Overview of remote sensing and flood mapping

Thirty meter resolution imaging (once considered high resolution, but currently considered moderate resolution) provided by optical sensors aboard Landsat satellites has been valuable to monitor seasonality and long-term changes in surface water over four decades [24]. With the launch of NASA’s Terra (1999) and Aqua (2002) satellites, daily tracking of wide areas (250 m resolution) of the Earth’s surface water has been made possible by the optical sensor MODIS (moderate-resolution imaging spectroradiometer) [6, 25–27]. Synthetic aperture radar sensors (SAR) such as those aboard Sentinel-1 satellites, operated by the Copernicus program, have the advantage of penetrating cloud cover [19, 27, 28]. Multi-satellite precipitation data products from NASA’s Global Precipitation Measurement Mission are used in flood detection models [29].

Identifications and monitoring of floods and long-term changes remains challenging, despite availability of data provided by satellites [24]. Landsat imagery is limited by a wide temporal coverage (8 days), however, this has been recently resolved by a new product that combines Landsat and Sentinel-2 images to derive daily surface water extent maps with global coverage [24, 28]. While MODIS sensors can monitor large floods daily, it cannot detect floods where there is persistent cloud cover, complex terrain such as urban environments, or smaller flash floods [6]. Sentinel-1’s SAR sensors are limited in flood detection capability in the presence of wind or vegetation as the contrast between flooded and non-flooded areas is reduced [19, 30, 31]. Due to the temporal cycle of Sentinel-1 satellites (2 to 3 days at best), SAR flood detection is more accurate for longer duration events rather than flash floods [19, 30, 31]. Several indices derived from satellite imagery, such as NDWI (Normalized Difference Water Index), WI (Water Index), RSWIR (Red and Short Wave Infra-Red), or GSWIR (Green and Short Wave Infra-Red), have been used for flood mapping, but each have limitations (e.g. tendency to classify cloud shadow as water) [32, 33]. NDWI, for example, tends to underestimate inundated areas while GSWIR, in contrast, tends to overestimate [32].

In recent years many websites and data archives containing geospatial flood data products (i.e. flood data portals) have come into existence, each with their own mission and purpose [6, 24, 29, 34–38]. These portals use various sources of information from satellite imagery, government, or media sources to provide information or generate usable products about floods. Some portals such as USDA’s Global Agricultural & Disaster Assessment System are interactive web map viewers that allow users to visualize floods without any special software [24, 38–40]. Some offer data products in different formats [25, 29, 37, 38, 41, 42]. The Dartmouth Flood Observatory, for example, provides a downloadable database of flood events [6, 25, 40]. Others, such as the Global Surface Water Explorer, provide high resolution map products of surface water changes [24].

Interdisciplinary example: Quantifying flood in studies examining the impact on nutrition and food security

Measurement of flood varies widely within the literature on nutrition and food security. One challenge is measuring flood at the community or household level. Another measurement challenge is distinguishing the type of flood–e.g. river, pluvial, coastal, or multiform floods–which can have different impacts [43, 44]. Many studies that have examined nutrition and food security impacts of floods have focused on single flood disasters, comparing pre/post or flooded and non-flood areas [45–56]. Some studies have measured flood through self-report using surveys, having the unique advantage of directly identifying flood-affected households but raising the risk of bias due to potential overreporting of flood exposure [46, 57, 58]. Others have used direct measures from satellite or other spatial data, aggregated at various levels [10, 59–61]. These studies typically do not differentiate the type of flood or main cause. Few studies in this body of literature have linked spatial data to household surveys. One recent study examined the effect of flooding on food security across broad regional, country, and all-continent scales in Africa and found both positive and negative impacts of floods and their related meteorological conditions. This study highlighted the important data gaps for assessment of flooding and food security outcomes across different spatial and temporal aggregations [44].

Linking spatial and survey data is challenging due to usability aspects. While flood data portals classify whether an area is flooded, with varying approaches, the differing spatial and temporal attributes make it difficult to link this information to surveys which include community, sub-district, or district level boundaries. In contexts where flooding is recurrent and commonplace, such a measure could deepen understanding of the impacts and how to respond. The heterogeneity in the literature makes it difficult to compare studies across contexts to determine, broadly, the human impacts of floods. But also, this heterogeneity makes it difficult to compare across studies within the same context. Currently there is no universal definition or classification system of flood for use in disaster relief efforts or research.

