Surface Water and Floods Detection with Multispectral Satellites: History
View Latest Version
Please note this is an old version of this entry, which may differ significantly from the current revision.
Subjects: Engineering, Civil
Contributor:
Cinzia Albertini
,
Andrea Gioia
,
Vito Iacobellis
,
Salvatore Manfreda

The use of multispectral satellite imagery for water monitoring is a fast and cost-effective method that can benefit from the growing availability of medium–high-resolution and free remote sensing data. Since the 1970s, multispectral satellite imagery has been exploited by adopting different techniques and spectral indices. The high number of available sensors and their differences in spectral and spatial characteristics led to a proliferation of outcomes that depicts a nice picture of the potential and limitations of each. Satellite remote sensing applications for water extent delineation and flood monitoring are reviewed to highlight trends in research studies that adopted freely available optical imagery. In addition, the performances of the most common spectral indices for water segmentation are qualitatively analyzed and assessed according to different land cover types to provide guidance for targeted applications in specific contexts. Finally, common issues faced when dealing with optical imagery are discussed, together with opportunities offered by new-generation passive satellites.

  • multispectral satellite imagery
  • water detection
  • flood mapping
  • multispectral indices

1. Introduction

Remote sensing techniques play a crucial role in monitoring water bodies and assessing river dynamics, providing effective support to surface water management. Considering the growing availability of medium–high-resolution and free remote sensing data, both active and passive sensors onboard different satellites are being extensively used for water segmentation and flood mapping. Satellite data are, in fact, of particular interest when observations over wide areas are necessary, for example, to assist with large-scale flood phenomena. Active and passive satellite sensors operate in the microwave segment of the electromagnetic spectrum, while passive satellites include both sensors operating in the microwave spectrum segment and in the portion from the visible to the thermal infrared (optical imagery). Although active microwave sensors provide their own source of illumination, thus being in operation day and night and under all weather conditions, the use of passive multispectral sensors is relatively straightforward [1], and observations are easily interpretable. The availability of several spectral bands with different wavelengths, in fact, makes it possible to derive valuable information from each band and surfaces’ spectral signatures, or through direct interpretation of true- and false-color composites. In addition, with simple math algebra, vegetation indices exploit the reflectance characteristics of different objects.
Multispectral images have been successfully used for water body and river monitoring, change detection, and water feature extraction [2][3][4][5]. In addition, when cloud cover does not represent a major issue, the application of optical remote sensing to maps of flooded areas [6][7][8] offers the possibility of quickly and reliably identifying hazardous areas and supporting the implementation of flood coping strategies and response activities. In this context, the use of Unmanned Aerial Systems (UASs) may also provide high-resolution data in the presence of clouds, given the low-altitude range of these systems that allow flying below the cloud layer to be achieved (e.g., [9][10]). Moreover, the advent of Global Navigation Satellite System (GNSS) and, in particular, the Cyclone GNSS (CYGNSS) constellation, has recently demonstrated the potential of passive microwave instruments to estimate flood inundation and map inland surface water [11][12][13][14][15][16]. In fact, CYGNSS benefits from the L-band microwave capabilities of penetrating cloud cover and low–medium vegetation density, thus avoiding scattering problems due to precipitation and clouds [17].
Multispectral satellite data have also been recently proven to be suitable for monitoring the long-term spatio-temporal dynamics of river planform morphology and vegetation coverage, as well as land use/land cover evolution [18][19][20][21]. Boothroyd et al. [18] used vegetation indices derived from multispectral images to assess changes in the wetted river planform morphology and vegetation coverage along Po River Basin (Italy) with a multi-temporal Landsat image analysis (from 1988–2018). Similarly, Henshaw et al. [19] used Landsat scenes to determine the spatio-temporal trends of vegetation extent and channel position of six different sites along Tagliamento River (Italy). The incorporation of morphological processes into the evaluation of flood hazard has been carried out over both long- [18][19] and short-term timescales [22] and has shown to be of relevance for flood risk management. In addition, the hydrodynamics characteristics of the flow during a flooding event can modify vegetation coverage and flow resistance through erosion and sediment deposition [23][24]. In this regard, the information that can be inferred from optical remote sensing data represents a unique opportunity for monitoring planform, vegetation, and land cover changes, improving and supporting flood hazard predictions and modelling.
Over the past few years, different methods for water segmentation and flooded area mapping using multispectral satellite images have been proposed in the literature. These are generally divided into single-band and multi-band approaches. While the former discriminates water features from other surfaces by thresholding a single band, usually the near infrared (NIR), which is highly absorbed by water, the latter uses a combination of two or more bands for deriving spectral indices. The multi-band method, also called water index method, is a promising approach for surface water mapping both in data-rich and data-poor catchments [25]. It performs better than the single-band method in detecting land surface water, since it exploits differences in the reflectivity of the selected bands [26].
Multispectral remote sensing-derived indices allow flooded areas and water bodies to be quickly and effectively recognized. However, this ability can be highly compromised by the presence of scattering noise, built-up areas, clouds, and shadows. In addition, spectral index performances can be different according to the geomorphological conformation of the territory, as well as the land cover conditions, i.e., crops, forests, artificial surfaces, or open water.

