi et al. [24] shows that the most commonly applied method for the detection of flooded areas under vegetation is the identification of increased backscatter values compared to other objects. The penetration depth of the SAR signal into vegetation is higher for longer wavelengths, so the use of the L-band has been recommended [26–28]. However, several studies [20,29,30] have demonstrated the capabilities of C-band and X-band data in the identification of flooded vegetation, especially in the case of sparse forests and leaf-off conditions. Copolarized signals (HH or VV) are preferred over cross-polarized signals for mapping water under vegetation. Studies have indicated that the use of HH-polarization leads to more accurate results compared to VV-polarization [31,32]. Moreover, the use of polarimetric decomposition and/or interferometric SAR coherence has been utilized for the mapping of floods under vegetation [33]. However, the availability of full polarimetric data is often limited in terms of spatial extent and temporal coverage.

Estonia is known for its large seasonal riverside areas that are flooded over annually. The surface area of the Estonian floodplain grasslands with a high nature conservation value is estimated to be 16,000 hectares. According to the EU Habitats Directive, northern boreal alluvial meadows (habitat type code 6450) are grasslands situated on the banks of large rivers, in sections with slow flow, which are frozen in the winter and flooded in the spring–summer period. However, extremely warm winters in Estonia during the last five years have also caused large flooding during winter [34]. Extreme changes in inundation extent, depth, and duration define phonological patterns, animal migration routes, and human living spaces [35]. Therefore, it is important to monitor the temporal and spatial changes in flooded areas.

The boreal forest encompasses approximately 30% of the global forest area and provides critical services to local, regional, and global populations. Communities benefit from ecosystem services provided by forests for fishing, hunting, leisure activities, and economic opportunities [36]. Countries such as Canada, Finland, Sweden, and Russia extract wood from boreal regions for their forest industries [36]. Flooding causes disturbances in forest management, resulting in economic losses. The vulnerability of the forest ecosystem in a changing climate has been discussed in Gauthier et al. [36] and Hari and Kulmala [37]. Previous studies have expressed the importance of flood monitoring in areas with emerging vegetation for a comprehensive evaluation of the economic and environmental costs of floods [38–40]. Recent mild winters in Estonia have affected the forest industry. Forest management is impossible due to unfrozen soils and floods [41]. However, the spatial extent and duration of floods during the winter period in Estonia is still unknown.

At the European scale, two flood-monitoring services are provided. The (1) Copernicus Emergency Management Service (EMS) [42] provides a free-of-charge mapping service in cases of natural disasters, man-made emergencies, and humanitarian crises throughout the world. This service can be triggered by request in the case of an emergency. The (2) Copernicus Land Monitoring Service (CLMS) provides a pan-European, high-resolution product known as Water and Wetness. This product shows the occurrence of water and wet surfaces over the 2009–2018 period. Thematic maps were produced for the years 2015 and 2018. These layers are compiled from multi-temporal high-resolution optical and radar satellite imagery [43].

However, these services do not provide information about the inter-annual variability of water extent on the floodplains, nor information about the flooded forest areas. Therefore, the current study was initiated with the following aims:

• Set up an optimal automatic workflow for open-water and flooded forest mapping from S1 data.