Goals of this paper

We share a usability framework for flood data portals for research on human outcomes that incorporates the end user perspective. To showcase this framework, we conducted a pilot study using NASA Worldview (NW), Global Surface Water Explorer (GSWE), Global Flood Awareness System (GloFAS), Global Flood Monitoring System (GFMS), and Dartmouth Flood Observatory (DFO). This framework can be adapted and aids researchers wanting to link spatial and survey data to study how natural disasters and environmental changes are impacting human outcomes, e.g., health, economics, migration, food security, or nutrition.

Flood data portals included in this pilot study

All flood data portals included in this review are publicly accessible. This is not an exhaustive list or systematic review of all flood monitoring or flood data portals in existence, of which there are many. We selected five commonly used flood data portals that had global coverage based on a literature review and exploratory search for data in consultation with an expert in the application of Remote Sensing and Geographic Information Systems. Our criteria were that the data were freely and publicly available. We included:

• NASA Worldview (NW): NW is a large satellite data archive with both an interactive, dynamic (spatiotemporal) map viewer and data repository offering download of multiple spatial data layers. A few of the many data products it offers are related to floods or surface water extent [37]. Hosted by the United States government’s National Aeronautics and Space Administration (NASA) and built by the Earth Science Data and Information System Project, NW includes satellite derived data from all aspects of the earth system and aims to support monitoring of natural hazards by providing access to current, daily satellite imagery data [37].
• Global Surface Water Explorer (GSWE): GSWE is an interactive, dynamic map viewer allowing visualization of long-term changes in the Earth’s surface water over 37 years [24, 42]. GSWE displays datasets comprising the Global Surface Water collection (since 1984) produced using Landsat satellite imagery. While these data are continuously being updated with no end date, the data products are made available periodically. Currently, the 1984–2021 dataset is available. Data are accessible via several download options provided through a separate data access page. The European Commission’s Joint Research Centre developed this portal and corresponding datasets under the Copernicus Program to support a range of applications such as water resource management, conservation, and food security [42]. While not specific to flood events, this portal provides information on the extent and change of surface water including development of new permanent water bodies, e.g. from seasonal to permanent inundated areas.
• Global Flood Awareness System (GloFAS): GloFAS provides an interactive map viewer to visualize current flood forecasts and detailed information about how to access the underlying data [35, 41]. Data are derived from satellite imagery and are accessible through the Copernicus Climate Data Store and can also be downloaded through the map viewer by creating an account. Implemented by the European Commission and Copernicus Program, the Global Flood Awareness System aims to support flood disaster preparedness efforts through global flood monitoring and forecasting of large river basins [41]. This is a component of the Copernicus Emergency Management Service.
• Global Flood Monitoring System (GFMS): GFMS is an experimental, quasi-global flood monitoring system using real-time, multi-satellite precipitation data to create a gridded flood model that users can view on 1/8^th degree resolution maps [29, 62]. GFMS uses TRMM (Tropical Rainfall Measuring Mission) multi-satellite precipitation analysis and global precipitation measurement (GPM) integrated multi-satellite retrievals for GPM (IMERG) to create the flood model which is based on land surface and river tracing routing models [29]. The maps allow users to change the different parameters and view time series of regional areas. The portal provides several variables that can be viewed, but the only data available for download is flood intensity (in depth) above threshold. This portal is developed and managed by the University of Maryland and funded by NASA.
• Dartmouth Flood Observatory (DFO): DFO is a resource of information about large flood events worldwide, provided through a range of dynamic maps and visualizations and a downloadable historical archive developed from media reports and other sources [25]. DFO originated at Dartmouth College in 1993 and then transferred to the University of Colorado in 2010 where it is hosted currently. DFO is often the first to map major flood events. Its mission is to support research and other humanitarian and water management applications.

Ethics statement

Ethical approval for the external user survey was obtained from the Tufts Social, Behavioral & Educational Research Institutional Review Board. The study met criteria for exempt status. All participants were informed of the study by reviewing a consent script on the first page of the webform and selecting “I consent to participate” to begin the survey.

