1. Introduction
Warning systems are key to resilience for hydrometeorological hazards as they alert people to the risk of potential hazards and encourage protective action to be taken (Basher 2006). To improve early warning systems, the World Meteorological Organization (WMO) is encouraging nations to adopt impact forecasts and warnings, as they argue that communicating the potential social, economic, and environmental hazard impacts allows individuals and communities to adjust their plans and better manage the potential consequences of the hazard [International Federation of Red Cross and Red Crescent Societies (IFRC) and Met Office 2020; WMO 2015].
Traditionally, warnings have relied solely on weather-based factors (e.g., wind speeds, snowfall depth) and hazard timing and location, but “impact based” warnings also consider exposed and/or vulnerable populations and infrastructure (IFRC and Met Office 2020; WMO 2015). Impact forecasting and warning (IFW) systems differ from traditional warnings by communicating what the hazard(s) will do, rather than only what they will be. They are built on impact-based thresholds rather than hazard-based thresholds (IFRC and Met Office 2020; WMO 2015) that may misrepresent the impacts of the hazard(s) (F. Sai et al., unpublished manuscript, available at https://doi.org/10.5194/nhess-2018-26). Furthermore, hazard-bsed warnings may lack messaging about risk, leading to potential inaction (F. Sai et al., unpublished manuscript, available at https://doi.org/10.5194/nhess-2018-26).
The risk of hydrometeorological hazard impacts depends on the vulnerability of the people, infrastructure, and environment, and the exposure of these “assets” multiplied by the likelihood of the hazard (IFRC and Met Office 2020; Poolman 2014; Tarchiani et al. 2020; WMO 2015). Accurately communicating the risk of hydrometeorological hazard impacts in IFWs thus requires combining likelihood and hazard severity with exposure and vulnerability (Poolman 2014; Tarchiani et al. 2020; WMO 2015).
Notable historic severe weather events revealed communication gaps between warning services and recipients. These have been attributed to various underlying social behaviors such as warning fatigue (Mackie 2013; Wagenmaker et al. 2011), and understanding warning terminology (Ching et al. 2015). IFW systems may help reduce the effects of these factors and increase warning compliance (Morss et al. 2018; Potter et al. 2018; Weyrich et al. 2018). However, their effectiveness depends on several caveats in design and implementation (see Morss et al. 2018; Potter et al. 2018; Ripberger et al. 2015; Scolobig et al. 2015; Weyrich et al. 2018).
Implementing IFW systems can be costly, requiring careful cost–benefit analysis (Merz et al. 2020). Potter et al. (2021) found that key benefits include a perceived increase in understanding impacts, added awareness of antecedent conditions, possible reductions of “false alarms,” and increased interagency communication. Others found benefits to be enhanced situational awareness (Kox et al. 2018a), and improved planning and response efforts (Terti et al. 2015). Alternatively, Potter et al. (2021) identified key challenges to be lack of data, the potential for conflicting messages, and an increased burden on agencies providing information to forecasters. Verification and conflicting roles and responsibilities are other significant challenges identified by Hemingway and Robbins (2020) and Kaltenberger et al. (2020), respectively. Data challenges can also be a roadblock, including their availability, processing capabilities, and management (Hemingway and Robbins 2020; Potter et al. 2021; Wei et al. 2018).
There is a growing need to address challenges with data availability and access and identify appropriate data sources. A key challenge is in understanding the data needs and requirements for implementing IFWs, such that we can identify appropriate datasets and data sources. To achieve this, we first present a review of IFW system elements and data needs, followed by findings from a series of qualitative interviews, the majority of which were producers of severe weather warnings and users of hazard, impact, vulnerability, and exposure data in Aotearoa (the Māori name for New Zealand).
a. Elements of impact-based forecasting and warning systems
Traditional (hazard based) severe weather warning systems rely on hazard forecasting and observations, with warning thresholds based on measurable characteristics of the hazard (e.g., minimum/maximum wind speeds, snow depth) (Harrison et al. 2014; Obermeier and Anderson 2014). IFW systems introduce the human element to early warning systems.
The terminology of IFWs varies across the literature. In this paper, we will use the definitions in Table 1, largely from the WMO guidelines on IFW systems (see WMO 2015).1 Kaltenberger et al. (2020) proposed an additional term of “impact-oriented warnings” to refer to warnings that are independent of the production process (i.e., the warning thresholds or criteria) used to issue warnings. These may be hazard-based or impact-based and include both a tangible and understandable description of expected impacts and clear advice on what to do. Impact-based warnings and impact warnings differ based on the production process used to issue them. The goal of the WMO Guidelines is to evolve severe weather warning services into the final “impact forecasting and warning” form, where vulnerability and exposure are integrated into the thresholds. Herein, we refer to all as IFW systems to align with the goal of the WMO. However, when the type of warning is specified, we will refer to its proper term according to the WMO definitions.
Definition and examples of the warning terms used in this paper.
IFW systems require knowledge of hazards and their likelihood, as well as the underlying vulnerability and exposure of the assets (e.g., people, infrastructure) at risk (Poolman 2014; Tarchiani et al. 2020; WMO 2015). Thus, the four data types needed for an IFW system are hazard (incorporating likelihood), impact, vulnerability, and exposure; defined in Table 2. These data types form the conceptual basis of IFWs. The lack of data and knowledge around impacts and risks relating to hydrometeorological hazards among meteorologists has been identified as a key barrier to implementing IFWs (Harrison et al. 2014; Obermeier and Anderson 2014; Potter et al. 2021). In particular, exposure and vulnerability data appear to be largely left out of weather warning systems (Potter et al. 2021). Thus, the WMO highlight the need for better collection and access to vulnerability and exposure information and understanding of how vulnerability and exposure data can be integrated into an IFW system (WMO 2015). The warning value chain provides a framework for understanding how hazard, impact, vulnerability, and exposure data fit into an IFW system.
The four types of data needed for IFWs.
b. The impact forecasting and warning value chain
Golding et al. (2019) outlined the warning value chain approach for representing the flow of information among actors in the warning chain for designing and operating hydrometeorological warnings. The value chain consists of six components for an operational IFW system: 1) observation, monitoring, and detection; 2) weather forecasting; 3) hazard forecasting; 4) impact forecasting; 5) warning; and 6) decision/action, as shown in Fig. 1.
A visual representation of the severe weather warning value chain based on Golding et al. (2019), where blue boxes relate to hazards, red boxes relate to impacts, and green relates to the outcomes.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
The first component of the warning value chain is observation, monitoring, and detection. Meteorological monitoring and observation are routine practices (WMO 2018). Observation, monitoring, and detection traditionally use ground-based observations (e.g., from weather stations) and remote sensing tools such as satellite and radar imagery for meteorological phenomena (e.g., Brotzge et al. 2013). Social media and crowdsourcing have also become useful tools for detecting and monitoring hydrometeorological phenomena such as tornados and hail (Harrison and Johnson 2016).
