Enhancing Flood Risk Management: A Comprehensive Review on Flood Early Warning Systems with Emphasis on Numerical Modeling

Abstract

During recent decades there has been an increase in extreme flood events and their intensity in most regions, mainly driven by climate change. Furthermore, these critical events are expected to intensify in the future. Therefore, the improvement of preparedness, mitigation, and adaptation counterparts is mandatory. Many scientific fields are involved in this task, but from a meteorological and hydrological perspective, one of the main tools that can contribute to mitigating the impact of floods is the development of Early Warning Systems. In this sense, this paper presents a scientific literature review of some of the most representative Flood Early Warning Systems worldwide, many of which are currently fully operational, with a special focus on the numerical modeling component when it is developed and integrated into the system. Thus, from basic to technically complex, and from basin or regional to continental or global scales of application, these systems have been reviewed. In this sense, a brief description of their main features, operational procedures, and implemented numerical models is also depicted. Additionally, a series of indications regarding the key aspects of the newly developed FEWSs, based on recent trends and advancements in FEWSs development found in the literature, are also summarized. Thus, this work aims to provide a literature review useful to scientists and engineers involved in flood analysis to improve and develop supporting tools to assist in the implementation of mitigation measures to reduce flood damage for people, goods, and ecosystems and to improve the community resilience.

1. Introduction

As many scientific publications are bringing to light, during the last few decades, the world has experienced one of the most intense flood-rich periods, mainly forced by climate change [1,2]. It seems to be clear that the number of extreme flood events, as well as their intensity, are currently increasing in most regions worldwide [3,4]. Therefore, as a direct consequence, it is expected that the number of people potentially affected by these events will increase in the future [5]. The floods registered in July 2021 in Germany and Belgium, where more than 150 people died as a direct consequence of these events [6]; the more recent flood registered in Peru caused by cyclone Yaku in March 2023 [7]; or the flood registered in northern India in April 2023 [8] show the devastation scenarios left by this type of event. Therefore, one of the most significant challenges that the scientific community has to address is the understanding of the mechanisms involved in the dynamics of such extreme events, as well as the anticipation of them. This is crucial to increase preparedness, improve flood mitigation and, essentially, adapt to these new scenarios derived from climate change [5,9]. There are many fields involved in these tasks. From meteorology to hydrology, encompassing soil, terrain, or climatic analysis, all these fields of study contribute and are necessary to achieving the aforementioned aims [10]. In this sense, Flood Early Warning Systems (FEWSs) are one of the most important tools that can contribute to fulfilling these objectives [11]. In turn, the integration of numerical models within them is essential for incorporating the main features of the areas of interest which play a key role in hydrological processes, thereby enabling a more precise and comprehensive delineation of floods. Thus, in response to a certain event of extreme precipitation, numerical models are capable of providing the expected river flow and some of them also are able to generate the associated flood maps. In the present work, a scientific literature review is provided on some of the most representative FEWSs worldwide, most of them currently operational, paying special attention to the utilization of numerical models in their operation. This synthesis of the existing knowledge in the field of FEWSs seeks to support researchers and practitioners in gaining a better understanding of the current state of the art; identify strengths, weaknesses, and the gaps that still exist in ongoing research; as well as become aware of the most important advancements in the field. Therefore, this comprehensive review of FEWSs, by synthesizing lessons learned from past and current implementations as well as best practices from different systems and regions, contributes towards knowledge sharing among the various actors involved in flood management, thereby enhancing collaborative development. Additionally, it aids in the dissemination of valuable insights, promoting standardized or common approaches to FEWSs implementation. Thus, the purpose is to incorporate FEWSs ranging from the most basic, usually developed in locations where scarce resources and data are available, to advanced systems, where multiple sources are implemented in hydrologic–hydraulic models, including real-time measurements of precipitation, river flow, and water levels as well as information on radars or satellites, among others. The most advanced FEWSs allow a complete and detailed analysis of the possible flood evolution from precipitation to flow dynamics. Regardless of the system and its level of complexity, the ultimate objective of all these systems is to forecast flood events with sufficient advanced notice. This can assist responsible authorities to act before the event occurs and mitigate flood damage. A brief description of the internal features, numerical models implemented, and operations and applications involved in these systems, along with other important related information such as documentation, access, or the location where they are implemented, are presented. Thus, as mentioned above, this review contributes to improving some important aspects related to FEWSs, such as identifying the best practices and technological trends in the design and implementation of FEWSs. It also contributes to establishing the lessons learned from the development and implementation of different FEWSs, including some gaps in current research. All of this helps in better understanding the successes and challenges, in order to adequately address future related projects. Also, the interdisciplinary nature of FEWSs (from meteorology to hydrology), along with the necessary cooperation of different institutions (from universities to governments), makes this type of review very useful and necessary in order to improve the global perspective on the implementation of these systems. Therefore, conducting this inventory of scientific publications on FEWSs can aid knowledge consolidation, thereby contributing to the future developments.

The paper is structured as follows. First, in Section 2, the methodology followed to perform the review is provided. In Section 3, a review of the most representative FEWSs is presented along with a description of their main components and features. In Section 4, a discussion focused on the advances and evolution of FEWSs is conducted, accompanied by the proposal of a series of recommendations regarding the key aspects to be included in the development of new FEWSs. Finally, in Section 5, a series of conclusions are briefly depicted.

2. Methodology

The main focus of this review work is to provide a comprehensive and complete basis encompassing the different and most representative types of existing FEWSs, with particular attention to those that make use of numerical modeling, although not exclusively. Consequently, given the proliferation of several FEWSs worldwide [12], selection criteria should be developed to ensure manageable article length and representativeness. Thus, the objective was to incorporate examples of FEWSs that span the spectrum, from those that include fewer data and components, thereby offering less detailed predictions (henceforth referred to as basic), to those that incorporate multiple sources of data and information, as well as various components, thereby offering more detailed predictions (henceforth referred to as advanced). Moreover, the objective also included covering scales ranging from basin or regional to continental or global. Additionally, there was an effort to ensure comprehensive global coverage when making the selection, including in the analysis examples of FEWSs developed in different regions worldwide.