Usability framework

Our framework incorporated data usability aspects as well as user feasibility. The data usability component included an in-depth review of the portals for data accessibility and availability, approaches to flood identification through the data sources and flood detection methods, and data quality through alignment, velocity, and variety (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Usability characteristics for flood data portals.

https://doi.org/10.1371/journal.pclm.0000511.t001

The user feasibility component included a survey of graduate students reviewing and rating each of the portals for several criteria. The survey captured participants’ opinions about the portals’ goals and scope, data and visualization quality, and platform accessibility from an external user perspective. We developed the survey using metrics adapted from a previous study which were based on principles of evidence, efficiency, emphasis, and ethics (Table 2) [20]. Under the evidence principle are metrics assessing the quality of a portal’s data products and whether the data is presented in line with the goals and objectives. Under the efficiency principle are metrics assessing the content organization and visual properties of a portal. Under the emphasis principle are metrics assessing a portal’s “user-friendliness” and ability to recognize the needs of a target audience and tailor the content to their perceptions, abilities, and interests. Finally, the ethics principle seeks to evaluate whether best practices are applied.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Assessment metrics for external user survey of flood data portals derived from principles of evidence, efficiency, emphasis, and ethics¹.

https://doi.org/10.1371/journal.pclm.0000511.t002

From the above principles, Zhou et al. designed metrics to assess functionality and operation features of nutrition dashboards, which communicate nutrition data through interactive visualizations (Table 2) [20]. While these metrics were designed mainly for the evaluation of visualization dashboards, they can be applied more broadly to other types of data platforms. The functionality and operation aspects aligned with our purposes related to the practicality of flood data portals and provided the opportunity and structure to develop a survey instrument to capture end-users’ opinions in a quantifiable way, which is critical for making recommendations to improve the portals. We developed the survey by reviewing the criteria from Zhou et al. and tailoring it to the application of flood data portals.

Pilot study

From February to March 2022, we explored the flood data portals in depth and completed a structured form containing the usability characteristics defined in Table 1 (see Appendix for data gathering form). The forms were completed with as much detail as possible that could be obtained from the websites and associated articles. We summarized the information in a table.

From May to September 2022, we recruited graduate students from Tufts University and Boston University to review each portal and complete a webform survey (see Appendix). For each portal, participants were asked to spend no more than 20 minutes exploring the portal and rating the assessment metrics using 5-point Likert responses. The survey was created and conducted using the Kobo Toolbox platform.

We used Stata 17 (StataCorp LLC, College Station, TX) for analysis of the external user assessment survey. Survey responses were recorded using a 5-point Likert scale defined as: Strongly Disagree, Disagree, Agree, Strongly Agree, Not clear/ Difficult to assess. For each of the 25 metrics in the survey, we calculated percent agreement (i.e. rated either ‘Strongly Agree’ or ‘Agree’) for each portal. Then, we assessed interrater agreement using a generalization of weighted Kappa allowing for the situation of more than 2 raters, more than 2 ratings, and an incomplete design. We calculated the Kappa statistic for each flood data portal and interpreted the level of agreement as 0 to .20 = slight, .21 to .40 = fair, .41 to .60 = moderate, .61 to .80 = substantial, and .81 to 1 = perfect. Survey respondents were also asked to leave notes about each of the portals they reviewed. We summarized common themes from these notes inductively.

Usability characteristics

Table 3 shows the suite of flood-related data products offered by the flood data portals. A comparative summary of the usability characteristics across the flood data portals is presented in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of flood-related data products provided by the portals in this pilot study.

https://doi.org/10.1371/journal.pclm.0000511.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Usability characteristics of flood-monitoring data portals in this pilot study.

https://doi.org/10.1371/journal.pclm.0000511.t004

Data accessibility and availability

All five portals in this review are in the public domain and free to users. All are currently maintained but it was unclear whether they receive long-term funding. The aims and missions varied and included flood monitoring (NW, GloFAS), flood forecasting (GloFAS), disaster preparedness (GloFAS), humanitarian (DFO), water-management decision-making and applications (DFO, GSWE). Only DFO specified a research purpose. The GFMS is an experimental system and did not specify an aim or mission.

All portals make their data products accessible directly from the site but NW and GloFAS require login. The portals offer 18 different flood-related data products (Table 3). Most provide information about flood occurrences from satellite imagery. For example, the MODIS NRT Global Flood Product from NW is a daily global map of flooding derived from Terra and Aqua satellite imagery. The collection of Global Surface Water datasets from GSWE capture changes in surface water occurrence at 30m resolution (previously considered high, but currently considered moderate resolution) from Landsat imagery since 1984. While not specifically measuring flood events, GSWE datasets can be used to identify flooded areas. GloFAS data products measure ongoing and upcoming river floods from Sentinel-1 imagery. While the other portals offer multiple data products, GFMS offers only 1, flood intensity (in depth), derived from multi-satellite precipitation data. DFO data provide information on occurrence of large flood events, compiled mostly from media sources.