Weather forecasting makes heavy use of computer modeling, which has seen technological advances in the last half-century to increase accuracy (Bauer et al. 2015). A shift occurred from deterministic forecasting to probabilistic forecasting, with the advent of approaches such as ensemble prediction systems (Global Facility for Disaster Reduction and Recovery 2016; WMO 2015). In South Africa and the United Kingdom, the implementation of an impact-based forecasting and warning system is solely based on probabilistic forecasting (e.g., Neal et al. 2014; Poolman 2014). With probabilistic forecasting, warnings can be issued earlier using low probabilities that can increase or decrease as confidence in the likelihood increases (Neal et al. 2014). Adding probabilistic uncertainty estimates in forecasts was found to improve both compliance and decision quality among technical forecast users (LeClerc and Joslyn 2015).
Upon producing weather forecasts, hazard forecasting may occur. Weather-induced hazards include flood and flash flood, storm surge, drought, wildfire, snowfall, ice, extreme temperatures, damaging winds, and so on. Weather forecasts are typically used in conjunction with other data/information about a hazard to produce a hazard forecast. For example, storm surge forecasts in the United Kingdom are produced using wind forecasts, sea level models, and wave/tidal models (Flowerdew et al. 2013). Hazard forecasting is challenging due to errors in predicting atmospheric conditions (Golding et al. 2019; National Research Council 2006). For example, forecasting for heavy rainfall in a nearby catchment, rather than the catchment that experienced the downstream flooding (Majumdar et al. 2021).
Impact forecasting is the next step in the warning value chain to identify and understand the potential impacts of a given hazard. Impact forecasting involves combining meteorological and hazard information with information about exposed and/or vulnerable assets (e.g., people, property) (Merz et al. 2020). Risk modeling can support impact forecasts by providing quantitative assessments of impacts and loss based on the hazard and asset information using a vulnerability/fragility function (Schmidt et al. 2011). However, the use of risk models by practitioners has been met with key challenges around data availability and investment priorities (Crawford et al. 2018). The data needs for this part of the warning value chain are thus the main topic of this paper.
The warning may be considered the end product or service of the warning system. Warnings are intended to communicate information about an oncoming hazard and the associated risk(s) to exposed and vulnerable audiences (WMO 2018). Design and delivery are two key factors in the success of a warning (Golding et al. 2019). Design involves having organizational and decision-making processes, and operational communication systems and equipment (WMO 2018). Delivery involves disseminating and communicating the relevant information to the target audiences in terms that are understandable and actionable by the audiences (WMO 2018).
Upon receiving a warning alert or message, recipients face many decisions. The decision-making end of the warning value chain has received a plethora of social science research. IFWs emerged from that research as a recommendation for inciting the desired responses to warnings (e.g., Ching et al. 2015; Wagenmaker et al. 2011). Much research has since been done to evaluate IFWs on their effectiveness for recipient decision-making, indicating that behavior to impact-oriented warnings varies across studies and hazards (see Casteel 2018; Morss et al. 2018; Potter et al. 2018; Ripberger et al. 2015; Taylor et al. 2019; Weyrich et al. 2018, 2020a,b).
Evidence has shown that including impact information with warnings can improve understanding of and response to warnings (Potter et al. 2018; Ripberger et al. 2015; Taylor et al. 2019; Weyrich et al. 2018, 2020a,b) by aligning with how people already tend to interpret warnings based on the impacts of the severe weather (Williams et al. 2017) and to contextualize the information (Schroeter et al. 2021). However, caution is advised when formulating the messages as research has found that the terminology used in impact-oriented warnings may induce fear and can cause undesired responses to warnings (Ripberger et al. 2015). For example, impact-oriented warnings can incite people to evacuate from hurricanes (the desired behavior for those at risk) (Morss et al. 2016, 2018; Ripberger et al. 2015). However, impact-oriented warnings and fear-based messages also have the potential to increase “shadow evacuations”2 during hurricanes (Morss et al. 2016), causing gridlock on motorways, increasing exposure, and thus risk, of even more people (Baker 1991; Lamb et al. 2012; Yin et al. 2016).
For tornadic hazards, Casteel (2016, 2018) found that impact-oriented warnings led to greater behavioral intentions of sheltering in place (a desirable outcome). Conversely, Ripberger et al. (2015) found that impact-oriented messages have the potential to “backfire” as their results showed a decreased probability of sheltering in place and an increased probability of evacuating for events with tags of higher impacts such as “devastating,” or “incredible.” This behavior could result in people putting themselves at even more risk in situations where sheltering in place is advised (Ross et al. 2015).
The implications for these effects must be seriously considered when designing an IFW system. Limiting the use of fear-inducing messages can enhance the perceived credibility of the messages and reduce the effects of shadow evacuations (Morss et al. 2018). Including clear prescribed actions can also increase perceptions and appropriate intended responses to impact-oriented warnings (Weyrich et al. 2018).
Partnerships and sharing information among the various agencies involved is an important aspect of the warning value chain framework (Golding et al. 2019). It has been argued that poor linkages between warning system components have been major causes for warning systems to fail, resulting in disasters (Garcia and Fearnley 2012). Thus, linking the components into an integrated early warning system is critical to their effective performance (Garcia and Fearnley 2012). This requires coordination and collaboration across the various agencies and levels that have the relevant information about hazards, impacts, vulnerability, and exposure (Garcia and Fearnley 2012; Golnaraghi 2012).
As outlined in relevant guidelines and resources (e.g., IFRC and Met Office 2020; WMO 2015), there is a need for not just hazard data but also impact, vulnerability, and exposure data. However, it is unclear where each of these data types is used in the warning value chain. Thus, the objective of this research is to identify the uses and gaps for hazard, impact, vulnerability, and exposure data within an IFW system for each component of the warning value chain framework. We adopt a grounded theory method, described next, to collect and analyze interviews to address the question, “what are the data uses and gaps for impact forecasts and warnings?” This will be explored in a New Zealand context to support the country’s efforts toward fulfilling the WMO’s objectives for member nations to implement IFWs.
2. Research method
To address the objective of this research, we adopt a qualitative research approach to explore the issues in depth with participants both within and outside of Aotearoa–New Zealand. This is opposed to a quantitative approach, which is more relevant for understanding the breadth of results across a population (Mack et al. 2005). We adopted a grounded theory method (GTM) to collect and analyze the data. As an exploratory strategy, GTM is particularly useful for research on areas with little to no prior research and where theory building is needed or desired (Fernandez 2005; Lehmann 2010; Seidel and Urquhart 2016). This study employed the evolved/Straussian grounded theory method (ES-GTM) due to its wide use in relevant fields (Matavire and Brown 2013; Urquhart et al. 2010), its ease of use (Charmaz 2006; Kelle 2005), and its allowance for a priori literature review (Hughes and Jones 2003; Matavire and Brown 2013).