2. Multispectral Satellite Remote Sensing for Flooded Area and Wetland Inundation Mapping

Since the mid-1970s, several optical satellites have been monitoring the Earth’s surface. The Landsat program represents the oldest and longest-running satellite mission, operating since July 1972, when Landsat 1 was first launched. Over the years, new satellites equipped with different instruments were placed in orbit, and the most recent mission, represented by the Landsat 9 satellite, was recently launched in September 2021. Satellites still in operation are Landsat 7, Landsat 8, and the aforementioned Landsat 9. Optical sensors onboard these satellites capture images at different spectral (i.e., wavelengths and bandwidths) and spatial resolutions, with a common revisit time of 16 days.
The Moderate Resolution Imaging Spectroradiometer (MODIS) instrument operates on the Terra and Aqua satellites, respectively launched in December 1999 and May 2002, and captures images in 36 spectral bands. Despite its coarse spatial resolution (see Table 1), MODIS has a daily revisit time, which is one of the most important factors when data obtained during the flood event are needed for flood extent delineation [27].
Sentinel-2 is a constellation formed by two twin satellites, S-2A and S-2B, respectively launched in June 2015 and March 2017, that together ensure a revisit time of 5 days. A multispectral instrument (MSI) mounted on Sentinel-2 provides images in 12 spectral bands (see Table 1). Together with Landsat, Sentinel-2 constitutes the family of medium–high spatial resolution multispectral satellites.
The National Oceanic and Atmospheric Administration’s (NOAA) Advanced Very High-Resolution Radiometers (AVHRR) sensor has been equipped on polar orbiting weather satellites since 1978. Since then, different instruments onboard several satellites have been imaging the Earth. The last version, AVHRR/3, currently operating on NOAA-15, -18, and -19 satellites, acquires daily images in six spectral bands (from the visible to the infrared portions of the electromagnetic spectrum) with a spatial resolution of 1.1 km, making it possible to perform flood mapping at very large scales.
These are examples of passive remote sensing programs offering free-of-charge data. In this work, we reviewed the use of medium–high-spatial resolution imagery for flooded area and wetland inundation mapping. In addition, as reviewed by Zhao et al. [28], Sentinel, Landsat, and MODIS are the most significant Earth observation satellite missions, based on the remote sensing impact index (RSIF) built by the authors, that will likely still be on the front line for environmental monitoring. For these reasons, we focused on the Landsat, MODIS, and Sentinel-2 satellites in our review.
The following table illustrates the main spectral and spatial resolution characteristics of Landsat 4- and 5-Thematic Mapper (TM), Landsat 7-Enhanced Thematic Mapper Plus (ETM+), Landsat 8-Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS), Sentinel-2 MSI, and Terra–Aqua MODIS instruments, with a particular focus on spectral bands commonly used in surface water monitoring studies. In addition, references and links to the data portals from which it is possible to visualize and freely download multispectral images from these satellites are provided in the same table.
Table 1. Main spectral and spatial resolution characteristics of the selected satellites: Landsat 4- and 5-Thematic Mapper (TM), 7-Enhanced Thematic Mapper Plus (ETM+), 8-Operational Land Imager (OLI), Sentinel-2 multispectral instrument (MSI), and Terra–Aqua Moderate Resolution Imaging Spectroradiometer (MODIS). Note that only bands commonly used in water detection methods are reported. (W and R stand for wavelength and spatial resolution, respectively).
  Landsat 4- , 5-TM Landsat 7-ETM+ Landsat 8-OLI Sentinel-2 MSI Terra–Aqua MODIS
Band Band Number W (μm) R (m) Band Number W (μm) R (m) Band Number W (μm) R (m) Band Number W (μm) R (m) Band Number W (μm) R (m)
Blue Band 1 0.45–0.52 30 Band 1 0.45–0.52 30 Band 2 0.45–0.51 30 Band 2 0.46–0.52 10 Band 3 0.46–0.48 500
Green Band 2 0.52–0.60 30 Band 2 0.52–0.60 30 Band 3 0.53–0.59 30 Band 3 0.55–0.58 10 Band 4 0.55–0.57 500
Red Band 3 0.63–0.69 30 Band 3 0.63–0.69 30 Band 4 0.64–0.67 30 Band 4 0.64–0.67 10 Band 1 0.62–0.67 250
NIR Band 4 0.76–0.90 30 Band 4 0.77–0.90 30 Band 5 0.85–0.88 30 Band 8 0.78–0.90 10 NIR 1
Band 2		 0.84–0.88 250
NIR 2
Band 5		 1.23–1.25 500
SWIR 1 Band 5 1.55–1.75 30 Band 5 1.55–1.75 30 Band 6 1.57–1.65 30 Band 11 1.57–1.65 20 Band 6 1.63–1.65 500
SWIR 2 Band 7 2.08–2.35 30 Band 7 2.09–2.35 30 Band 7 2.11–2.29 30 Band 12 2.10–2.28 20 Band 7 2.11–2.16 500
Data Access USGS EarthExplorer data portal [29]
https://earthexplorer.usgs.gov/
(accessed on 4 February 2022)		 Sentinel Scientific Data Hub [30]
https://scihub.copernicus.eu/
(accessed on 4 February 2022)		 USGS EarthExplorer data portal [29]
https://earthexplorer.usgs.gov/ (accessed on 4 February 2022)
NASA Earthdata Search [31]
https://search.earthdata.nasa.gov/search
(accessed on 4 February 2022)
LAADS DAAC Archive [32]
https://ladsweb.modaps.eosdis.nasa.gov/
(accessed on 4 February 2022)
 