In addition to data on flood occurrence, NW and GSWE capture recurrence and NW, GloFAS, GFMS, and DFO provide data on other characteristics of floods (Table 3). The Flood Hazard: Frequency and Distribution gridded data from NW measures frequency of floods between 1985–2003 using DFO data. The Recurrence dataset from GSWE measures inter-annual frequency of surface water between 1984 to present time, capturing areas that were episodically flooded. NW offers Mortality Risk and Economic Risk Flood Hazard layers, providing information on mortality and economic losses. Both are derived from the International Emergency Events Database (EM-DAT). GloFAS flood monitoring has data layers of the affected population and affected land use. GFMS provides data on flood depth. DFO provides the most information about floods in its Global Archive of flood events including main cause, area affected, detailed locations, names of rivers, duration, known number of deaths and displaced, and damages in USD. None of the portals’ data products distinguish or classify the type of flood, but DFO does provide the main cause and a measure of flood severity. GloFAS data only includes river floods.

Most of the data products are geospatial, provided in raster or vector data formats. Only DFO provides a non-geospatial format–an Excel sheet of flood events.

Approaches to flood identification and their potential errors

The portals had different approaches to flood identification and minimizing type 1 (false positive) or type 2 (false negative) errors by: compositing (NW), ensemble of different flood detection algorithms (GloFAS), classification over an extended time (GSWE), or combining land surface and river runoff models (GFMS).

NW classifies floods via the MODIS Near Real Time Global Flood Product and the Flood Hazard: Frequency and Distribution layer. The Global Flood Product is derived from the MODIS (Moderate Resolution Imaging Spectroradiometer) sensor datasets from both Terra and Aqua satellites. Flood identification is based on time compositing water detections over 3 periods (1, 2, and 3-day) and differentiating them from surface water using a reference mask [63]. If a water detection does not match the reference layer, it is classified as flood. The compositing approach minimizes false flood detections by combining multiple observations, thus reducing error from clouds. For each period, all observed water detections are combined, and pixels are classified as water if they exceed thresholds of 1, 2, or 3 observations based on the periods, respectively. While this approach minimizes error, the Global Flood Product still has limitations and false detections cannot be entirely removed, especially if cloud cover is persistent over multiple observations [63]. False-negative detections may result from obscuration from buildings, canopy, or clouds, while false-positives may still result from cloud or terrain shadow or dark rock or soil. Users are encouraged to compare flood maps to reflectance imagery to determine potential sources of false-positives or false-negatives [63]. Another potential source of error is the reference mask used to differentiate surface water from floods, which is outdated and planned for improvement [63].

The Flood Hazard: Frequency and Distribution layer from NW is derived from DFO data (discussed further below).

GSWE uses high resolution Landsat satellite imagery to classify changes in surface water from 1984–2021 [24]. This portal differs from the others in its goal to detect and classify surface water, not flood events, over multiple decades. Challenges include variations in measurement across the Landsat sensors (TM, ETM+, OLI) and in the conditions in which they’re observed [24]. False water detection may be caused by shadows (e.g. from buildings, terrain, or cloud), or lava flow. GSWE applies a combination of expert systems, visual analytics, and evidential reasoning techniques to classify pixels and prevent water classification errors [24]. This is followed by manual visual inspection and cleaning of the maps to remove false detections, and further validation to handle remaining errors [24]. An important feature of GSWE data is its classification over an extended period. Because classifications are made over time, the temporal behavior of the pixels is also used in determining their likelihood of being water, thus strengthening the accuracy of the classification.

GloFAS captures river flooding using Sentinel-1 SAR satellite data and a combination of 3 separate flood detection algorithms [64]. A final ensemble algorithm generates the final flood product based on a consensus of the 3 flood detection algorithms [64]. Like NW, a reference water mask is used to differentiate flooding from permanent surface water [64]. The ensemble approach helps prevent flood classification errors attributed to any one of the flood detection algorithms. However, classification errors are possible due to limitations inherent with SAR data. False positives can occur from land characteristics such as very dry soil, frozen ground, wet snow, or impervious surfaces or from radar shadowing [64]. False negatives can occur from urbanization or dense vegetation or strong, heavy winds and rainfall [64]. To address these classification challenges, GloFAS produces exclusion mask (for effects due to ground surface characteristics) and advisory flag (for effects due to weather conditions) layers along with their flood product layers [64]. Users can apply these layers to determine which pixels to disregard.

GFMS uses a real-time multi-satellite precipitation data product from NASA’s Global Precipitation Measurement Mission to identify floods using a dual land surface with river runoff model–the Dominant River Tracing-Routing Integrated with Variable Infiltration Capacity Environment (DRIVE) model [29]. Flood is defined when the routed runoff is greater than the flood threshold of that grid cell. Precipitation biases and the size and duration of flood events can impact GFMS flood detection capability [29]. False positives can occur in areas with dams or related to winter surface features in higher latitudes. False negatives can occur due to underestimation of orographic rainfall (due to hills or mountains). Generally, GFMS has improved flood detection performance with longer, larger floods.

