Feasibility study of flood risk monitoring based on optical satellite data

Flood hazard monitoring covers a wide range of activities: from forecasting rainfalls to monitoring flood infrastructures. The research of flood risk monitoring based on the optical satellite data was conducted as part of the SAFEDAM project – “Advanced technologies in the prevention of flood hazard”. The project is financed by the National Centre for Research and Development in Defence and Security Programme. The aim of the SAFEDAM project is to develop an innovative system enabling the risk of river floods in Poland to be monitored using, inter alia, unmanned aerial platform, satellite and aerial data. For this purpose, different photogrammetric and remote sensing techniques were tested and will be used in monitoring the flood risk, the flood phenomenon and the condition of levees. A comprehensive system will enable the multi-source collection of data, automatic data analysis and visualization for hydrological services and crisis management professionals.

For the purpose of flood risk monitoring, depending on the scale on which the analysis is performed, a variety of satellite data is used. In order to carry out effective monitoring, several conditions must be met: 1. data should be collected at the appropriate frequency depending on the mode (preventive/ operational); 2. the data should be adequately detailed to provide information on the state of the water and the status of the flood embankments.

Data from meteorological systems is useful to forecast precipitation and its intensity. Very high-resolution satellite images (VHRS) and high-resolution satellite images (HRS) can be used to monitor flood embankments and their condition, and to monitor floodplains. This applies to both radar and optical data. Radar data is weather independent and generally considered to be better suited for water detection (Pluto-Kossakowska et al. 2017), while optical data is easier for photointerpretation and acts as a compaction of measurements from radar data.

The method of performing flood risk maps and flood risk monitoring has been presented by Skakun (2012) for Namibia with the use of multi-temporal LANDSAT images. Shaker et al. (2012) determined one of the best methods of using SPOT PAN satellite data to estimate the risk of flooding. Malinowski et al. (2015) conducted a study in which the total accuracy of twelve maps of flood ranges obtained by the decision tree method ranged between 77% and 95%. Asare-Kyei et al. (2015) used RapidEye satellite data to produce land cover maps in order to obtain flood risk indicator maps. An analogous approach which presents the usage of optical and radar satellite data to produce flood risk maps (where optical data is used for the implementation of land cover maps) and, based on radar data, identified the extent of the flood presented, was introduced by Stancalie et al. (2007). Byun et al. (2015) present a methodology for detection of flood changes based on VHRS KOMPSAT images from two dates without the need for user interaction. Cundill’s (2014) article presents the usage of remote sensing as a tool for the inspection of grass-covered embankments. Summarising the existing studies, various research teams use different methods and indices to identify flood risk areas depending on both the type of the area of interest and the data availability (the issue of costs and availability of data in the archive). Optical data is typically used to obtain information about the current status of land cover, while the extent of flood waters has frequently been obtained on the basis of radar data.

Optical satellite remote sensing systems are characterized by a variety of technical parameters, including very important spatial and spectral resolution. Both of these parameters are crucial for monitoring the condition of embankments and forecasting the occurrence of floods. A detailed analysis of currently available VHRS and VHR satellite data was carried out, both in terms of their spatial resolution and their spectral resolution (Tables 1 and 2). The tables present the currently operating satellite systems as well as the systems operating in 2010, because this was a year in which there was a flood threat in Poland and data from this period can be used as part of the research to test various image processing methods in terms of flood risk monitoring.

Table 1

VHRS optical satellite data parameters

Table 1 presents all available VHRS optical satellites, along with their characteristics. Among these systems, WorldView-2 data was analysed, mainly due to the availability of data for the area of interest during the flooding and Plèiades data was analysed because of the daily revisit time and future use of data in the SAFEDAM system.

Table 2 presents selected HRS optical satellites, along with their characteristics, as a complement to the VHRS data. Sentinel-2 was chosen for research because of the free data policy and convenient access to the imagery database.

Table 2

Selected high resolution (HRS) optical satellite data parameters

The temporal resolution of the satellite systems means the revisit time and the possibility of acquiring new imageries. For the area of medium latitudes, this ranges from 1-2 days (Plèiades and SPOT constellation) up to several days (e.g. LANDSAT, Sentinel). It depends on the construction of the satellite system (a single satellite or a constellation, a scanner operating only in a nadir or a scanner with the possibility of tilting along and across the direction of flight), and the possibilities of its programming. In the case of individual satellite systems with HVRS, where programming is possible, it is usually a few days (3-4 days). For preventive monitoring this is a sufficient revisit time, but the possibilities of the usage of satellite optical systems are limited by the weather conditions and cloud cover. The experience so far shows that it is generally possible to obtain optical data every 7-10 days. However, there are periods when obtaining optical data is impossible due to unfavourable weather conditions.