2.1 Flood Mapping Approaches Using Optical Remote Sensing

Satellite observations from optical sensors have long been used for river monitoring and to assess inundation dynamics in the pre- and post-flood emergency. In detail, flooded areas and wetland inundations have been mapped using Landsat [41-48], MODIS [49-52], and Sentinel-2 imagery [53-56] in different contexts, including urban [57,58] and coastal environments [59-62]. Satellite-derived flood maps also represent support for emergency responses and the evaluation of flood severity and costs. Several studies assessed post-flood damages to infrastructures, built-up, and agricultural areas [63-68], as well as the estimation of flood impacts on natural environments and floodplain ecosystems (e.g., [61,62]).

Different methodologies were applied to detect flooded areas. A comprehensive review of optical images classification methods was carried out by Gómez et al. and can also be found in Radočaj et al. [70] and Foroughnia et al. [71].

Most of the studies adopted segmentation approaches to discriminate between water and non-water pixels. They included density slicing to a single band by visually inspecting the grayscale histogram of the satellite images or thresholding the different band ratios or compositions that form the spectral indices. Sims and Thoms [72] analyzed the vegetation response to floodplain inundation using Landsat TM images. The authors applied a ratio of Band 1 to Band 7 to map deep open water, while the change detection technique applied to Band 5 for processing preflood images and those acquired during the flood event was used to identify shallow water. NDVI maps were then used to study the relation between vegetation growth and flood frequency, from which it emerged that where vegetation was more vigorous, occurring where high NDVI values could be detected, flooding was registered less frequently. Frazier and Page [73] applied density slicing to Landsat TM infrared Band 5 to classify water and non-water regions and relate wetland inundation with river flood peak discharge. By observing Digital Number (DN) values in the red band of MODIS images, Thito et al. [48] identified threshold values for classifying inundated and non-inundated areas to ultimately define flooding frequency and duration from 2001 to 2012 in Okavango Delta (Botswana, Africa). Similarly, image thresholding was applied by Ludwig et al. [55] to multi-temporal Sentinel-2 imagery for large-scale wetland mapping mainly based on spectral indices. Thresholding represents an easily implemented technique for flood mapping based on the selection of a single appropriate threshold value from a bimodal intensity histogram that partitions the image into two classes, i.e., water and non-water. One of the most common thresholding approaches is Otsu’s method, aimed at maximizing the inter-class or, similarly, minimizing the intra-class variance [74]. Difficulties arise because in most of the cases, the histograms are not bimodal, but several classes are present in the flooded scene, and disturbance factors, such as dense vegetation canopies, do not allow one to identify the presence of water beneath them [71].

Supervised classification was also found to be one of the most widely used approaches for flood extent delineation and/or land cover/land use classification to assist with water detection. This method includes maximum likelihood (e.g., [66,75]), random forest (e.g., [42,76–78]), and support vector machine (e.g., [79,80]). Despite the large use of supervised classifiers, their use requires a priori knowledge of the classes to be identified. Such information assumes the form of large training datasets that must adequately describe the classification problem and contain representative class samples [20,69,70]. An alternative is represented by unsupervised approaches, mainly based on Iterative Self-Organizing Data Analysis (ISODATA) clustering [81]. For example, Thomas et al. [46] identified inundated areas in Macquarie Marshes in central–eastern Australia, using Landsat Multispectral Scanner System (MSS), TM, and 7-ETM+ images over 28 years and applying ISODATA, while Jung et al. [82] used the same classification algorithm to extract flood extent from Landsat 5-TM data and estimate the relationship between the flood discharge and the elevation extracted using a Digital Elevation Model (DEM) at the flood extent boundaries. Since unsupervised methods do not require training samples, they are especially adopted when there is knowledge about the classification problem [69] and can easily be transferred to different contexts [20].

In many studies, remote sensing data and derived flood extents were integrated or used in conjunction with ancillary data to help to avoid flooded area underestimation and misclassifications, especially caused by vegetation and forest cover [62]. Auxiliary information included DEMs and Light Detection and Ranging (LiDAR) products (e.g., [62,65,83–85]), or derived geomorphic indices, such as the Height Above the Nearest Drainage index (HAND [86]; also defined as the elevation to the nearest stream, H) and the Geomorphic Flood Index (GFI [87,88]). Totaro et al. [89] carried out a comparative analysis of geomorphic descriptors (i.e., H and the GFI) and satellite-based spectral indices derived from Landsat 8-OLI images for flood-prone area delineation. More recently, Mehmood et al. [90] implemented an innovative Flood Mapping Algorithm (FMA) on the cloud platform Google Earth Engine (GEE) using the MNDWI for water classification and filtering dark and steep vegetated hilly areas with NDVI and HAND maps, respectively.