Initially, a literature review of papers published in indexed journals was conducted to identify FEWSs whose methodology was published in scientific journals. The databases used to perform the literature search were SCOPUS (Elsevier) and Google Scholar. Then, the keywords “flood early warning system” were used in the search engines. Also, the search was restricted to publications published during the time frame of the last 15 years, from 2009 to the present day. From the obtained results, a selection process was performed for the inclusion in the article. The first criterion was the utilization of numerical modeling in the publication; for this an in-text search of the keywords “numerical modeling”, “numerical model”, “hydrological model”, “hydraulic model” or “hydrodynamic model” was performed. In order to reduce the geographical bias, examples of representative FEWSs from most continents were manually selected, considering that some areas may be under-represented in the literature. Then, the systems were classified depending on the scales they covered and their architecture. The final selection aimed to gather manageable set of FEWSs covering all the considered typologies. This selection was finally disclosed in this article, highlighting the general operation, the context, and the particularities of each FEWS. Finally, a discussion of the main trends, advances, and areas for improvement was developed. Note that a considerable number of FEWSs are internally developed in collaboration with various institutions and/or authorities responsible for river basin administration, primarily focusing on the functioning of the respective systems in their territories. In these cases, some of their developments and implementations may not be published in scientific journals since the priority is the management of the basin. To address this gap and provide a more comprehensive global perspective on FEWSs development, detailed and rigorous information available in publicly accessible reports and technical documentation was also analyzed and revised for some of the FEWSs. Once the documentation and information available for the selected FEWSs were identified, the most representative ones, following the aforementioned criteria, were incorporated into the review. We acknowledge that, despite the fact that the selection made provides broad coverage of the different types of FEWSs and offers a comprehensive global perspective, including implementations in different regions worldwide, some FEWSs exhibiting high similarities with those included were left out of this review to ensure the article was a manageable length.

Then, the main different types of FEWSs; their implementations; the description of their main characteristics, applications, and numerical models incorporated; as well as the methodologies applied are presented and described in the next section.

3. Flood Early Warning Systems

The UNDRR [13] defines Early Warning Systems as “an integrated system of hazard monitoring, forecasting and prediction, disaster risk assessment, communication and preparedness activities systems and processes that enable individuals, communities, governments, businesses and others to take timely actions to reduce disaster risks in advance of hazardous events”. In particular, the World Meteorological Organization (WMO) [14] defines flood forecasting and warning systems as “linkage between the basic structures” that “include provision of specific forecasts with magnitude and timing of rainfall, establishment of a network of hydrometric stations, operation of real-time flood forecasting model software and issuance of early flood warnings”. In this sense, Flood Early Warning Systems (FEWSs) are one of the most capable risk management tools and, nowadays, many countries count with operational FEWSs [12].

Some studies estimate that the number of operational FEWSs doubled in the first two decades of this century [12]. This substantially contributed to diminishing the damage associated with floods [15,16]. In fact, in spite of the negative effects associated to climate change, with an intensification of floods in many regions worldwide, a declining trend was detected for flood disasters and mortalities, as well as the individuals affected, in the last decades [12,17,18]. The decrease in populations affected by floods is partially associated with the continuous improvement in flood mitigation due to FEWSs developments [12,17,18].

In general, a comprehensive and complete FEWS is based on four components (see Figure 1): (i) flood risk knowledge; (ii) detection, monitoring, analysis, and forecasting of hazards; (iii) warning dissemination and communication; and (iv) preparedness to respond [12,19]. In this sense, most of the scientific publications related to FEWSs are focused on developing and addressing the first two components [16,20,21]. In general, adequate communication of the flood warning and the proper response to the risk tend to be less developed. This is mainly due to the fact that the implementation of these components requires collaboration among many different stakeholders, including local authorities, river management authorities, communities, etc. [12,22].

However, the level of development of each of these integrated parts is highly variable. In a recent study [12], a worldwide survey was conducted in order to establish the level of development of FEWSs. Although there is no information on all the countries, some conclusions can be drawn. In general, developing countries have been progressing well in FEWSs implementation in recent years and, in fact, most of them present FEWSs with similar technical capabilities compared to developed countries. This was possible, in part, due to international aid [12,15,16]. However, as highlighted by Perera et al. [12], a challenge that still needs to be addressed in some developing countries is to provide sufficient training to local staff that allows them to maintain, or even improve, FEWS capabilities for long-term operations. In spite of the fact that FEWSs of developing countries present similar technical capabilities to those of developed countries, the casualties and economic losses are higher in the former, which indicate that the components of FEWSs related to communication and timely respond to flood risk should be improved in these countries [16,23]. Even in some developed countries, the part of the FEWSs related to the communication and respond to alerts should be reinforced [12]. On the contrary, the least-developed countries are left behind because many of them only have FEWSs with very limited capabilities [12,24].

Thus, focusing on the development of the technical part, where there is much more literature, the FEWSs can range from basic approaches, where the limited data available are usually obtained from rudimentary stations complemented with qualitative information based on people observations, to advanced systems, which include a large amount of information and data from automated stations as well as remote sensing data; this is complemented by hydrological–hydraulic models that provide detailed maps with the flood forecast, essential for a good response from the respective authorities [12,16,20]. As commented above, the integration of numerical models into FEWSs is crucial for providing more accurate and comprehensive flood information. The numerical models used in these systems can be categorized into three types: lumped, semi-distributed, and distributed models (see Figure 2). Lumped models do not explicitly consider the variations in topology but instead use averaged characteristics [25]. Thus, lumped models use mathematical representations to simulate the hydrological processes of a watershed by simplifying it into a single conceptual unit, with average variables representing the mean hydrologic characteristics of the basin [26,27]. This makes them computationally efficient, providing the river flow associated with different assimilated conditions (precipitation, evapotranspiration, etc.). However, the provided information is quite limited since they assume a homogeneous hydrological behavior of the basin and, therefore, do not capture its spatial variability [28]. A refinement of these kinds of models are the semi-distributed models, which divide the watershed into multiple sub-basins. Here, each sub-basin is also treated as a distinct unit, considering its average characteristics and mathematically resolving the hydrological procedure in a similar manner to the lumped models. However, dividing the basin into sub-basins, normally taking into account their topographic characteristics, allows to better capture the heterogeneity of the basin. In addition, a manageable level of complexity is maintained with high computationally efficiency. Lumped and semi-distributed models were mostly used traditionally in FEWSs because they were able to perform simulations at the catchment scale in a reasonable time frame, thereby offering a suitable response for the purpose of these systems [29]. Some popular models in this category include HEC-HMS [30,31], SWAT [32], or HBV [33]. However, these models only carry out the hydrologic procedure, transforming rain into the corresponding runoff and providing only river flow information, not the specific areas that can be flooded. To solve these constrains, static flood maps associated with different river flow thresholds could be used to approximate the flooded areas associated with the predicted river flow [34]. One of the main advancements recently included in FEWSs is the integration of distributed numerical models. The distributed numerical models divide the basin into numerous computational cells where the mathematical equations that characterize one or more natural processes are solved, given an approximation of the real behavior of a system [35]. Thus, distributed models explicitly consider the entire topology of the basin, using discretization in the form of a mesh. Usually, two-dimensional (2D) structured or unstructured meshes are used for this purpose, although 1D approaches like HEC-RAS [36], which only takes the river cross sections into account, are also popular due its computational efficiency. In any case, the resolution of the equations in each mesh cell can be a computationally expensive process, which can present a barrier in certain applications that require a rapid response, as is the case for FEWSs [37]. However, the progressive development of computational capabilities has allowed the utilization of these methods in more applications and with a higher resolution, including FEWSs [20,38,39,40,41]. In particular, distributed numerical modelling constitutes a useful tool for flood analysis, since it can provide the extent, depth, velocity, and arriving time, among others, of a flood given a forecasted river flow [20,42]. In addition, using these techniques, it is possible to study multiple flooding scenarios and obtain data that are difficult to measure in real-world scenarios and they can be useful for hazard assessment. As commented above, distributed models are computationally more expensive than lumped or semi-distributed models, although their cost depends on the resolution of the mesh. However, these numerical tools have improved using HPC (High Performance Computing) techniques, which facilitate their incorporation as a component of FEWSs. Some codes, such as Iber [43], have been improved recently using such techniques [38,44], obtaining an increase in the velocity of simulation near 100 times. There are many other examples of distributed 2D models in the literature, such as JFLOW [45], MIKE 21 [46], G-FLOW [47], or LISFLOOD-FP [48], among others.