Unlike the other portals, the DFO historical archive is based mainly on media sources (news reports, governmental and international relief agency websites). DFO captures large flood events spanning large geographic extents which are mostly hand drawn based on where the media reported flooding. In some instances, the archive might also have satellite observations to determine the flood extents. While it is a good source for identifying where large flood events occur globally, it is not useful for identifying which specific locations or communities are affected, making it prone to type 1 error. The differences in news coverage by country and region may impact the accuracy of information about a particular flood event.

Alignment, velocity, variety

We assessed data quality through alignment, velocity, and variety (Table 2). All portals provide data products structured by spatiotemporal attributes, but only DFO provides a dataset of flood events. All provide daily or Near Real Time data except for GSWE which classifies the whole timespan. DFO is the most important source of historical data about flood events from the 1980s. NW and GSWE also provide historical data products from the 1980s, but the products available through NW are derived from DFO data and EM-DAT. While GSWE has historical data, it captures changes over time and is not specific to flood events. GSWE and GloFAS provide the highest resolution data products, but only DFO provides detailed locations including names of rivers and towns.

All portals update their data products regularly. GFMS, NW, GloFAS, GFMS, and DFO (River and Reservoir Watch), provide daily updates. DFO may have some poor or missing data in the 1990s. The flood data products provided by NW cover the 1985–2003 and January 2022-present timespans, but there is a gap between 2003 to January 2022.

External user feasibility assessment

We present an exploratory descriptive and qualitative assessment of graduate student raters from the external user assessment survey. Eight raters participated in the survey, with 6 completing ratings for all 5 portals. Fig 1 shows the percent agreement. According to raters’ opinions, GSWE and GloFAS were ranked most favorably for their usability on average (73% average agreement), while DFO and NW were ranked least favorably (39% and 37%, respectively) and GFMS was in the middle (58%). Fig 1 shows the breakdown by the assessment metrics and areas where the portals can improve their usability. For goals and scope, raters agreed for GSWE, GloFAS, and GFMS, but did not seem to agree for DFO and NW. Ratings for data and visualization quality were mixed across the portals, with raters agreeing on different aspects for each portal. For platform accessibility, raters agreed for GSWE, but less so for the other portals. For contacts, communication, and conflicts of interest, raters agreed for GFMS, GloFAS, and GSWE and did not seem to agree for DFO and NW. Inter-rater reliability for each of the portals ranged from slight to fair. When examined across all 5 rating categories and across 2 rating categories (either agree or disagree), respectively, Kappa coefficients were 0.06 and 0.40 for GSWE, 0.03 and 0.16 for GloFAS, 0.25 and 0.40 for GFMS, 0.05 and 0.08 for DFO, and 0.08 and 0.11 for NW.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Percent agreement among raters of external user assessment of flood portals in this pilot study.

https://doi.org/10.1371/journal.pclm.0000511.g001

Notes provided useful feedback about the portals. For DFO, comments indicated the difficulty of navigating the website. E.g.,

“The website is not intuitive and difficult to navigate.”

“There is a wealth of useful data on this portal, however, the site is not user-friendly.”

For GFMS, the ease of navigating the portal was commonly described. E.g.,

“This website was easy to navigate.”

“This portal is well organized and provides easy to-use instructions.”

But improvements could be made to the visualization tools, e.g.:

“The visual presentation and design leaves somewhat to be desired—there are missing spaces and grammatical errors, and the formatting changes throughout the webpage.”

“Some data links were dead, and some visualization options resulted in blank white areas without any stated reason as to why data could not be produced.”

Comments about GloFAS were positive and indicated that it was clear and user-friendly, e.g.:

“This website is the most clearly laid out and the outputs are all described well.”

Comments about GSWE were mixed. Comments indicated the site was well-designed. One rater felt the data download process was difficult, while another commented on the difficulty of the graphs:

“You are able to explore the data but downloading it is a bit more complicated. The website is well designed though and easy to use!”

“The tool to explore global surface water was impressive… however, either I was not able to properly understand the graphs, or the graphs weren’t working properly.”

Finally, comments about NW were focused on difficulty of accessing and downloading flood-specific data, e.g.:

“While this tool is visually impressive, I couldn’t even find a way to view flooding or surface water after several minutes of exploration.”