Finally, from the proposed review, it was interesting to note some emerging trends in the last few years, especially the increasing interest in cloud computing for processing remote sensing products. The already mentioned GEE, in fact, allows data visualization and the analysis of ready-to-use satellite data and geospatial products to be performed at the planetary scale [91], and these can be collected from a vast archive, including Landsat imagery since 1982, and MODIS and Sentinel collections. Since 2016, several authors have developed tools and automated flooded area mapping algorithms on GEE to delineate the extent of the event or generate time-series flood maps [90,92–95]. Others just exploited the GEE environment capabilities of managing large amounts of data and offering parallel computations to process satellite data [96,97].

3. Multispectral Indices for Water Segmentation

Spectral indices are a combination of two or more spectral bands through which the water class is detected and separated from other features by exploiting the reflectance characteristics of different bands. Since 1974, several spectral indices have been proposed in the literature, each formed by a different band composition to enhance water features from other objects. The NDVI was introduced by Rouse et al. [33] to detect the vegetation greenness using Landsat 1-MSS images. This index exploits the contrast between the reflectance in the infrared and red bands, where negative values indicate the presence of water. To enhance water features, McFeeters [34] proposed the NDWI, which uses the green and NIR bands. In this case, the index assumes positive values in water regions. Using the same name, Gao [35] introduced an index that uses the NIR and SWIR1 bands. To avoid confusion, Gao’s NDWI was later renamed as Normalized Difference Moisture Index (NDMI) by Xu [26]. The same author, using Landsat TM images, developed a modified version of the NDWI, the so-called MNDWI, which exploits reflectance differences between the green and SWIR1 bands. This index was successfully introduced to better discriminate water surfaces in regions where built-up land areas dominate the background [26].
Other features can lead to the misclassification of water pixels, including dark surfaces and the presence of shadow, which can be predominant especially in mountainous areas. To reduce such environmental noises and improve surface water mapping, Feyisa et al. [36] proposed an Automated Water Extraction Index (AWEI). Two indices were formulated by the authors to distinguish situations in which shadow does not represent a major problem (AWEInsh) from those in which shadows and dark surfaces predominate (AWEIsh).
A modification of the MNDWI was introduced by Ji et al. [37] for Landsat 7-ETM+ and MODIS sensors. The author proposed to substitute the Landsat 7-ETM+ SWIR1 band with the SWIR2 band to form an index named NDWIL2,7. The same substitution was proposed for the MODIS SWIR1 band, replaced by the SWIR2 band to form the NDWIM4,7. Hereinafter, the index formed by the green and SWIR2 bands, regardless of the sensor, is referred to as MNDWI7, as proposed by Colditz et al. [38].
A further investigation of the literature was carried out to identify the most employed spectral indices derived from Landsat, Sentinel-2, and MODIS data. In particular, the NDVI, NDWI, NDMI, MNDWI, AWEIsh, AWEInsh, and MNDWI7 were found to be applied to both flooded area detection and water body mapping, while the Water Ratio Index (WRI) introduced by Shen and Li [39] for Landsat ETM+ imagery, was also found to be used in surface water delineation studies. Among these indices, the most frequently used are the NDWI (around 65.2% of applications with Landsat, 21.4% with Sentinel-2, and 16.1% with MODIS) and the MNDWI (66.7% of use with Landsat and 16.7% with both Sentinel-2 and MODIS). In Figure 1, the percentages of use of the MNDWI (left panel) and NDWI (right panel) from 2002 to 2021 for each satellite are shown. Similar trend patterns on their applications with Landsat, Sentinel-2, and MODIS were found both for the MNDWI and NDWI, i.e., an increasing use of Landsat and Sentinel-2 (pink and green lines, respectively) and a decreasing interest in spectral indices for MODIS (blue line) since 2015.
Remotesensing 14 06005 g002 550
Figure 1. Percentages of applications of the Modified Normalized Difference Water Index (MNDWI, left panel) and Normalized Difference Water Index (NDWI, right panel) from 2002 to 2021 using Landsat (pink line), Sentinel-2 (green line), and MODIS (blue line) data.
The following table illustrates the indices selected after an in-depth screening of the available documents. For each index reported in the table, both the general formula with the original bands composition from the authors that first proposed it and the specific formula for each selected sensor are provided.
Table 2. Spectral index formulas according to the original band’s composition and formulation for each selected sensor.
Index Formula   NDVI NDWI NDMI MNDWI WRI MNDWI7
Reference Rouse et al. [33] McFeeters [34] Gao [35] Xu [26] Shen and Li [39] Ji et al. [37]
 
Landsat
5-TM
7-ETM+
Landsat
8-OLI
Sentinel-2
MSI
Terra–Aqua
MODIS
Index Formula   AWEInsh AWEIsh
Reference Feyisa et al. [36] Feyisa et al. [36]
  4⋅(GREEN−SWIR 1)−0.25⋅(NIR+2.75⋅SWIR 2) BLUE+2.5⋅GREEN−1.5⋅(NIR+SWIR 1)−0.25⋅SWIR 2
Landsat
5-TM
7-ETM+		 4⋅(B2−B5)−0.25⋅(B4+2.75⋅B7) B1+2.5⋅B2 −1.5⋅(B4+B5)−0.25⋅B7
Landsat
8-OLI		 4⋅(B3−B6)−0.25⋅(B5+2.75⋅B7) B2+2.5⋅B3−1.5⋅(B5+B6)−0.25⋅B7
Sentinel-2
MSI		 4⋅(B3−B11)−0.25⋅(B8+2.75⋅B12) B2+2.5⋅B3−1.5⋅(B8+B11)−0.25⋅B12
Terra–Aqua
MODIS		 4⋅(B4−B6)−0.25⋅(B2+2.75⋅B7) B3+2.5⋅B4−1.5⋅(B2+B6)−0.25⋅B7