Optical data, due to frequent cloudiness occurring in the period of increased flood hazard, is less commonfor operational use in monitoring damage to levees. It is possible to use such data only in cloudless conditions or in the presence of slight cloud cover. This means that the identification of damage to flood embankments with the use of optical data can be carried out as quasi-continuous monitoring, conducted in order to detect damages early. Examples of optical data with very high spatial resolution (IKONOS, QuickBird) being used to identify damage to flood embankments are shown in the literature - these cover, for example, damage to the sod in the main body of the embankment and on the embankment slopes (Hossain et al. 2004). Optical data is much more frequently used to assess the state of damage after a flood event or to determine the extent of flooding, or for flood risk assessment (Stancalie et al. 2007; Ho et al. 2010; Skakun 2012; Shaker et al. 2012; Asare-Kyei et al. 2015; Byun et al. 2015; Hu et al. 2015; Malinowski et al. 2015).

In order to clarify the nomenclature and build a coherent vocabulary, the following definitions are provided: ‘identification’, as the possibility of indicating the location of the damage, but without specifying its type, and ‘recognition’, as providing the type of damage, which seems quite a limited task in the context of satellite images, especially those with high resolution. As a result of the analysis of the type and size of the potential damage to flood embankments and the available satellite data (Tables 1,2), the possibilities of using individual types of optical data to identify damage to flood embankments have been indicated (Tables 3, 4, 5). Most often with this analysis, it is possible to point out a potential place of damage, but without indicating the cause. Satellite data is only used to monitor and indicate places for more accurate measurements. Supplementary measurements based on more precise technologies are recommended.

Table 3

Damage identification based on VHRS 0.3 m × 0.3 m pixel optical data

Table 4

Damage identification based on VHRS 0.5 m × 0.5 m optical data

Using optical data with a maximum spatial resolution of 0.3 m × 0.3 m it is possible to detect damages of size equal to or bigger than 0.3 m × 0.3 m. Table 3 presents the estimated accuracy of results based on the use of airborne imagery or satellite data with the highest available resolution, such as WorldView-3. It is possible to see a smaller artefact on a VHRS image, but it is not feasible to determine details such as type of damage.

Table 4 presents the estimated accuracy of results based on the use of optical VHRS data with a 0.5 m × 0.5 m pixel size. An example of satellite sensors are Plèiades satellites.

Figure 1 presents damaged flood banks caused by animals. The holes were caused by the beavers living in the area. Grazing animals on the levees and in their immediate vicinity violates the construction of levees and causes changes in soil mass.

Figure 1

Table 5 presents the estimated accuracy of results for optical data with a 2 m × 2 m pixel size. It is possible to use SPOT 6 and 7 data to identify damage with approximate dimensions of 2-5 metres and higher. Large-scale damage, with dimensions of several or several dozen metres can be detected on the basis of optical data with a 10 m × 10 m pixel, such as Sentinel-2.

Table 5

Damage identification based on HRS 2 m × 2 m optical data

Figure 2 presents an example of a landslide in Poland in 2015 and levee surface damage caused by people. Wheel rims are clearly visible, probably from cars, quads or agricultural machinery. The levee must maintain the correct geometry. Changes in the soil mass cause shifts in the geometry and changes in shaft stability. In the case of traces, water is collected in them and the filtration is disturbed, which reduces the stability of the shaft.

Figure 2

The main goal of the SAFEDAM project is to create an IT system for managing spatial data related to levees. The system will enable data collection with a non-invasive method of data acquisition (UAV and satellite data), automatic data analysis, visualization for hydrological services and for society, and provide tools for crisis management professionals and rescue teams during the intervention phase.

Earth observations are one of the key elements of the system, providing continuous information about a large area of interest at the same time. Analysis based on satellite data is used in both preventive and intervention mode. Satellite data can be used for large-scale image analysis to pre-select areas dedicated to unmanned platform measurement. Using processing and the observation of time series, the multi-temporal analysis tool may indicate simple analyses based on vegetation indices, to identify landslide areas, the occurrence of illegal objects, anthropopressure and the impact of animals.

As a result of the conducted analysis it was found that NDVI is the most suitable index for the automatic determination of a water range with a threshold value of -0.22 (for cloudless satellite scenes). This threshold value makes it possible to detect water with a slight overestimation error of 1-1.5% of the reference water mask area. In Rokni et al. (2014), where Landsat imagery was used for water surface detection, similar NDVI values were acquired (-0.21 and -0.29). However, the detection accuracy in Rokni et al. (2014) was slightly higher, which may have been caused by the selection of a more homogeneous, water-covered area than the test area selected for this study (Fig. 5). Moreover, whereas the NDWI and modification of this index gives accurate results in many studies (El Asmar et al. 2013; Rokni et al. 2014), in the analysis presented above the NDWI-based water detection was not satisfactory. Additionally, according to the proposed approach, in the event that high-level clouds (e.g. Cirrus) or middle-level clouds (e.g. Altus) occur over the analysed area, it is necessary to apply a two-stage processing workflow and two different threshold values for cloudless and cloudy parts of the scene.

Damage and changes on the surface of the embankments are detectable, but it is difficult to clearly determine the type and cause. The presented research has shown that damage visible as bare soil can be detected using the MSAVI or NDVI indices. However, if the threshold value is too high, the bare soil is overestimated.

This feasibility study will support the SAFEDAM System by providing a clear overview of the existing optical satellite systems and their capabilities, as well as taking into account their limitations.