In the following lines, some examples of different FEWSs, developed around the word, are provided to offer a general view of the different existing types, developments, methodologies, and implementations. The main characteristics of the analyzed FEWSs are shown in

Table 1. General information about the FEWSs incorporated in this work.

ID Number	Name	Scale	Location	Reference
1		Regional	Red Sea Mountains (Egypt)	[49]
2		Regional	Inner Niger Delta (Maly)	[16]
3		Regional	Kaijuri Union (Bangladesh)	[50]
4	MERLIN	Regional	Galicia Region (Spain)	[51]
5	MIDAS	Regional	Galicia Region (Spain)	[20]
6		Regional	Flanders Region (Belgium)	[16]
7		National	Australia	[52,53,54]
8	HEFS	National	USA	[52,55]
9		National	Brazil	[56,57]
10		National	India	[58]
11	EFAS	Continental	Europe	[59]
12	GloFAS	Global	Worldwide	[60]
13	HYDRATE	Continental	Europe	[61]
14	FLASH	National	USA	[62]
15		Regional	China	[63,64]
16		Regional	Barranquilla (Colombia)	[65,66]

and the locations where these systems are implemented are shown in Figure 3.

3.1. FEWSs Developed at Regional (River Basin) Level: From Basic to Advanced Systems

With respect to basic systems, especially applicable in areas where data are scarce, the local FEWS developed in Egypt (Red Sea Mountains, the Waidi Watier) was one of the first initiatives in North Africa and the Middle East [16,49]. The development of this FEWS was constrained by the limited availability of field data [49], proving a good example of how, even with limited resources, a FEWS can be developed to assist in flood prevention and mitigation. Thus, this FEWS was developed based on the combination of some quantitative information complemented with qualitative data based on the opinions of different experts and the knowledge provided by the local stakeholders [49]. Rainfall measurements collected using manual rain gauges, rainfall measurements of some digital rain gauges (although only for the most recent period), and satellite information of precipitation derived from the Tropical Rainfall Measuring Mission (TRMM) and the Global Precipitation Climatology Center (GPCC) were used as quantitative data for this FEWS. Precipitation forecasts generated by the Weather Research and Forecasting (WRF) model assimilating initial and lateral boundary conditions from the Global Forecast System (GFS) were also taken into account [49]. A discrete event lumped rainfall–runoff model was developed, linked to a hydraulic model. An important limitation of the model is the scarcity of reference data, usually obtained from historical river events registered in the area of study. This limitation prevented adequate calibration of the system. Thus, model parametrization was based only on expert knowledge and literature information, as well as on visual observations made by local experts or available videos and photos. Flood volumes based on some peak discharge estimations from the Water Resources Research Institute (WRRI) in Cairo were also used. In spite of the commented limitations, the authors showed that the implemented FEWS was able to forecast, with acceptable accuracy, some of the historical floods [49]. However, as mentioned by the authors, some improvements, in several aspects, are necessary. On the one hand, increasing the availability of field data to enhance the parametrization of the models employed in the FEWS. On the other hand, improving the institutional capacity to operate this system, along with better communication to the authorities responsible for implementing mitigation measures, as well as the development of emergency procedures. More detailed information about this FEWS is provided in [49].

Another example of FEWS development under limited and simple available data is the case of the local FEWS developed for the Inner Niger Delta (Mali). This case, which presents some problems similar to the previous one in terms of data scarcity, has other conditions that can be exploited to develop a useful FEWS for the region, especially for the local community affected [16]. In this case, the floods of the Inner Niger Delta are key to the local economy, since fishing and agriculture are correlated with flood intensity [67]; therefore, the anticipation of these events by means of a FEWS can help to improve the socio-economic development of the area. In this case, rainfall forecasts have low influence on floods in the Inner Niger Delta, since the river level is mainly affected by what occurs upstream, in more than 4000 km of the river’s length [16,68]. Thus, accurate knowledge of the upstream levels can be applied to provide an adequate flood forecast of the Inner Niger Delta [16]. At present, the system has been improved and it is provided via a website where the information can be consulted (https://www.opidin.org/en, accessed on 11 April 2024). This shows that, in this case, the system was not only technically developed according to the resources and necessities of the area under scope but also presents a high degree of dissemination and communication, especially focused on informing the potentially affected people. Currently, the forecasting of floods in the Inner Niger Delta does not employ any numerical model but is instead based on the calculations of a set of curvilinear regression equations that relate the daily measurements of the water level in hydrometric stations upstream to floods in the Inner Niger Delta. Measurements of upstream rainfall during the preceding weeks are also considered to increase the prediction window.

Fakhruddin et al. [50] discussed the regional scale FEWS implemented in the Kaijuri Union (Bangladesh). This system, more complex than the previous ones, performs medium-range forecasts focused on the needs of the farmers, giving crucial information for decision-making [50]. This supposes an effective measure to reduce the economic impacts of floods in the agricultural sector. The system is based on the Hopson and Webster [69] methodology. The input data sources used by the FEWS are the ensemble forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), precipitation observations from several sources (including satellite and rain gauges), and river discharge data provided by the Flood Forecasting and Warning Center (FFWC). Lumped and semi-distributed hydrological models are used to perform multi-model simulations, providing 10-day-ahead discharge forecasts. In this way, for each point and each day, 51 sets of ensemble forecasts are generated [50]. Using these forecasts, along with field data and community interviews, probabilistic flood risk maps are developed following a risk assessment framework. However, the authors highlight the difficulty of the end-users to understand the non-deterministic data. The interviews performed also show that many farmers do not completely trust the forecasts and tend to rely in their own experience instead. The authors show the necessity of teaching the end-users to understand the provided flood information given the ability of the system to effectively predict extreme events along a 10-day horizon [50].