 3.1 Spectral Index Performances

Among flooded area detection studies (42% of total studies), Boschetti et al. [30] proposed a comparative analysis of several spectral indices to map flooded rice cropping systems using MODIS data. Validation on pure water pixels showed that among 11 selected indices, the best mapping accuracy was achieved by those based on the SWIR and visible bands, particularly the MNDWI (overall accuracy - OA - equal to 97%). Similarly, Munasinghe et al. [32] compared different inundation mapping methodologies, including supervised and unsupervised classification techniques, a change detection approach, and two spectral indices, i.e., the MNDWI and NDWI, based on Landsat 8-OLI satellite imagery. Despite the fact that other methods led to better performances, both indices achieved satisfactory results (OA values equal to 77.3% and 77.1% for the MNDWI and NDWI, respectively). Asmadin et al. [6] assessed the performances of seven water index algorithms, including, among others, the MNDWI, NDWI, NDMI, NDVI, and AWEInsh, derived from Sentinel-2A MSI and Landsat 8-OLI for coastal surface inundation mapping. In this case, indices from both sensors showed good accuracy (OA values above 94%). Using Sentinel-2 MSI data, Li et al. [111] performed an MNDWI-based segmentation to separate water and land features to ultimately characterize extreme flood impacts on the channel–floodplain morphology and sediment regime of the river system. The classification accuracy achieved 96%. More recently, Li et al. [94] combined Landsat images with precipitation data and high-resolution satellite imagery in the GEE environment to evaluate channel–floodplain dynamics. The authors used the MNDWI to extract the flooding extent, based on visual interpretation, achieving 93% of accuracy.

A higher number of studies that employed multispectral indices were found for surface water body detection (58% of total studies). One of the most significant works is the one proposed by Li et al. [112], in which Advanced Land Imagery (ALI) data and Landsat 5-TM and ETM+ images were selected to compare land surface water mapping based on the MNDWI, NDWI, and MNDWI7 in three different study sites. Despite the aim of the authors being to demonstrate the superiority of ALI data, Landsat imagery succeeded in delineating water features in all three regions. In particular, the indices based on the SWIR bands (i.e., the MNDWI and MNDWI7) showed very high and similar performances (OA values equal to 94.6% and 93.9% on average, respectively), while the accuracy of the NDWI was slightly lower (OA = 92% on average). Zhou et al. [29] evaluated the performances of six spectral indices, including the NDWI, MNDWI, AWEIsh, and AWEInsh, derived from three different sensors, i.e., Landsat 7-TM+, Landsat 8-OLI, and Sentinel-2 MSI, for water body mapping in Poyang Lake Basin, China. The authors showed the superiority of Landsat 8 and Sentinel-2 over Landsat 7 data and the higher performance of the NDWI in the selected study region (OA values equal to 95.7% and 95.6% for Landsat 8 and Sentinel-2, respectively). Ogilvie et al. [5] applied the spectral index segmentation method using Landsat imagery to map small lakes (from 1 to 30 ha) in Tunisia. The selected indices for validation purposes using Landsat 8 scenes of six different lakes included the MNDWI, NDWI, NDMI, and NDVI. The MNDWI had higher performances in four out of six cases (OA values above 89%), while in the other two sites, the NDWI and NDVI performed better (OA values equal to 80% and 88.1%, respectively).

To identify the best-performing spectral indices, we carefully analyzed the selected literature. No distinction was made among different satellite sensors, while accuracy values were evaluated separately for surface water detection and flooding extent delineation. The selected spectral indices for the performance assessment were the MNDWI, NDWI, NDMI, NDVI, AWEInsh, AWEIsh, WRI, and MNDWI7. As regards flood mapping studies, the most used spectral indices were the MNDWI, NDWI, and NDMI (39.4%, 30.3%, and 9.1% of use among flood studies, respectively). Only two studies were found that employed the WRI and NDVI (12.2%) and one study that used the AWEInsh, AWEIsh, and MNDWI7 (9%). Among the MNDWI, NDWI, and NDMI, the former was shown to be the best index both in terms of OA median value (93.03%) and the spread of the data. The NDMI had a median value very close to that of the MNDWI (93%); however, the data were more spread out. The same occurred for values of the NDWI, whose median was equal to 87.85%. Regarding surface water detection, the MNDWI and NDWI were the most common indices (27.8% and 26.4% of use among open water studies, respectively), followed by the AWEInsh (15.3%), NDVI (11.1%), AWEIsh (9.7%), WRI (4.2%), NDMI, and MNDWI7 (5.6%). In this case, the MNDWI had an OA median value of 95.37% and a relatively small spread of the data. However, some outliers were observed. The NDWI and AWEInsh had similar performances to those of the MNDWI, in terms of median value (94.41% and 94.80%, respectively), data spreading, and outlier values. The NDMI showed the lowest median value among all selected spectral indices (median OA value of 88.80%). Although the highest OA median value was observed for the WRI, only five studies were available, which may not be sufficient to satisfactory interpret its performance.