A purposive sampling method (Chun Tie et al. 2019) was used to target and recruit participants based on their roles in severe weather risk communication and response, and use of impact, vulnerability, and exposure data. After the initial interviews, theoretical sampling guided the rest of the data collection process for this research, which is “the process of data collection for generating theory whereby the analyst jointly collects, codes, and analyses [their] data and decides what data to collect next, and where to find them, in order to develop [their] theory as it emerges” (Glaser and Strauss 1967, p. 45).
Thirty-nine (n = 39) experts in weather forecasting, warning, response, risk modeling, and data collection and management were interviewed between November 2018 and May 2021 (described in Table 3). These 30–60-min in-person or virtual semistructured interviews aimed to address the gaps in the literature about IFW data needs and sources. Questions asked for participants’ thoughts on 1) IFWs (e.g., what they know about IFWs, perceived challenges and benefits, requirements for implementation); 2) what kind of impact, vulnerability, and/or exposure data they use or need, why, and how; 3) the life path of the data (e.g., how it is obtained, used, stored, what happens to it after its intended use); 4) experienced and/or perceived challenges obtaining data required for IFWs and other uses; and 5) thoughts on collecting and using alternative data sources (e.g., social media and crowdsourcing). Interviews were audio recorded and transcribed verbatim. In support of achieving the objective of this paper, findings about the data uses and gaps for IFWs are reported here, while findings about data sources and access will be reported in separate papers currently under preparation (see Harrison et al. 2021).
Summary classification of interview participants.
A “low risk” ethics notification was lodged with the Massey University Human Ethics Committee before data collection in 2018. All participants remain anonymous and are assigned an alphabetic code (A, B, C, etc.), being identified only by the area of expertise and/or practice, industry, location, or governance level (Table 4). Interviews were audio-recorded and transcribed verbatim. Qualitative analysis (including coding and memo-writing) was conducted using NVivo 12 (Bergin 2011) following the ES-GTM.
Participant codes. All participants remain anonymous and are assigned an alphabetic code (A, B, C, etc.), being identified only by the area of expertise and/or practice, industry, location, or governance level.
3. Findings and discussion
We identified two key themes from the interview data: 1) data uses for IFWs and 2) the need for more understanding on different types of impacts, discussed next.
a. Data uses in the impact forecasting and warning value chain
The actors and data uses for IFWs are discussed in the following sections using the warning value chain as a framework for guiding the discussion. The focus of this paper is on the data uses and gaps of the impact forecasting and warning portion of the warning value chain. As such, only a brief description of results on the data uses and actors in the first three components of the warning value chain (observation, monitoring, and detection; weather forecasting; and hazard forecasting) will be provided. Following this, results and discussions will be presented for each of the impact forecasting and impact warning components.
1) Weather and hazard observation and forecasting
Extensive research has been conducted for weather and hazard observation and forecasting that documents more comprehensive explanations of the data gaps and uses for weather and hazard observation and forecasting (e.g., Gneiting and Raftery 2005; Saima et al. 2011).
In Aotearoa–New Zealand (NZ), meteorological services (e.g., the NZ MetService), hydrological services (e.g., regional and local council hydrologists in NZ), and hydrometeorological services [e.g., New Zealand’s National Institute of Water and Atmospheric Research (NIWA)] are the primary actors in this value chain component. The NZ MetService monitors the weather, council hydrologists monitor river networks while simultaneously monitoring the weather themselves, and NIWA collaborates with councils to set up monitoring and modeling scenarios for flood early warning systems. Forecasters also use media and social media to monitor areas where radar coverage is lacking (Met. Int. C; Met. NZ. K; see Table 4 for descriptions of participant codes).
Weather forecasting is primarily done by the meteorological services (public and private), as outlined by an NZ meteorologist: “our responsibility is for the forecast of rain until it hits the ground. After that, it is the responsibility of other agencies to deal with what that rain does” (Met. NZ. F). By using numerical weather prediction (NWP) models, weather forecasts can provide meteorologists, hydrologists, and lifelines sectors such as transportation with advance notice, such as 48 h, 3 days, or 5 days, of a potentially hazardous event, which allows for extra preparation time (Met. Int. I; Met. NZ. K). “Nowcasts” use observational data (e.g., spatiotemporal extrapolation of lightning strikes or radar-estimated precipitation) or very short-range NWP for near-real-time forecasting out to a few hours (Farnell et al. 2017; Kotsuki et al. 2019; Srivastava and Bhardwaj 2013). It is important to note the differentiation between the data needs for IFWs in terms of time scale; the needs, type, and availability of data for real-time operations and for developing warning systems can differ. For example, social media and crowdsourcing have value in their real-time contribution to a warning system for hazard and impact detection, forecasting, and warning verification (Harrison et al. 2020), but may be less useful due to the qualitative nature for defining impact warning thresholds. This should be explored in future research.
“are there particular concerns for your region? Is there anything that we should be on the lookout for?” And it’s really those interactions that also feed into the decision for “should this be red?” because if we talk to them after we issue a Watch, but before initiating the Warning, and we say “look, we’re considering a warning for 300 mm,” they might go “oh, that’s going to flood our entire region” and that might spark that two-way discussion to say “well, okay, that’s really interesting, how concerned are you, what level of flooding do you anticipate?” Which might lead us down the path of “this should be red” (Met. NZ. K).
From this discussion, a watch or warning decision is made, as well as the warning level. The hydrologist possesses more knowledge of the current conditions (e.g., river levels, soil moisture content) and is thus able to contribute to a more informed warning decision (Met. NZ. A, K).
The NZ MetService primarily uses hazard-based thresholds for their watches and warnings, such as “widespread (over an area of 1000 square-kilometers or more) rainfall greater than 50 mm within 6 hours or 100 mm within 24 hours” (MetService 2021). In some cases, the MetService uses more dynamic thresholds depending on the region, antecedent conditions, and feedback from emergency management (EM) groups and councils. For example, in Auckland, the minimum speed of strong winds was lowered for northeasterly winds as these winds tend to be more damaging in Auckland than southwesterlies (Met. NZ. K). Furthermore, senior meteorologists also use their tacit knowledge developed over the years to make a judgement call on issuing hazard forecasts and warnings (Met. NZ. K). The meteorologists also use media reports and social media to detect hazards and impacts and inform their warnings (Met. NZ. K).