More complete FEWSs have been developed in other areas, where developers have taken advantage of the large amount of data available, which are normally used to feed hydrological and hydraulic models that allow the provision of detailed flood forecast maps, helping in flood response and mitigation. This level of data also allows for in-depth calibration and validation of applied models to produce reliable numerical simulations [70]. A good example of an advanced FEWS that use the numerous resources available is provided by [51], where the authors present a local FEWS, called MERLIN, developed for river basins of the Galician coast (NW Spain). In this case, authors take advantage of the high-density network of in situ meteorological and hydrological stations available in Galicia, whose data are provided, respectively, by MeteoGalicia (the Regional Meteorological Agency of Galicia) and Augas de Galicia (Hydraulic Administration of Galicia) [51]. The system also integrates other information such as the soil moisture data from the Soil Moisture Active Passive (SMAP) satellite provided by the National Snow and Ice Data Center. Thus, the hydrological procedure, executed using the HEC-HMS hydrological model, is carried out by assimilating a high number of variables during the hindcast stage (soil moisture, river discharge, rainfall, wind speed, air moisture, and solar radiation), in order to provide the most accurate representation of the reality on the date when the forecast is performed. During the forecasting stage, the outputs of the Weather Research and Forecasting (WRF) model provided by MeteoGalicia, including rainfall, temperature, air humidity, solar radiation, and wind speed, are assimilated for the hydrological model and the river flows are forecasted for the next 96 h [51]. Then, the obtained river flow hydrographs are used as inputs in the freely available and large tested 2D Iber+ hydraulic model [38], in order to predict the possible flooded areas in important vulnerable locations. In this sense, and taking into account that some villages under interest are coastal, the predicted tidal levels forecasted by the Regional Ocean Modelling System (ROMS), operated by MeteoGalicia, are also integrated into the hydraulic model [51]. Every day the system is restarted to perform the most recent forecast, providing detailed maps of the predicted floods in vulnerable villages. The system is operative and provides relevant information to decision-makers to implement adequate measures to minimize the flood impacts.

Another local FEWS, called MIDAS, was developed for the Miño-Sil river basin (NW Spain), where the available network data are similar to the above for the rivers of the Galician coast, but the concept of the FEWS is different [20]. In this case, the FEWS was developed by also integrating hydrological–hydraulic models but using procedures that only require a few number of parameters that can be obtained from global databases. The authors took advantage of the availability of data but only to calibrate and validate the system, since the purpose was the simplification of parameters so that the system can be extrapolated and applied in locations where data are scarce, thus facilitating the transfer of the developed methodology [20]. The semi-distributed hydrological model used to transform the rainfall in the corresponding river flow is the HEC-HMS [30,31]. In turn, the methodology selected to carry out the infiltration process is the curve number method [71], although adapted to run in a continuous mode. This procedure only needs the knowledge of one parameter, the curve number, which characterize the terrain features and for which there are global databases such as the one developed by [72]. Thus, the only variables needed as inputs for the system are the previous and the forecasted precipitations. Although the system is fed by the data provided by MeteoGalicia, the required precipitation data can be obtained from global databases for its application in other worldwide areas. In this sense, previous precipitation can be obtained from global datasets, some of them mentioned above, if no measurement stations are available in the area under scope, and the forecasted precipitation can be also obtained from global databases, such as the Global Forecast System (GFS). The system also uses information on previous river flows, provided by the Miño-Sil river basin authority (Confederación Hidrográfica del Miño-Sil, https://www.chms.es, accessed on 11 April 2024), to establish initial conditions more precisely. However, the river flow from previous days can also be calculated using the hydrologic system itself, attending to previous precipitation, if no measurement data are available. Some additional required parameters, such as lag times or baseflow coefficients, can be obtained via calibration, while others, such as river routing coefficients, can be obtained from Digital Elevation Models (DEMs). In spite of the simplicity of this method, it provides good results, especially when simulates extreme events, and, therefore, it is especially suited for FEWSs [20,21,73,74]. Then, once the river flow prediction is available, the generated hydrographs also feed the Iber+ hydraulic model in order to calculate the extent of the flood, along with important characteristic such as water depth or velocity, caused by the forecasted river discharge in vulnerable locations [20]. Although the information and parameters required to launch the hydraulic model implemented in MIDAS, such as the manning coefficients associated with land uses or the terrain topography, are supplied by the Spanish government and are only available for Spain, they can be obtained from global datasets. In this first case, certain databases, such as CORINE Land Cover [75], offer information on land uses. In the second case, several freely available DEMs with global coverage are available, such as, for example, ASTER-DEM (downloadable from https://earthexplorer.usgs.gov/), SRTM-DEM (downloadable from https://earthexplorer.usgs.gov/), or Copernicus DEM (downloadable from https://land.copernicus.eu/imagery-in-situ/eu-dem/eu-dem-v1.1?tab=download). In this sense, it is important to highlight that the use of accurate DEMs is highly important for determining the water dynamics over the surface. The FEWS designed was tested using the numerous available data, including the river flow evolution and also the levels reached by water in different points of vulnerable villages. The developed system showed an accurate forecast of real flows and also an efficient detection of flooded areas well in advance [20]. This is an example of a simple FEWS with accurate functioning that can be fed with freely and globally available data and, therefore, could be applied and implemented in basins with scarce data.

MERLIN and MIDAS represent examples of FEWSs developed for nearby areas with similar data availability (NW Spain) but by applying different approaches, since each one was developed for different purposes. While the former FEWS was constructed by taking advantage of all the information available for the area under scope (MERLIN), the latter (MIDAS) was focused on testing the efficiency of a less complex system that could be applicable in other areas with scarce data availability. Both systems show a high capability to detect floods far enough in advance, demonstrating that different FEWSs implementations can be equally valid for predicting floods and, therefore, helping to mitigating them.

Another example of a complete FEWS using all the available information is located in the Flanders Region, in Belgium (waterinfo.be, accessed on 11 April 2024). In this case, the hydrologic–hydraulic models are fed with ensemble weather forecasting data together with rainfall radar data, resulting in river flow forecasts as well as the predicted flood hazard maps [16]. The FEWS offers forecasts in two temporal windows of 48 h and 10 days, respectively. In addition, an extensive network of rainfall and water stage gauges is available, which allows for complete validation of the results, as well as increasingly efficient calibrations, with consequent permanent improvement in the system. The information provided by the system is available via a public web portal (waterinfo.be, accessed on 11 April 2024) and is continuously transferred to the authorities responsible for the response to flood events in order to improve the mitigation of its impacts. This persistent interchange of information also allows improving the flood forecast since field information of actual flood development is also provided to the FEWS operators [16].

3.2. FEWSs Developed at the Country Level

In this section, some interesting examples of FEWSs on a national scale, that is, covering entire countries, are explored.