3.2 Classification According to Land Cover

Multispectral indices performances can vary according to the land cover, which can influence the ability to discriminate between water and other features. To achieve a better understanding of multispectral applications and identify the spectral index that is best suited for a specific land cover setting, the MNDWI and NDWI performances were qualitatively analyzed by classifying them according to different land cover types. Three main classes were considered, namely, crops, forests, and mixed land. The latter indicates heterogeneous areas that include the contemporary presence of the first two classes, wetlands, urban areas, and/or artificial surfaces. Figure 2 shows the accuracies of the MNDWI and NDWI for the three selected land cover categories. The MNDWI showed better performances than the NDWI, both in terms of median values and data variation (left panel). Moreover, the MNDWI always showed the best performance in all three land cover categories (upper-right panel) compared with the NDWI in the same categories (lower-right panel). In terms of OA, the median values for the MNDWI were always above 93% for both crops and mixed lands (94.30% and 93.36%, respectively) and above 90% for forest areas (91.40%), while for the NDWI, values never exceeded 90%, with the highest value being observed in forests (88.70%) and the lowest in mixed land (85.98%). For croplands, the NDWI had an OA median value equal to 87%. As regards data variation, MNDWI values showed the lowest spread, while higher variation was observed for NDWI values, with the highest spread of mixed land data.

Figure 2. Results of the performances of the MNDWI and NDWI in detecting flooded areas in different land cover settings, expressed in terms of overall accuracy (OA). The blue line represents the median value of the general OA values of MNDWI and NDWI (left panel) and of the OA values of the MNDWI and NDWI in each land cover category (right panels).

4. Discussion and conclusions

In this work, an up-to-date review of remote sensing applications for water mapping was carried out focusing on satellite remote sensing programs that offer free-of-charge optical data. Optical satellites represent a straightforward instrument for flooded area and water body mapping. In fact, the multi-band sensors allow the spectral signatures of different objects to be exploited and information to be derived through a direct visual interpretation of scenes of specific bands or color composites. In addition, multispectral imagery allows information about river morphology dynamics and land cover changes to be integrated in flood risk modelling and provides good spatial resolution for flood management applications, although with some limitations. The proposed analyses had the more general aims of identifying sensors and methods most used and best suited for monitoring water-related processes, highlighting the potentials of spectral indices, and providing some general practical guidance for targeted applications in different contexts. We believe that this could be beneficial not only for satellite-based remote sensing but also for UAS-based environmental monitoring, which in addition allows the cloud cover issue that affects optical remote sensing to be overcome. The analyses presented here can, indeed, be transferred to airborne-based applications to identify the best methodology that can ensure a reliable flood-prone area delineation using multispectral sensors. The opportunities offered by UAS technology are not only related to the possibility of flying below cloud cover layers and vegetation canopies, as it also enables submeter-level spatial resolution acquisition necessary for a detailed understanding of flood processes [see 9]. On the other hand, UASs have the limitation of not being able to survey vast areas. Although this issue could be overcome with multipleUASs flying simultaneously over the same area, proper regulations that allow multi-UAS flight to be performed are still lacking. Nevertheless, the potentiality offered by such a system represents a versatile support to satellite imagery.

Promising applications of passive remote sensing under cloud conditions are also those from the CYGNSS constellation, which presents the advantages of similar microwave sensors (such as Synthetic Aperture Radar (SAR)) of seeing through clouds but at higher temporal resolutions. New-generation passive satellites are, in fact, characterized by a higher revisit time over land and ocean, which increases the chance to have closer post-flood observations. The aforementioned CYGNSS is one example, but valuable applications, still at their dawn, are also those offered, for example, by the family of small satellites of CubeSat. Although suffering from cloud cover problems, as with optical data, it provides both high-temporal (nearly daily)- and -spatial-resolution (~3 m) multispectral imagery on the global scale [126-128].

Future research directions should expand the unprecedent opportunities offered by the new generation of passive satellites, which could be enhanced by merging data from multiple sources, thus combining the potential of multispectral imagery at high temporal resolution (e.g., CubeSat data) with the capabilities of radar data (e.g., CYGNSS). Moreover, hybrid approaches, also including NOAA AVHRR data, which increase the possibility of obtaining cloud-free data, could also be explored. Finally, the advent of emerging technologies such as hyperspectral imagery (e.g., the recently launched Environmental Mapping and Analysis Program (EnMAP) German satellite) offers the possibility to better discriminate the spectral signature of different surfaces (thanks to the many different spectral bands) and to be used in conjunction with multispectral data, allowing detailed information to be retrieved, which is not achievable with multispectral data alone. Regarding multispectral indices, future opportunities can be represented by the application of the WAVI in flood mapping studies.

In conclusion, both optical and passive microwave satellite sensors will likely continue to represent a reference for Earth imaging applications, enabling a better understanding of surface water dynamics. Indeed, European Union’s Earth Observation Copernicus Programme makes a large use of these remote sensing technologies for monitoring the Earth’s surface and supply operational products (e.g., the free-of-charge flood maps produced by the Rapid Mapping module of Copernicus Emergency Management Service (CEMS)) to the end-user community [129].