Floods were the main hydrometeorological hazard that participants discussed. In NZ, participants described how regional councils typically conduct flood hazard forecasting using meteorological forecast data combined with their hydrological data (e.g., river level and flow gauges) (Met. NZ. A), and sometimes their local and historical knowledge of past impacts (EM NZ. Reg. A, B, C). These hazard forecasts usually do not provide impact information, as one NZ regional council flood warnings specialist stated, “we don’t then do any quantitative work to . . . determine how much impact that flood financially damaged, that type of thing. We’ve not done that in the past to my knowledge” (Hyd. Gov. NZ. Reg. A).
2) Impact forecasting
The WMO Guidelines briefly suggest that impact forecasting “could be done in a subjective way working alongside [partners], or in an objective way through developing an impact model using vulnerability and exposure datasets as well as meteorological information” (WMO 2015, p. 3). Interview findings corroborate this suggestion. Figure 2 provides a conceptual visualization, based on our interviews, of the four data types used as inputs for impact forecasts, which then support either the more objective risk/impact modeling approach, or the more subjective impact-oriented discussion approach. These two approaches will be discussed next.
Data inputs and uses for impact forecasts and the two main approaches. The data inputs are represented in the top-left with arrows pointing to the impact forecast component. Blue, red, orange, and yellow are associated with hazard, impacts, vulnerability, and exposure, respectively. The two approaches of impact forecasting are listed on the left side of the impact forecast box, and the actors involved are listed on the right. More details on the activities used for each impact forecasting approach are provided in the bottom box. The multidirectional arrow between the two impact forecasting approaches signifies that these two approaches are best used complementarily with each other.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
(i) Model-based impact forecasting approaches
Model-based impact forecasting approaches use computer models, algorithms, and quantitative data to identify potential impacts of varying severity. Specifically, risk models require detailed data about the fragility/vulnerability of assets (including buildings, infrastructure, people, vehicles, etc.), dynamic exposure data, and hazard data4 (Schmidt et al. 2011). In the United Kingdom, under the Natural Hazards Partnership (NHP), several models have been developed, or are under development, for real-time forecasting of various hazards and impacts. Two of these models were discussed in interviews.
The vehicle overturning model (VOT) is an impact forecasting model for U.K. highway networks during high wind events (Hemingway and Robbins 2020). The model uses a “classic risk assessment approach” (Met. Int. A), by defining a hazard footprint (i.e., the wind affecting vehicles), and using vulnerability and exposure indices to calculate the risk of vehicles overturning (Met. Int. A). Vulnerability factors are road altitude, number of lanes, road direction relative to the wind, and road attributes (e.g., tunnel, bridge) (Met. Int. A). The exposure index “is split out by vehicle type because we recognise that different vehicles are more susceptible to being overturned than others” (Met. Int. A; see Hemingway and Robbins 2020 for more). The Met Office does not traditionally possess knowledge or data on traffic flows (Met. Int. A), thus they partnered with agencies that do, namely, transportation agencies in England and Scotland (Met. Int. A).
We needed a lot of information on different property types, and we needed information on how people move around the city and things like that to account for the fact that people commute in and out […] And […] within our broader partnership under that Natural Hazards Partnership we have an organisation called the Health and Safety Executive and they do a lot of work in the risk assessment space and have a lot of available data. And that allowed us to really produce that whole second part […] Had we not had that; I think we would’ve really struggled (Met. Int. A).
The approach of the SWFM is slightly different from the VOT as vulnerability is included but not via a vulnerability index layer; rather via classification of the property itself (e.g., commercial properties vs residential properties) (Met. Int. A). The output is impact grades ranging from lesser to greater impacts to life and safety (Met. Int. A; see Aldridge et al. 2016; Cole et al. 2016a,b for more).
Similarly, a flood early warning system (FEWS) is under development in collaboration with the Samoan Ministry of Natural Resources and Environment for the Vaisagano River in Apia, Samoa, which incorporates outputs from RiskScape.5 An impact-forecasting component has been integrated into this FEWS to assess life and vehicle road closure safety needs (Risk Modeling NZ. C, D). After a postevent analysis of flooding in Wellington, New Zealand, it was suggested that risk modeling would have been useful to identify the risk of flood damage to emergency vehicles that drove through flood waters and were subsequently damaged (Risk Modeling NZ. C). The data/tools used in this system are near-real-time rain intensity forecasts based on available NWP models [e.g., Pacific Islands Ocean Observing System (PacIOOS) 3 km grid], rainfall–runoff predictive relationships, exposure of assets (e.g., roads, buildings), lidar topography, flood inundation models for different annual exceedance probability events, and vulnerability functions (Risk Modeling NZ. D). More exploration into characterizing these data sources will be covered in future research.
These impact forecasting systems are still under development and testing and thus are not yet fully operational (Met. Int. A; Risk Modeling NZ. C, D). Consequently, impact forecasting models are not currently the sole tool used for designing and issuing IFWs (Met. Int. A). “Impact-oriented discussions” (Met. Int. A) between different agencies and groups are still the main approach for designing and issuing IFWs, which aligns with the suggestion for a subjective approach in the WMO Guidelines (WMO 2015), discussed next.
(ii) Impact-oriented discussions
Impact-oriented discussions typically involve many stakeholders that possess the different types of knowledge and information needed to understand and forecast impacts. In New Zealand, the impact forecasting approach tends to be an impact-oriented discussion. For example, the red warnings from the MetService’s new warning system (see MetService 2019) are issued based on impact-oriented discussions with hydrologists and EM groups (Met. NZ. K). A red warning does not have “fixed thresholds” (Met. NZ. K); it is only issued if the event is expected to produce significant impacts. The information used to inform a red warning decision is gathered from directly communicating with the local EM groups and regional/local councils where the impacts are expected to occur (Met. NZ. K). Impacts reported in the media and on social media platforms then help to verify and update warnings (Met. NZ. K).
Similar impact-oriented discussions occurred when ex–Tropical Cyclone (TC) Debbie and ex-TC Cook made landfall in New Zealand in 2017 and caused significant flooding in Edgecumbe. After ex-TC Debbie impacted the Bay of Plenty Region in NZ from 3 to 6 April 2017, causing widespread flooding (Cullen et al. 2017): “[the Bay of Plenty] were suddenly faced with ex-TC Cook coming in after that [on 13 April 2017]” (EM NZ. Reg. A). In preparation of ex-TC Cook, the Bay of Plenty EM group had a team conduct a probability assessment to “look at what … the potential impact [might] be of this event coming in, given the current situation that we already had been impacted” (EM NZ. Reg. A).
This risk assessment helped to identify which neighborhoods needed to be evacuated in response to the forecast for ex-TC Cook (EM NZ. Reg. A). It relied on discussions and collaboration where decision-makers and stakeholders shared knowledge of recent impacts caused first by ex-TC Debbie that made some areas and assets more vulnerable to damage/flooding from ex-TC Cook (EM NZ. Reg. A). Additional hazard information such as river and stream levels were also shared (EM NZ. Reg. A).