The FEWS developed by the Bureau of Meteorology (BoM) of Australia provides flood forecasts for the entire country [52,53]. This service, operational for several decades, evolves and improves its technology and forecasts, continually adapting to the new implementations and developments [52,53]. The system performs the rainfall–runoff procedure using measured real-time data from a dense network operated by the Bureau of Meteorology and partner organizations (water management agencies and local councils), along with precipitation forecasts from the numerical weather predictions (NWPs) of the Bureau of Meteorology’s ACCESS (Australian Community Climate and Earth-System Simulator) system [52,54]. This information is used as an input in semi-distributed hydrological event-based and continuous models [52,54]. In this case, hydrologists can also force models with fixed “what if” scenarios to simulate different possible situations. Another difference of this system in comparison to other FEWSs is that, while most of the FEWSs are entirely operationally automated, this system, although also largely automated, involves additional interpretations of river flow and flood forecasts made by expert meteorologists and hydrologists, in order to provide more useful information to the population [52,53,54]. Thus, forecasters can perform subjective data assimilation, adjusting some model parameters to better fit the observed hydrographs, and can also produce additional warnings in real-time based on their previous experiences and by attending to some other facts such as, for example, the filling level of reservoirs in the affected area, thereby more actively participating in the process. The flood forecasts are disseminated to the interested users via several media platforms (web interface, email, fax, telephone, etc.) [52]. In addition, detailed national arrangements are also developed to enable a better response to the floods [53].

The United States National Weather Service also operates a FEWS covering the entirety of the continental United States, the Hydrologic Ensemble Forecasting Service (HEFS), which runs at several regional River Forecast Centers (RFCs) that forecast river flow for the corresponding locations [52]. The system provides uncertainty-quantified forecasts, especially useful for deriving the probabilities that an extreme event may occur [55]. The system procedure can be summarized as follows: First, seamless and bias-corrected weather forecast ensembles are produced from data provided from several weather forecasts from various sources, assimilated to address the uncertainty inherent to these predictions. Then, the weather forecast ensembles generated are used as inputs to run a suite of hydrologic, hydraulic, and reservoir models, to produce the corresponding ensembles of river flows and flood forecasts. These final forecasts are verified and a graphic generation is applied to produce visualizations of this information, which can then be communicated to different users to assist in decision-making and warning dissemination [52]. In this sense, some forecast data obtained are available via a public web interface (https://water.weather.gov/ahps/forecasts.php, accessed on 11 April 2024). The final forecast products include the expected values according to ensemble distribution (minimum, maximum, average, deviation, etc.), or the probability of exceeding critical thresholds, among others [52,55].

There are some other examples of FEWSs developed at the country level for different purposes. An example includes the FEWSs developed in Brazil. In this case, instead of a common system for the entire territory, particular FEWSs were developed on a regional scale for some locations vulnerable to floods, each with its own characteristics adapted to the vulnerable area under scope [12,56,57]. In general, Brazilian FEWSs use lumped and semi-distributed hydrologic models that are fed with the measured precipitation of the days preceding the forecast period, obtained from in situ stations and satellite technology. Hydrological models also assimilate forecasted rainfall from numerical weather predictions to obtain the river flow forecasts [56,57]. Thus, forecasted river flow hydrographs are obtained with lead times up to 14 days, and the information is made available to the potential users, especially to the authorities responsible for flood prevention [57]. The system is focused not only on flood forecasting to mitigate these events but also on facilitating hydropower reservoir operations [56,57].

In India, a country heavily affected by floods every year, Nanditha and Mishra [58] discuss the status of this country as a matter of FEWSs implementation. The Central Water Commission (CWC) of India provides deterministic flood forecasts 3 days ahead using different methods, including hydrological and hydraulic modelling tools. Here, the authors highlight that, although the current system exhibits a high accuracy, the sources of uncertainty present in deterministic weather forecasts can translate into the flood forecasts [58]. This is especially critical in areas prone to flash floods where relatively small inaccuracies can lead to very different flood developments. Making use of the available ensemble weather forecast information can help to measure and put into context those uncertainties, identifying plausible scenarios and their probability [58]. Authors also identify the dams as key actors in flood management, as they alter the natural river flow, having the capacity to mitigate the floods downstream. However, inadequate dam operation can also worsen flood effects. Therefore, dam inflow and weather forecasts are fundamental for proper decision-making and reservoir operation in flood management. Inaccuracies in inflow and weather forecasting were observed as a factor in worsening the flood impacts due to inadequate reservoir operations [76]. To overcome these limitations and address the recurring floods in India, the authors proposed a hierarchical FEWS based on ensemble precipitation forecasts, in which different forecast centers work at different scales [58]. The Central Water Commission (SWC) would coordinate the operations at the state level. In the following level, the Catchment-scale Forecast Centers (CFCs) and the coordinators of specific catchments will operate. In the last level, the diverse Regional Forecast Centers (RFCs) will operate at the sub-catchment scale, for specific urban areas and flash-flood-prone zones. The proposed system includes coordination between the different centers and the creation of guidelines for optimal reservoir operation; meanwhile, the RFCs would operate high-resolution hydraulic models for detailed flood damage assessment.

3.3. FEWSs Developed on a Continental and Global Scale

It is also especially remarkable that significant efforts have been made to develop FEWSs at larger spatial scales. One of the main motivations behind developing these systems is the recognition that important floods occur on a multi-country scale, typically resulting in more important and extensive damages [77]. Also, it is important to take into account that the overflow produced in upstream catchments, belonging to different countries, can have a critical effect downstream, thereby affecting other countries [77,78]. Thus, when floods occur across borders, they are usually managed by different authorities, and this can lead to a lack of or incomplete communication [77]. It is also possible that the different responsible authorities can assume different forecasting results since each one can have their own local FEWS for their area under scope. This can provoke situations where the coordination is inconsistent, making it difficult to respond to the risk [77]. Under this scope, some FEWSs have been developed at the continental scale—the European Flood Alert System (EFAS)—and even at the global scale—the Global Flood Awareness System (GloFAS). In the following text, some of the fundamentals upon which these systems are based are presented.

The EFAS project, developed by the European Commission’s Joint Research Centre (JRC), in collaboration with national hydrological and meteorological services, aims to provide early warning for potential flood events covering the entire European continent, particularly focused on large tans-national river basins [59]. The EFAS is based on the Lisflood hydrological model [79], especially adaptable to using the data available from the European Commission Joint Research Centre (JRC), which provides information describing soil, land use, topography, and river network, among others, as an input [59]. The historic and real-time observed meteorological data collected by the JRC, and also those provided by several meteorological and hydrological national services and river basin authorities, are integrated into the system as inputs in the hydrological model. In this sense, historic information is used to calibrate the model and real-time information is utilized to establish the initial conditions of the forecasts. In the latter case, the deterministic and probabilistic forecasts provided by the ECMWF are used as inputs for the forecasting stage, together with forecasts provided by national weather services [59]. Finally, the system provides forecast hydrographs from which flood warning information can be obtained for Europe up to 10 days in advance. The flood alerts emitted by the system are related to critical values of exceedance. These thresholds are calculated based on long discharge historical time series simulated at each grid point using the Lisflood model (using the same parametrizations as in the operational forecast system) fed with observed historical meteorological information [59,80]. These historical simulated discharges are statistically evaluated to determine thresholds that delineate different levels (coded using different colors), which indicate the flood possibility and its potential intensity [59]. When the predicted flows of the deterministic forecast and a certain number of ensemble forecast members surpass critical thresholds persistently over several forecasts, an alert is issued and the respective national authorities are informed in detail [52,77] (see detailed criteria of flood risk notifications in https://www.efas.eu/en/efas-notifications, accessed on 11 April 2024). In addition, although some information is only provided to EFAS partners (real-time forecasts and products), most of the information provided by the system, including past forecasts, is accessible via a public web portal (www.efas.eu, accessed on 11 April 2024). In this sense, EFAS developers stressed the need for a well-designed and accessible platform to communicate and disseminate the results on different levels to improve the usefulness and effectiveness of the system [59]. This approach is not as detailed as that provided by the hydraulic models mentioned above, which can only be applied to specific areas due to their high computational cost. Despite this, they provide useful information about the predicted flood risk in trans-national rivers at the continental scale. EFAS has been fully operational since 2012.