This entry is adapted from the peer-reviewed paper 10.3390/rs14236005

References

  1. Schumann, G.J.-P.; Brakenridge, G.R.; Kettner, A.J.; Kashif, R.; Niebuhr, E. Assisting Flood Disaster Response with Earth Observation Data and Products: A Critical Assessment. Remote Sens. 2018, 10, 1230.
  2. Buma, W.G.; Lee, S.-I.; Seo, J.Y. Recent Surface Water Extent of Lake Chad from Multispectral Sensors and GRACE. Sensors 2018, 18, 2082.
  3. Liu, D.; Li, Y. Extraction of Water-Body in Remote Sensing Image Based on Logic Operation. In Proceedings of the 2011 19th International Conference on Geoinformatics, Shanghai, China, 24–26 June 2011; pp. 1–4.
  4. Rokni, K.; Ahmad, A.; Selamat, A.; Hazini, S. Water Feature Extraction and Change Detection Using Multitemporal Landsat Imagery. Remote Sens. 2014, 6, 4173–4189.
  5. Ogilvie, A.; Belaud, G.; Massuel, S.; Mulligan, M.; Le Goulven, P.; Calvez, R. Surface Water Monitoring in Small Water Bodies: Potential and Limits of Multi-Sensor Landsat Time Series. Hydrol. Earth Syst. Sci. 2018, 22, 4349–4380.
  6. Asmadin; Siregar, V.P.; Sofian, I.; Jaya, I.; Wijanarto, A.B. Feature Extraction of Coastal Surface Inundation via Water Index Algorithms Using Multispectral Satellite on North Jakarta. IOP Conf. Ser. Earth Environ. Sci. 2018, 176, 12032.
  7. Ireland, G.; Volpi, M.; Petropoulos, G.P. Examining the Capability of Supervised Machine Learning Classifiers in Extracting Flooded Areas from Landsat TM Imagery: A Case Study from a Mediterranean Flood. Remote Sens. 2015, 7, 3372–3399.
  8. Memon, A.A.; Muhammad, S.; Rahman, S.; Haq, M. Flood Monitoring and Damage Assessment Using Water Indices: A Case Study of Pakistan Flood-2012. Egypt. J. Remote Sens. Space Sci. 2015, 18, 99–106.
  9. Manfreda, S.; Ben Dor, E. Unmanned Aerial Systems for Monitoring Soil, Vegetation, and Riverine Environments, Earth Observation Series, 1st ed.; Elsevier: Amsterdam, The Netherlands, 2023; ISBN 9780323852838.
  10. Manfreda, S.; McCabe, M.F.; Miller, P.E.; Lucas, R.; Madrigal, V.P.; Mallinis, G.; Dor, E.B.; Helman, D.; Estes, L.; Ciraolo, G.; et al. On the Use of Unmanned Aerial Systems for Environmental Monitoring. Remote Sens. 2018, 10, 641.
  11. Chew, C.; Reager, J.T.; Small, E. CYGNSS Data Map Flood Inundation during the 2017 Atlantic Hurricane Season. Sci. Rep. 2018, 8, 9336.
  12. Wan, W.; Liu, B.; Zeng, Z.; Chen, X.; Wu, G.; Xu, L.; Chen, X.; Hong, Y. Using CYGNSS Data to Monitor China’s Flood Inundation during Typhoon and Extreme Precipitation Events in 2017. Remote Sens. 2019, 11, 854.
  13. Gerlein-Safdi, C.; Ruf, C.S. A CYGNSS-Based Algorithm for the Detection of Inland Waterbodies. Geophys. Res. Lett. 2019, 46, 12065–12072.
  14. Chew, C.; Small, E. Estimating Inundation Extent Using CYGNSS Data: A Conceptual Modeling Study. Remote Sens. Environ. 2020, 246, 111869.
  15. Rajabi, M.; Nahavandchi, H.; Hoseini, M. Evaluation of CYGNSS Observations for Flood Detection and Mapping during Sistan and Baluchestan Torrential Rain in 2020. Water 2020, 12, 2047.
  16. Zhang, S.; Ma, Z.; Li, Z.; Zhang, P.; Liu, Q.; Nan, Y.; Zhang, J.; Hu, S.; Feng, Y.; Zhao, H. Using CYGNSS Data to Map Flood Inundation during the 2021 Extreme Precipitation in Henan Province, China. Remote Sens. 2021, 13, 5181.
  17. Ruf, C.S.; Chew, C.; Lang, T.; Morris, M.G.; Nave, K.; Ridley, A.; Balasubramaniam, R. A New Paradigm in Earth Environmental Monitoring with the CYGNSS Small Satellite Constellation. Sci. Rep. 2018, 8, 8782.
  18. Boothroyd, R.J.; Nones, M.; Guerrero, M. Deriving Planform Morphology and Vegetation Coverage from Remote Sensing to Support River Management Applications. Front. Environ. Sci. 2021, 9, 657354.
  19. Henshaw, A.J.; Gurnell, A.M.; Bertoldi, W.; Drake, N.A. An Assessment of the Degree to Which Landsat TM Data Can Support the Assessment of Fluvial Dynamics, as Revealed by Changes in Vegetation Extent and Channel Position, along a Large River. Geomorphology 2013, 202, 74–85.