Impact-oriented discussions also make use of decision support systems and integration of spatial data to identify exposed and vulnerable areas and assets. For example, an impact-based decision support system was set up between the National Weather Service and emergency managers in New York to support forecast updates, response decisions, and public and partner messaging (Hosterman et al. 2019). Integration of spatial layers has also been used to analyze social vulnerability to extreme precipitation in Colorado (Wilhelmi and Morss 2013) and to support flood risk management decision-making in Brazil (Horita et al. 2015). More work is needed to explore these similar efforts for integrating hazard, impact, vulnerability, and exposure data to support IFWs in New Zealand.
if you’re doing the collaborative approach with very broad scenario type impacts, then perhaps the update frequency can be less. If you’re doing a more robust, sort of bespoke risk algorithm, then I would hope the update frequency would be higher because you are trying to actually capture risk given the current circumstances. And if you’re not updating frequently, then you’re using a historical dataset to inform the current risk and I don’t think that’s very accurate (Met. Int. A).
In these examples, all four data types (hazard, vulnerability, exposure, and impacts) come into play for impact forecasting (Fig. 2). Model-based impact forecasting uses quantitative datasets, typically in spatial format (e.g., GIS shapefiles). Alternatively, impact-oriented discussions utilize the forecasts along with more holistic qualitative information in the forms of tacit knowledge and experience, observations from social media and/or media reports, meteorological observations, hydrological observations, etc. Discussions between stakeholders bearing different knowledge are critical for the subjective approach. While objective model-based approaches may be considered the more “sophisticated” approach (e.g., Schroeter et al. 2021), our participants indicated that both approaches have value, and the most effective approach is to use them together rather than preference one over the other (Met. Int. A; EM. NZ. Reg. D). This exemplifies the importance of incorporating different scientific epistemologies into transdisciplinary science and research to inform policy and decision-making (see International Science Council 2021).
5) Impact warning
Findings from the interviews show that impact warnings introduced new needs for setting warning thresholds, discussed below.
(i) Defining impact thresholds
Hydrometeorological warnings are based on thresholds traditionally set around physical hazard characteristics, such as “x” amount of rainfall in “y” amount of time. Participants indicated that previous event impact data can help determine warning thresholds (Met. Int. A, B, D; EM. NZ. Reg. B, C). For example, investigating the drivers of the resulting impacts from a recent event can help redefine thresholds such that they are specifically impact-related (Met. Int. A). This can involve identifying the rainfall and wind speed leading up to the event that resulted in the impacts to determine which thresholds result in such impacts (Met. Int. A), such as “x” wind speed will result in “y” type of damage to “z” structures. An NZ EM official highlighted the value of cataloguing impact data to prevent “history from repeating itself,” by allowing officials to identify trigger points (or thresholds) based on past events (EM. NZ. Reg. B). Like the impact forecasting approaches, countries and regions appear to use different approaches to defining impact-based thresholds, varying in how systematic they are (Fig. 3).
Data inputs and uses for defining impact thresholds for the impact warning component. This component uses similar approaches to forecasting impacts: the objective risk/impact modeling approach, or the subjective impact-oriented discussion approach, as shown at the bottom of the figure. All four data types feed into both approaches. Asset data are represented in gray because they overlap with both vulnerability and exposure data. The actors are listed on the right side of the impact warning box. Dynamic vulnerability (orange) and dynamic exposure (yellow), such as antecedent physical conditions (e.g., drought or oversaturated soil), human response capabilities, or human locations, are needed to update thresholds so that the risk estimates are accurate.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
Tacit knowledge and experience are less systematic ways to identify future event thresholds. NZ EM officials use tacit knowledge to identify different river level impacts (EM. NZ. Reg. B, C). The Northland EM agency works with council hydrologists to informally forecast these flood impacts and inform responders and communities (EM. NZ. Reg. B). Similarly, West Coast EM officials have identified and logged river level thresholds for specific points in the river network and are looking to do this more systematically with GIS.
The Met Office have a slightly more systematic approach to documenting tacit knowledge. They use a series of “impact tables” (Met. Int. A) that were developed in consultation with local EM to “discuss what impacts are observed at different levels” to identify response capacity levels (Met. Int. A). These tables helped the Met Office understand “how different impacts have occurred related to different hazards” (Met. Int. A). The Met Office then uses a risk matrix (see WMO 2015, p. 14), with impacts on one dimension and likelihood on the other to relate the level of impact from the tables to the level of impact in the matrix (Met. Int. A). The Met Office still relies on contacting EM authorities to verify the most appropriate warning level from the impact matrix (Met. Int. A). This remains a “continual collaborative process” where the warning can be updated based on feedback from the authorities on the ground (Met. Int. A).
Still, this is less systematic than model-based techniques, and participants envisioned using impact data in risk/impact models to classify what is hazardous. Engineers often use impact data to inform the vulnerability function in risk or impact models by associating levels of damage to hazard intensities for a given asset (Pita and de Schwarzkopf 2016; Tarbotton et al. 2015). Postevent damage surveys can be used to develop and calibrate vulnerability functions and identify impact thresholds (Met. Int. A, B). For example, an event database with empirical impact data from damage surveys can provide evidence about how hazard intensity will cause damage (Met. Int. B). Eventually, these models can help identify the causes of certain impacts. This includes determining which wind speed or rainfall amount led to high impact events (Met. Int. A). By understanding the mechanisms that led to the impact, forecasters can improve their models and identify the specific impact and warning thresholds (Met. Int. A).
Currently, risk modelers do not appear to be involved in the warning operations of some of the participating meteorological, hydrological, and emergency management agencies, such as in NZ. The VOT developed by the Met Office is undergoing evaluation with operational meteorologists (Hemingway and Robbins 2020). Furthermore, the FEWS that is currently under development for the Vaisagano River in Apia, Samoa, has been developed in collaboration with risk modelers and developers of the RiskScape software and with the Samoan Ministry of Natural Resources and Environment, indicating that risk modeling will play a role in the operational FEWS. There is further interest and future work planned for developing the capabilities of risk model frameworks like RiskScape to conduct rapid impact modeling assessments for warning systems and response and recovery (Risk Modeling NZ. A).
I think there can be provided a bit of a false sense of what’s going to happen? Like people think that that’s going to be the impact, but there’s always a chance that a) the forecast, there might be an area that might get more rain than that, also we’re in a climate change, so things are just a little bit more intense now. And there might be another factor that has never been considered, and it could just be something that’s just a change in the landscape, that sort of meant that the storm might just have a bit more sting in its tail and might, you know the local experience might, it can tell you about what’s happened, but it won’t tell you about the really extreme event (Met. Research NZ. J).