The GloFAS is a probabilistic FEWS that was developed to provide global flood forecast information regarding large river basins [60]. It produces daily flood probability forecasts over the proceeding 30 days and has been operational since 2011. Initially, in the first versions, the system assimilated the ensemble data provided by the land surface component of the ECMWF. This component applies the distributed Hydrology Tiles ECMWF Scheme for Surface Exchange over Land (HTESSEL) to compute the land surface response to different atmospheric conditions and to carry on the transformation of precipitation into runoff, obtaining surface and sub-surface runoff [60]. HTESSEL was fed by the ensemble atmospheric forecasts obtained from ECMWF, whereas ERA-Interim meteorological data were used to establish the initial conditions of the forecasts. Then, GloFAS used the Lisflood hydrological model but only for routing and groundwater river processes [52,60]. Thus, ensemble streamflow forecasts are obtained operationally, providing global coverage of river flow predictions [60]. In addition, the system is also fed with a long-term dataset to achieve a reference climatology that is used to establish the alert thresholds. For this purpose, the Lisflood model was fed with historical ERA-Interim/land runoff data (HTESSEL forced by using ERA-Interim meteorological data) [60]. Finally, the predicted river flows for each ensemble member at daily scale are compared with reference streamflow climatology to detect the probability of exceedance of critical flow thresholds, associated with return periods, with the issue of different levels of alerts [60]. GloFAS has registered different types of users, from governmental institutions, other public authorities, and non-governmental organizations to academic and research institutions and the private sector [52].

Both EFAS and GloFAS operational systems are constantly being evaluated and updated, in order to improve their efficiency. This includes assimilating more information and with increased accuracy, as well as enhancements in the temporal resolution of forecasts (from daily to 6 hourly), among other improvements. For instance, currently, GloFAS uses the Lisflood model as the main hydrological model, calculating the complete water balance from meteorological forcing data (HTESSEL was used until version 2.2) (https://confluence.ecmwf.int/display/CEMS/GloFAS, accessed on 11 April 2024). In the most current versions, ERA5 reanalysis is used as a proxy of meteorological observation information instead of ERA-Interim (the near-real-time version of ERA5 is specifically used to define the initial conditions of forecasts) (https://confluence.ecmwf.int/display/CEMS/Global+Flood+Awareness+System, accessed on 11 April 2024) [81].

Additionally, GloFAS also currently provides static inundation extent maps associated with different return periods in order to estimate the possible flood inundation areas that a certain forecasted event can cause. EFAS also provides map information about the risk of flooding in the latest versions (https://www.efas.eu/en, accessed on 11 April 2024). While these maps are not real-time and lack the precision of the maps provided by the hydraulic models implemented in some of the local FEWSs presented above, they represent a step forward in these more global simulation forecasts compared to the first versions.

3.4. FEWSs Focused on Flash Floods

There are some examples of FEWSs specifically focused on flash flood events. These floods are characterized by extremely rapid development velocities, reducing the time available to take measures and mitigate their impact. Unlike river floods, which usually develop gradually and cover larger areas, flash floods are characterized by their sudden onset and localized impact, making them particularly dangerous and more challenging to predict and mitigate. Some examples of FEWSs focused on flash floods are detailed below.

The HYDRATE (Hydrometeorological Data Resources and Technology for Effective Flash Flood Forecasting) project was devised with the aim to develop a common strategy across Europe to develop FEWSs focused on flash floods with the capability to adequately forecast far enough in advance these events [61]. The idea was to develop a common framework that could be applied in basins with an important quality and quantity of data and then take advantage of the information provided in these well-gauged basins to extrapolate a validated methodology to ungauged basins [61]. This can solve, at least partially, one of the main constraints of flash flood forecasting, that is, the small spatial and temporal scales of these events that prevent the accurate monitoring of rainfall and streamflow. Therefore, the project included actions to collect flash flood data across Europe by combining hydrometeorological monitoring and complementary information obtained from post-event surveys [61]. In this sense, the processes behind runoff generation in moderate floods can differ from the dominant processes in extreme events, hence the importance of analyzing past floods in basins where information is available [82,83]. Thus, the concept consisted of improving the understanding of the dynamics and development of past flash flood events in areas with available data to extend and transfer this knowledge to the development of tools able to adequately predict flash floods in ungauged basins. To achieve this aim, taking advantage of the flood information collected, relationships can be established between the catchment area and the unit peak discharge, or other aspects such as the runoff coefficient or response time, among others, taking into account the terrain (orography, soil moisture conditions, etc.), climate, and rainfall characteristics, which can provide valid approximations to forecasts in ungauged basins [61]. However, authors highlight the need to provide highly accurate and real-time rainfall estimations for each area under scope as well as accurate numerical weather prediction models able to provide short-range forecasts and to adequately simulate local atmospheric processes, due to the local nature of these events [61]. The application of distributed hydrological–hydraulic models would be also welcomed to forecast flash floods, especially in ungauged basins, where they can contribute to the detection of flash flooding in tributaries. The use of these models also allows the application of methodologies such as the Flash Flood Guidance, based on the estimation of the rain necessary to cause flooding, taking advantage of hydrologic simulations, which allows to anticipate possible flash flood scenarios related to a precipitation event (see Refs. [84,85]). Finally, the authors also highlight the need to have well-established protocols and actions in place to adequately address risk management, taking into account the short time available to anticipate the hazard [86]. Lastly the standardization of methods to analyze flash floods is also recommended as it facilitates meaningful comparisons between the events that have occurred in different locations and would contribute to creating a cohesive collection of flash floods [61]. This would improve the understanding, monitoring, forecast, and mitigation of flash floods.