  20. Cavallo, C.; Papa, M.N.; Gargiulo, M.; Palau-Salvador, G.; Vezza, P.; Ruello, G. Continuous Monitoring of the Flooding Dynamics in the Albufera Wetland (Spain) by Landsat-8 and Sentinel-2 Datasets. Remote Sens. 2021, 13, 3525.
  21. Soomro, S.; Hu, C.; Boota, M.W.; Soomro, M.H.A.A.; Jian, S.; Zafar, Z.; Li, X. Mapping Flood Extend and Its Impact on Land Use/Land Cover and Settlements Variations: A Case Study of Layyah District, Punjab, Pakistan. Acta Geophys. 2021, 69, 2291–2304.
  22. Radice, A.; Rosatti, G.; Ballio, F.; Franzetti, S.; Mauri, M.; Spagnolatti, M.; Garegnani, G. Management of Flood Hazard via Hydro-Morphological River Modelling. The Case of the Mallero in Italian Alps. J. Flood Risk Manag. 2013, 6, 197–209.
  23. Bertoldi, W.; Drake, N.A.; Gurnell, A.M. Interactions between River Flows and Colonizing Vegetation on a Braided River: Exploring Spatial and Temporal Dynamics in Riparian Vegetation Cover Using Satellite Data. Earth Surf. Process. Landf. 2011, 36, 1474–1486.
  24. Gurnell, A.M. Trees, Wood and River Morphodynamics: Results from 15 Years Research on the Tagliamento River, Italy. In River Science: Research and Management for the 21st Century; Gilvear, D.J., Greenwood, M.T., Thoms, M.C., Wood, P.J., Eds.; John Wiley & Sons: Chichester, UK, 2016; pp. 132–155.
  25. Masocha, M.; Dube, T.; Makore, M.; Shekede, M.D.; Funani, J. Surface Water Bodies Mapping in Zimbabwe Using Landsat 8 OLI Multispectral Imagery: A Comparison of Multiple Water Indices. Phys. Chem. Earth Parts A/B/C 2018, 106, 63–67.
  26. Xu, H. Modification of Normalised Difference Water Index (NDWI) to Enhance Open Water Features in Remotely Sensed Imagery. Int. J. Remote Sens. 2006, 27, 3025–3033.
  27. Notti, D.; Giordan, D.; Caló, F.; Pepe, A.; Zucca, F.; Galve, J.P. Potential and Limitations of Open Satellite Data for Flood Mapping. Remote Sens. 2018, 10, 1673.
  28. Zhao, Q.; Yu, L.; Du, Z.; Peng, D.; Hao, P.; Zhang, Y.; Gong, P. An Overview of the Applications of Earth Observation Satellite Data: Impacts and Future Trends. Remote Sens. 2022, 14, 1863.
  29. U.S. Geological Survey (USGS) EarthExplorer. Available online: https://earthexplorer.usgs.gov/ (accessed on 4 February 2022).
  30. Sentinel Scientific Data Hub. Available online: https://scihub.copernicus.eu/ (accessed on 4 February 2022).
  31. Earthdata Search—NASA. Available online: https://search.earthdata.nasa.gov/search (accessed on 4 February 2022).
  32. LAADS DAAC. Available online: https://ladsweb.modaps.eosdis.nasa.gov/ (accessed on 4 February 2022).
  33. Rouse, J.W.; Haas, R.H.; Schell, J.A.; Deering, D.W. Monitoring Vegetation Systems in the Great Plains with ERTS (Earth Resources Technology Satellite). In Proceedings of the Third Earth Resources Technology Satellite Symposium, Greenbelt, ON, Canada, 10–14 December 1973; Volume SP-351, pp. 309–317.
  34. McFeeters, S.K. The Use of the Normalized Difference Water Index (NDWI) in the Delineation of Open Water Features. Int. J. Remote Sens. 1996, 17, 1425–1432.
  35. Gao, B. NDWI—A Normalized Difference Water Index for Remote Sensing of Vegetation Liquid Water from Space. Remote Sens. Environ. 1996, 58, 257–266.
  36. Feyisa, G.L.; Meilby, H.; Fensholt, R.; Proud, S.R. Automated Water Extraction Index: A New Technique for Surface Water Mapping Using Landsat Imagery. Remote Sens. Environ. 2014, 140, 23–35.
  37. Ji, L.; Zhang, L.; Wylie, B. Analysis of Dynamic Thresholds for the Normalized Difference Water Index. Photogramm. Eng. Remote Sens. 2009, 75, 1307–1317.
  38. Colditz, R.R.; Souza, C.T.; Vazquez, B.; Wickel, A.J.; Ressl, R. Analysis of Optimal Thresholds for Identification of Open Water Using MODIS-Derived Spectral Indices for Two Coastal Wetland Systems in Mexico. Int. J. Appl. Earth Obs. Geoinf. 2018, 70, 13–24.
  39. Shen, L.; Li, C. Water Body Extraction from Landsat ETM+ Imagery Using Adaboost Algorithm. In Proceedings of the 2010 18th International Conference on Geoinformatics, Beijing, China, 18–20 June 2010; pp. 1–4.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
This entry is offline, you can click here to edit this entry!
ScholarVision Creations
Feedback