This year the drought’s been so widespread that we’re really low on specific types of feed that are really common in a farmland system normally, really easy to source […] So […] that’s […] actually put the rural community at a really big risk of not being able to cope for events that otherwise would be quite easy to recover from (Agriculture/Rural NZ. A).
Thus, a low impact snow event can become high impact due to underlying vulnerability conditions caused by a separate event (e.g., drought); this is discussed further in section 3b(4), dynamic exposure and vulnerability.
This example also demonstrates the need to consider cascading hazards, as stated by Pascal et al. (2006), Potter et al. (2021), and WMO (2018).
(ii) Updating warnings
Real-time or near-real-time hazard and impact data streaming onto meteorologists’ screens are highly valuable to participating national meteorological services (NMSs), as real-time information allows meteorologists to adjust warnings on the fly (Fig. 4):
Data inputs and uses for updating warnings. Real-time or near-real-time data/information on the hazards, impacts, vulnerability, and exposure are key for updating warnings. Updating warnings starts from the initial impact warning, which leads to actions taken by the warning audience. These actions can determine what human impacts occur (e.g., injuries, fatalities, or traffic impacts). The hazard can change in space and time, also causing different impacts. Warnings can be updated with this up-to-date information. Dynamic vulnerability and dynamic exposure can also change as impacts unfold and as people take protective actions.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
the most important thing here is that [the European Weather Observer is] real-time data, and we will have this data available on our screens as forecasters wants [sic] it … So … in our view it's really important for the forecaster to have real-time data available in order to adjust warnings, for example, if things are more severe out there, … to upgrade the warning or vice versa (Met. Int. E).
This impact data can show forecasters and responders “what’s happening now versus what you predicted is going to happen” (EM. NZ. Reg. C). For example, in New Zealand, one MetService participant described how they can upgrade an orange warning to a red warning if they observe impacts that they were not expecting, “coupled with a forecast with a lot more to come, and that might then prompt you once you’ve seen those impacts to then escalate to red” (Met. NZ. K). Social media, crowdsourcing, and media reports are valuable for gathering real-time impact data, which is useful for updating warnings on the fly (Met. Int. E; Met. NZ. K; Krennert et al. 2018b), despite the limitations in quality and trustworthiness of social media and crowdsourced data (Harrison and Johnson 2019).
(iii) Verification
To continuously improve their services, warning organizations routinely verify and validate models, forecasts, and warnings. Participating NMS officials described how warnings have traditionally been verified based on the occurrence of the forecast hazard, with little investigation into what happened “on the ground” (Met. Int. A, B, D, I; Met. NZ. F). This verification involves meteorological observations to confirm the occurrence and timing of the hazard in comparison with the forecasts and warnings (Sharpe 2016; Wilson and Giles 2013). Participating NMS and EM officials indicated that verifying impacts is the next step toward improving forecasts and warnings, as shown in Fig. 5. This can help to determine if they are under or overwarned and strive for the “right buoyancy of [the] warnings” (Met. Int. E).
Verification of impact warnings relies on observed hazard and impact data. The impact warnings that were issued lead to certain actions, which then result in impacts (or no impacts). These impacts are then used to verify the warnings that were issued. The feedback loop from learning from verification to adjust procedures and impact thresholds is represented by the arrow pointing back from verification to impact warning.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
Along with observational hazard data, impact data from postevent assessments, media reports, social media, and other crowdsourcing or citizen science efforts are used for warning verification (Met. Int. D, E; Met. NZ. F). Postevent assessments offer credible, systematic ways of collecting impact data, useful for postevent analyses (Harrison et al. 2015; Pita and de Schwarzkopf 2016). For example, NMSs use assessments to determine whether they “cried wolf” (Met. Int. B). Most participating NMSs use these assessments for postevent analysis and evaluation (Met. Int. B, C, E, I). The NMSs in the USA and Austria use these data, along with crowdsourced storm reports, to build event databases that allow for event comparison, past event learnings, and future preparedness (Met. Int. E, I; Krennert et al. 2018a,b).
IFW verification challenges exist. Hemingway and Robbins (2020) questioned how verification can be done if the impacts are reduced due to the warnings themselves. These concerns were echoed by Potter et al. (2021), who suggested several approaches for verifying IFWs if and when impact data is not available. These include using predetermined, dynamic hazard-based criteria, or identifying the mitigation decisions that were made in response to the warning (Potter et al. 2021). An NZ meteorologist in our study indicated that verification is not binary (Met. NZ. K). NZ MetService meteorologists use a combination of observational hazard data, impact data from media reports, and experience in a “pragmatic” approach to assigning a verification score to the warnings. Thus, in line with Potter et al. (2021), a multifactored and multidata approach grounded in pragmatism and experience presents an alternative to verifying IFWs.
6) Decision/action
After a forecast or warning is issued, the recipients must decide on appropriate protective actions. Investigating the effectiveness of IFWs for warning recipient decision-making was out of the scope of this research, yet some notable comments were made by participants on this topic that should be acknowledged.
From just the base physical science aspect, I don’t know that … we’re at a level at least with consistent, calibrated probabilistic forecasts that we can predict those kind of events, or … consistently in a calibrated way predict what the intensity of a tornado is going to be or the size of hail (Met. Int. F).
More information on hazards and impacts such as hail size, tornado intensity, and resulting damage is needed to calibrate models for hazard forecasting (Met. Int. I). For example, reports of hail size from crowdsourcing and citizen science platforms can help calibrate radar algorithms used for forecasting precipitation types (Met. Int. I), which can then inform hazard and impact forecasts (e.g., damaging hail to crops) by overlaying the hazard forecast with asset data (e.g., crop fields, infrastructure). However, more work is still needed in this area (Met. Int. I).
IFWs may help individuals conceptually link hazards to impacts, thus increasing their risk awareness. For example, with regard to severe thunderstorm-induced asthma attacks, one NZ-based respiratory doctor indicated that people who experience severe thunderstorm-induced asthma attacks and have never experienced an asthma attack before may not realize what is happening to them, resulting in a delay of them seeking appropriate care (Health NZ. Reg. A). IFWs containing information on thunderstorm-induced asthma may benefit warning recipients by clearly drawing the links from the hazard(s) (i.e., the thunderstorm and pollen) to the impacts (i.e., asthma attacks), helping individuals to more quickly identify what may be happening to them and seek proper care.
people then can react to that in the way that they deem appropriate, and that then moves the responsibility back on the property owner, the residents to be able to react. Obviously, we’re there to help and guide and advise, but we just haven’t got the reach and the people so … you need enough bubbles of control out there that they’re able to self-manage and deal with it themselves … so really it’s get the message out as quickly as possible and let them react as they see fit. That’s sort of where our, for me anyway, our responsibilities should end, we can’t help everyone (Hyd. Gov. NZ. Reg. A).