The Flooded Locations and Simulated Hydrographs (FLASH) project was developed to enhance flash flood forecasting in the United States [62]. It was launched in 2012 and have been operationally used for flash flood forecasting by the U.S. National Weather Service since 2017 [87]. In this case, the FLASH project assimilates mainly the data from the Multi-Radar Multi-Sensor (MRMS), which provides precipitation estimate data with a spatial resolution near to 1 km, updated every 2 min, a very high spatiotemporal resolution especially suited to flash flood forecasting [62]. A distributed hydrologic modelling framework, the Ensemble Framework For Flash Flood Forecasting (EF5), is mainly fed with this data, maintaining the resolution of 1 km over the conterminous United States (over 10 million grid points) [62,88]. The developers of the model opted to apply simple implementations of required parameters (evapotranspiration, water balance approach, baseflow calculations, etc.), due to the purpose of the simulations and their scale, which required reducing the computational burden of the simulation to provide the forecast in a tight time frame for all the grid points considered [62]. The system runs different water balance approaches, providing an ensemble of results, all of which are usable to a greater or lesser extent; thus, each approach can provide valuable information about flash flood characteristics [89]. The system forecast products, include, among others, river discharge (m³s⁻¹) and also the discharge normalized to the upstream drainage area (m³s⁻¹km⁻²), which can be more useful for identifying the specific locations that can suffer hazardous flows [62]. Differing from other early warning systems for floods, this system mainly assimilates near-real-time estimated precipitation for forecasting [62], since its main objective is to provide streamflow forecasts for the hours immediately after the simulation, primarily impacted by the rain that has already fallen. In some cases, the system also assimilates rainfall forecasts from storm-scale numerical weather prediction models to extend the lead time of forecasts (https://www.nssl.noaa.gov/projects/flash/research/, accessed on 11 April 2024). In terms of flood risk analysis, firstly, the FLASH system compares MRMS estimated precipitation with static and dynamic thresholds associated with different frequency values (return periods), in order to estimate the intensity of the event and to determine if it can surpass a critical threshold that has been previously determined [62]. Additionally, the streamflow forecast products provided by the distributed hydrologic modeling system are also evaluated to estimate risk of flooding.

An important effort over the last few years in flash flood prevention projects was carried out in China, extended over all of the country but especially applied to cover some of its mountainous regions where these events occur more frequently [63]. To this end, an intense recompilation, survey, and evaluation of flood data was carried out by means of the Chinese Flash Flood Survey and Evaluation (CFFSE) project. This allowed researchers to determine the main causes of floods, as well as key parameters like their magnitude, and also to establish warning indices with their corresponding threshold values for each area susceptible to being affected by this phenomenon [63]. Thus, the flash flood mitigation system is mainly based on the development of early warning indices, which rely on the exceedance of flood thresholds referring to precipitation and also water level registered in a real-time monitoring system that was gradually improved in the areas of study [63,64]. These flood thresholds were calculated according to historical flood disasters, topography and landforms, storm characteristics, and engineering experience, as specified in [64]. This knowledge helps authorities to make decisions and take measures to diminish flooding impacts. In the first stage (implemented during the period between 2006 and 2012), the expected peak discharge associated with the rainfall that fell during a certain time period was obtained by means of empirical methods that considered the general characteristics of the terrain of the basin under consideration (runoff coefficient, concentration time, drainage area, etc.) [64]. Then, by establishing relationships between measured rainfall and river discharge, distributed maps with an early-warning index were developed to determine the flood risk associated with certain precipitation levels. However, the authors highlight that the empirical methods are not accurate enough in some places, because some important aspects such as the storm characteristics or the distribution of population, among others, are not considered. In the second stage (implemented from 2013 onwards), an in-depth investigation was developed to establish more accurate relationships between rainfall and the water level reached in each location under interest. In this case, the considered flood threshold was related to the minimum water level that was able to reach the house located at the lowest elevation in the village under study. In the next developments, the indices used were improved to better represent the real risk, producing dynamic warning indices with real-time adaptation to the conditions. In addition, in this phase, developers took advantage of new approaches, including distributed hydrological models that contributed to obtaining more accurate threshold runoffs in relation to the flood risk. Moreover, new indices related to different probabilities of flash flood occurrence, given a certain rainfall event, were also used to identify and forecast floods and their probable associated intensities [64,90]. At this stage, rainfall forecasts were also considered to establish meteorological warning indices. Lastly, the authors confirmed the improvement of the system by using hydrological models, fed with real-time rainfall observations from satellites or radars, and even weather forecasts, to obtain a more detailed forecast of flash floods, thereby improving the general performance of the system due to the limitations of using only indices [64]. The latest developments to improve flash flood forecasting in China included implementing hydrologic and hydraulic models as well as deep-learning-based tools [91,92,93,94].

Another example of a FEWS focused on flash floods and designed for ungauged basins is the one implemented in Barranquilla (Colombia) [65]. This city is heavily affected by flash floods due to the lack of a proper water drainage system; therefore, the implementation of a FEWS is a cost-effective measure to reduce the impact of the floods that affect the city [65]. The system does not require historical information, using real-time data from multiple rain gauges to feed a hydrological model based on the SCS (Soil Conservation Service) method [65]. Additionally, it employs a hydraulic model (HEC-RAS) to predict the water level and velocity at different cross sections, enabling the estimation of hazard levels within a time window from 15 min to 3 h. Then, a diffusion process is performed via a web platform and social networks, indicating the danger level for the pedestrians and vehicles at different points of the stream [66]. Two kinds of events were analyzed (short and long duration), given reasonable results in both cases. Some differences in water level and velocity, when compared with measured values, were detected, but they did not affect the hazard level [65]. Finally, it is important to note that, given the nature of the flash flood events in this area, the reaction time between the alert being issued and the occurrence of the hazard level can be as short as a few minutes [65], making it crucial to develop faster alert practices to reach the broadest audience as soon as possible. Here, the authors suggest the installation of displays along the stream indicating the alert, along with improvements in diffusion via web and social media platforms.

4. Advancements in FEWSs

From a technical point of view, the conducted review shows that most of the reviewed FEWSs include numerical models to simulate the main processes involved in the hydrologic procedure. In this sense, most of the authors involved in FEWSs development emphasize that the application of hydrologic and hydraulic models is essential to provide a better depiction of floods and the associated risks. In general, authors consider numerical models to be the most effective means to comprehensively incorporate the characteristics of the areas under flood risk, enabling a more precise delineation of flood development and its impacts. Therefore, it seems to be clear that FEWSs development should incorporate these types of models to provide the most useful forecasts. Particularly, the inclusion of models that provide flood maps (distributed models) is especially welcomed for the precise deployment of countermeasures. This has led the most advanced FEWSs to integrate a semi-distributed hydrological model to obtain the river flow forecasts throughout the basin, which, in turn, feeds a distributed hydraulic model to generate the associated flood maps over specifically vulnerable areas. In this sense, it is important to highlight that high-resolution DEMs are essential to perform an accurate hydraulic procedure. Despite the availability of some global DEMs, as commented above, higher resolution models are necessary to adequately address flood development in villages or areas under flood risk [95]. In this context, some efforts have been undertaken in recent times. An example is the case of Spain, where high-resolution digital terrain models, with a 2 m horizontal resolution, were developed to cover the entire territory and were made freely available (https://centrodedescargas.cnig.es/CentroDescargas/index.jsp, accessed on 11 April 2024).