This participant suggested having multiple “bubbles of control” across the community through self-management, rather than following prescribed actions that may not be situationally appropriate. They acknowledged this would require heavy early community engagement efforts, so communities and individuals understand their local flood risk and develop their emergency response plans for floods. This raises questions around the needs of the warning recipients and the communities, and the need to identify what they deem the most appropriate approach for warnings, as suggested by Thomalla et al. (2006), and done by Tarchiani et al. (2020) in their design of a flood IFW system in Niger. This would align with the concept of people-centered early warning systems (Basher 2006). Still, the evidence from the literature suggesting the efficacy and value of including guidance messaging cannot be denied (e.g., Weyrich et al. 2018).
In terms of data and information for the decision component of the warning value chain, the information produced from the previously described components (observation, weather forecasting, hazard forecasting, impact forecasting, and impact warning) and other social data such as past experience and behaviors of peers influence the recipients’ decision-making. Furthermore, outcomes of these decisions produce new impact information resulting from the event (Hemingway and Robbins 2020). Thus, rather than being an input, impact data can also be an output from the decision component.
Each of these components links together to complete the impact forecasting and warning value chain. Figure 6 provides a visual representation of the connected chain, starting with observation, monitoring, and detection, where meteorological services and hydrological services use raw observational hazard data. These observational hazard data are then used to formulate weather forecasts in the next component. Weather forecast data and raw observational data are then used to produce hazard forecasts, in conjunction with impact data if available. These hazard forecasts can then be combined with impact, vulnerability, and exposure data to produce impact forecasts, using models, impact-oriented discussions, or both. After an impact forecast is produced, the meteorological, hydrological, and EM services must decide on the appropriate impact warning level to issue to the appropriate audiences. Hazard, impact, vulnerability, and exposure data are all utilized to define the warning thresholds, update the warnings, and verify the warnings. Once a warning is issued, the warning audiences must decide on the appropriate action to take (if any). These actions (or lack thereof) may then result in forecast or unforecast impacts. Uncertainty begins at the beginning of the value chain with weather forecasts. From there, the uncertainty compounds, as demonstrated by the compounding uncertainty wedge, which can influence warning audiences’ decisions to act.
Warning value chain and the associated data inputs and outputs for each component, along with the activities and actors for each value chain component. Uncertainty compounds at every stage of the chain and is represented by the compounding uncertainty wedge.
Citation: Weather, Climate, and Society 14, 1; 10.1175/WCAS-D-21-0093.1
b. Need for more understanding on different types of impacts
Ranges of impacts occur from hydrometeorological hazards that can be physical (e.g., damage to infrastructure and the built environment), economic (e.g., financial losses), social/societal/cultural (e.g., traffic delays, disruptions to large events, educational services, health services), or environmental (e.g., pollution spills, wildlife endangerment). The impacts can be direct (e.g., injuries, casualties, building damage) or indirect (e.g., job loss, displacement, illness) (Lindell et al. 2006). Discussions with participants revealed a gap in modeling, forecasting, and communicating some indirect and social impacts.
1) Social impacts
Here in Argentina, football, soccer … it’s very, very important … And there were [sic] South American final soccer cup here, and the final soccer cup was between the two most important soccer clubs, and everybody was very crazy. There was a very, very heavy storm and the match had to be postponed. You cannot imagine how was [sic] that day for us, and it was just a soccer game, you know. High-level authorities were calling my boss to “tell me what’s going on, what do we have to do” you know, because people really got very, very angry. They blame us, on the match. So, what can I do? (Met. Int. D).
This example indicates the importance of identifying cultural impacts from hydrometeorological hazards and including this in communications. Further, messaging needs to be more culturally sensitive/aware and empathetic. As highlighted by Ayeb-Karlsson et al. (2019), several cultural and social limitations exist in disaster preparedness, particularly warning response, and there is a need to understand the cultural contexts influencing preparedness. Doing so would further align with the concept of people-centered EWSs (Basher 2006).
A shift is occurring in the NZ risk modeling space toward social impact assessments in response to needs identified by the EM sector. An NZ-based risk modeler described the importance of considering social impacts beyond just the direct, physical impacts, to provide a fuller picture (Risk Modeling NZ. B). For example, direct impacts such as power and other critical service outages may cause population displacement, which is often not modeled.
Indeed, an official from NZ’s National Emergency Management Agency (NEMA) echoed the need for documenting indirect impacts (EM. Gov. NZ. Nat. G). While NEMA collects impact and loss data for the Sendai Framework (United Nations Office for Disaster Risk Reduction 2015), the face value of the impact/loss data lacks the deeper meaning of the indirect, potentially longer-term, impacts because “recording a road outage as a 1 or a 0, you know it was out or it wasn’t out, doesn’t give you any scale of duration, the impact, lost productivity, stress on people’s lives” (EM. Gov. NZ. Nat. G).
Following flood events in NZ, risk modelers and other researchers have disseminated postevent surveys to the affected people. Challenges arose around collecting and using social impact data for modeling purposes (Risk Modeling NZ. C). A range of physical and nonphysical impacts occurred after the initial event, such as injuries (physical), and anxiety and stress (nonphysical). While they were able to capture some of this information, data collection was not well planned, thus it did not result in a high-quality dataset useful for modeling (Risk Modeling NZ. C). Conducting well-planned longitudinal surveys in collaboration with more experts such as social scientists were suggested to increase the usefulness of the surveys (Risk Modeling NZ. C). Furthermore, this participant suggested that risk modelers may not be the best suited for this type of social impact modeling as their expertise lies in modeling buildings and infrastructure. This highlights the need for multidisciplinary work for IFWs (Merz et al. 2020).
2) Health impacts
Health impacts appear to also have gone largely unnoticed in the IFW literature, aside from the health impacts of extreme heat (IFRC and Met Office 2020; Pascal et al. 2006; WMO 2015). Recent events have brought the risk of thunderstorm-induced asthma attacks to light in Australia and New Zealand (Sabih et al. 2020; Thien et al. 2018). Before the 2016 Melbourne epidemic thunderstorm asthma event, this risk was not forecast or warned for. Following this event recommendations included developing an epidemic thunderstorm asthma forecast (Hew et al. 2017; Thien et al. 2018), which is now provided by the Australian Bureau of Meteorology (Met. Int. B; Bannister et al. 2021). A similar event occurred on a smaller scale in Waikato, NZ in 2017 with no advanced warning (Health NZ Reg. A; Sabih et al. 2020). There has thus been a push to capture such severe weather health impacts (Aitsi-Selmi and Murray 2016), which must then be integrated into IFWs. Health impacts from poor air quality due to pollution (Gao et al. 2015), wildfires (Cisneros and Schweizer 2018), etc., are also increasingly important and must be considered when designing IFWs.