Another important advancement in FEWSs development is the integration of several ensemble forecasts to address the inherent uncertainty typically associated with predictions [55,58,69,88]. This enables the provision of not only flood maps indicating the areas affected and the risk level but also the risk probability, which allow a better understanding and ability to address these events.

Finally, it is important to remark that the phases of “warning dissemination and communication” and “preparedness to respond”, which are often less developed, as commented earlier, are crucial for the effective application and mitigation purposes of FEWSs. These phases play a key role in successfully minimizing the impact of floods on communities. In this sense, the establishment of clear protocols for flood warning communication and the concrete and specific measures to be taken for the different alert issues, including evacuation, if necessary, is fundamental [96,97]. Additionally, the dissemination of FEWSs features, as well as a detailed explanation of result interpretation, is necessary not only for the authorities in charge of implementing preventative measures but also for the community in general, to enhance preparedness for flood risk situations. In the latter case, efforts to involve education communities in flood risk awareness and protocol procedures during a flood emergency, as well as the general population, are fundamental to effective responses to a flood event [96]. Another gap that has been identified during this research is that some FEWSs do not consider the significant role played by certain infrastructures in flood development, such as protection structures or dams. To the extent possible, all the structures and infrastructure with some role in flood development should be considered to improve the flood forecasts, since, in addition, some of them can contribute to flood mitigation, as in the case of dams [95,98].

Thus, based on the trends found in the literature and the advances in the field of data acquisition and assimilation, computation, and communications, among others, the following components have been identified as key aspects of the newly developed FEWSs:

• Accurate historical and real-time meteorological and hydrologic measurements: to perform historical analyses that contribute to a better understanding of flood development, robust calibration, and validation of the implemented numerical models, as well as to enable the adequate establishment of the initial conditions for the forecasts.
• Semi-distributed hydrologic models: to transform the atmospheric forecasts into river flow forecasts throughout the basin under consideration.
• Distributed hydraulic models: to obtain flood maps in vulnerable areas associated with the forecasted river flows.
• Ensemble forecasts: different atmospheric forecasts should be considered to address the uncertainty inherent in predictions. Additionally, hydrological ensembles, constructed by applying different hydrological approaches, are also encouraged.
• Adequate dissemination, communication, education, and community preparedness to respond: to translate the technical aspects of the FEWS into effective flood mitigation.

5. Conclusions

In this paper, a detailed review of the scientific literature on Flood Early Warning Systems (FEWSs) was carried out. The review mainly focused on high-impact journals, although comprehensive and rigorous information available in local documentation developed by institutional administrations was also compiled. The main purpose of FEWSs is to forecast flood events well in advance, which can help responsible authorities to act before the event occurs and mitigate flood damage. Thus, the review presented in this paper provides good overall representativeness of the different types of FEWSs worldwide, most of which are currently fully operational. However, the authors acknowledge that, in order to maintain a manageable volume of information, some other interesting FEWSs, with similar characteristic to the ones presented in this paper, had to be omitted from this review.

The analyzed FEWSs range from basin or regional to continental or global scales and from basic to technically advanced, providing a large sample of different approaches and implementations applied worldwide. Brief descriptions of the main characteristics, operational features, numerical models implemented, and the real application of these FEWSs, are also provided.

Therefore, a global perspective of these systems is provided, allowing for the identification of some key benchmarks and serving as a kind of best practice guide to help and assist in the development of new FEWSs. Some remarks can be made based on this work:

• According to the technological implementation of these systems, most of the analyzed FEWSs are based on a similar scheme: starting from observed data and precipitation forecasts, hydrologic and hydraulic models are applied to model runoff processes and predict flooded areas in advance. It is important to note that some FEWSs do not include the hydraulic component, mainly due to the size of the area under consideration and the computational cost, and flood maps are not provided. In these cases, some of the reviewed FEWSs use static flood maps corresponding to different flow thresholds (return periods), which can be useful to approximate the area expected to be flooded by the forecasted river flows. This review also indicates that some of the FEWSs use approaches based on flood threshold exceedance to determine the flood possibility and its intensity, especially when there are no hydraulic models implemented. These indices are usually obtained from the analysis of historical river discharge information.
• The most advanced systems usually require a large amount of data to function properly, but some approaches also demonstrate that the systems can work properly even with limited input data, helping in flood mitigation.
• The prediction of floods based on the ensemble of precipitation forecasts appears to be the best approach, as it can provide very useful additional information related to the probability of the occurrence of an event, thus reducing the uncertainty inherent to the forecasts. In this sense, the use of different hydrological approaches (hydrological ensemble) can also contribute to limiting the uncertainty.
• This literature review is mainly focused on the technical aspects of the FEWSs. This emphasis is due to the fact that most existing FEWSs have specifically developed this part, as the components related to communication and preparedness are more challenging to address due to the requirements of involving other stakeholders to handle them adequately. This is certainly an important gap in most of the FEWSs since the communication of risk and corresponding response planning are key factors to reduce the negative effects of floods and ensure the effective real application of these systems. Therefore, more in-depth research in these fields is required, although it is important to remark that, increasingly, new FEWSs developments are paying more attention to these FEWSs components.
• The lack of detailed data is one of the main reasons why wide areas of the world lack this tool. However, this paper shows some examples of FEWSs developed in areas with a low level of data availability. While these systems usually offer less detailed predictions than more advanced FEWSs, they can properly forecast floods, thus helping in mitigating the negative effects of these events. It is encouraged to increase FEWSs coverage, especially in areas where these systems are scarce. In this sense, efforts to improve data availability are imperative, not only to facilitate the development of these systems in such areas but also to conduct rigorous calibration and validation of the models and approaches implemented, which is essential to enhance the accuracy and robustness of flood forecasts.
• According to this literature review, a complete FEWS maximizing flood mitigation must incorporate the following components: accurate historical and real-time meteorological and hydrologic measurements; ensemble forecasts; a semi-distributed hydrologic model to obtain river flow forecasts; a distributed hydraulic model to generate the corresponding flooded areas; and robust and effective dissemination, communication, education, and community response preparedness components.

Thus, this work has aimed to provide a comprehensive and valuable literature review, describing the main characteristics, implementations, numerical models incorporated, methodologies, and applications of the reviewed FEWSs, while also identifying the necessary key components that must be integrated into FEWSs to ensure highly efficient performance. Thus, this review provides a basis upon which scientists and engineers can rely on to improve and develop new FEWSs. This can facilitate the implementation of more effective mitigation measures to reduce flood damage and improve community resilience.