3) Urban and rural impacts
exposure and vulnerability are the two key questions and for some warnings, a short lead time is fine, but for others, it’s not, and it very much depends on the user. A farmer for a snow warning is going to need a lot more time to get stock down out of the high country than you as an individual getting a wind warning and needing to go down and tie up the trampoline, right? So, if you want to tie down the trampoline, you can do that in a couple hours when you get home, but you know, if you’re a farmer, you might need to know that there’s snow coming the day after tomorrow to move the stock (Met. NZ. K).
This relates to the need for vulnerability and exposure to be included in IFWs and the identification of appropriate audiences’ relevant trigger points (or thresholds) (Zhang et al. 2021). It is not enough to collect only past event impact data and expect the impacts to be the same for future events or different populations. However, as Potter et al. (2021) found, vulnerability and exposure are often overlooked when designing EWSs. This may be due to challenges in capturing the dynamic nature of vulnerability and exposure, discussed next.
4) Dynamic exposure and vulnerability
Exposure and vulnerability change over space and time (World Bank 2014). Thus, up-to-date datasets for this information are critical for accurately representing the level of risk (Met. Int. A, D; Met. NZ. F, K). If exposure data are not current, the risk assessment will be inaccurate, resulting in improper warning buoyancy (i.e., overwarning or underwarning will occur) (Met. Int. A, B). Thus, a U.K.-based meteorological risk specialist hopes “we all recognize the importance of updated vulnerability and exposure data” (Met. Int. A). However, gathering and integrating dynamic vulnerability and exposure data is a challenge for participating NMSs such as the Met Office that are determining how best to maintain up-to-date data and integrate them into an IFW system (Met. Int. A).
In recent years, research has emerged investigating the dynamic nature of exposure and vulnerability (Terti et al. 2015) and methods for integrating dynamic exposure and vulnerability into risk analyses (Gallina et al. 2016; Shabou et al. 2017; Wilhelmi and Morss 2013). Our findings echo those from Merz et al. (2020) with regard to the role of dynamic exposure and vulnerability in impact forecasting, and the need to integrate dynamic exposure and vulnerability into models through collaboration across sectors (e.g., social science, engineering, and natural sciences). Other options commonly used include visual overlays for integrated multihazard decision support platforms supporting impact-oriented discussions. There is a need to identify the potential sources, creators, and custodians of dynamic exposure and vulnerability data and investigate how it can be accessed and used for IFWs.
4. Conclusions
This exploratory study identified the uses and gaps for hazard, impact, exposure, and vulnerability data for implementing severe weather IFW systems within the warning value chain framework presented by Golding et al. (2019). The qualitative nature of data collection and analysis herein limits the generalizability of results beyond the participants. However, the qualitative approach offers an in-depth understanding of a problem not readily available from quantitative approaches (Blumer 1969; Miles and Huberman 1994; Patton 2002).
Our findings support existing research pointing to the need for creating, gathering, and using impact, vulnerability, and exposure data for IFW systems. This study builds on existing research by identifying where each data type (hazard, impact, vulnerability, and exposure) can be used for each warning value chain component and offers examples of alternative approaches for impact forecasting and for defining impact thresholds. For example, models provide an objective approach while impact-oriented discussions provide subjective and flexible approaches. These approaches should be used complementarily.
Findings from this research provide new insight into a growing need to identify, model, and warn for social and health impacts, which have typically taken a back seat to modeling and forecasting for physical and infrastructure impacts. The type of impacts to be forecast and warned for depends on the intended audiences of the warnings.
While this study identified the uses and gaps for hazard, impact, vulnerability, and exposure data, questions remain around who collects, creates, stores, and manages these data (particularly impact, exposure, and vulnerability). Future research should explore mapping the various sources, creators, users, and custodians of the relevant data needed for IFWs.
This study also has a heavy focus on hydrometeorological hazards and impacts due to the experiences of most of the research participants. Floods are the most frequent and costly hazard in NZ (Rouse 2011) and thus were the most common examples that NZ participants drew from during interviews. Further research should verify the findings of this research against other hydrometeorological hazards both within and outside of New Zealand.
Note that an update on the WMO Guidelines on Multi-Hazard Impact-Based Forecast and Warning Services is anticipated to be released soon. A more recent resource providing further guidance on IFW implementation was published by the International Federation of Red Cross and Red Crescent Societies and is cited as IFRC and Met Office (2020) throughout this paper for further support.
Shadow evacuation occurs when people who are not at risk or are not in the official evacuation zone evacuate unnecessarily (Zeigler et al. 1981).
Not all hydrological services are provided by councils; this depends on the governance structure of the location under study. In the case of New Zealand, hydrological services are provided by councils as per the Resource Management Act (New Zealand Government 1991).
For the purposes of this study, “hazard data” will refer to meteorological and hydrological data. For example, wind speed and rainfall data may be considered to be meteorological hazard data, whereas river level and river flow data are considered to be hydrological hazard data.
RiskScape is an open-source risk modeling software that was codeveloped in New Zealand by GNS Science and NIWA (see Schmidt et al. 2011).
Acknowledgments.
This research was funded by New Zealand’s Ministry of Business, Innovation and Employment via the GNS Science Strategic Science Investment Fund and Resilience to Nature’s Challenges Kia manawaroa—Ngā Ākina o Te Ao Tūrua and is an output of a doctoral thesis completed at Massey University. The authors are thankful to the participants of the research for providing their invaluable time and insights and to the NZ MetService for their insightful conversations and advice throughout this research. The authors also thank the reviewers, whose feedback greatly improved this paper.
Data availability statement.
Anonymized data are available by contacting the authors.
REFERENCES
Aitsi-Selmi, A., and V. Murray, 2016: Protecting the health and well-being of populations from disasters: Health and health care in the Sendai framework for disaster risk reduction 2015-2030. Prehospital Disaster Med., 31, 74–78, https://doi.org/10.1017/S1049023X15005531.
Aldridge, T., O. Gunawan, R. J. Moore, S. J. Cole, and D. Price, 2016: A surface water flooding impact library for flood risk assessment. FLOODrisk 2016—Third European Conf. on Flood Risk Management, Lyon, France, IRSTEA, Deltares, HR Wallingford, and Samui, 18006, https://doi.org/10.1051/e3sconf/20160718006.
Ayeb-Karlsson, S., D. Kniveton, T. Cannon, K. van der Geest, I. Ahmed, E. M. Derrington, E. Florano, and D. O. Opondo, 2019: I will not go, I cannot go: Cultural and social limitations of disaster preparedness in Asia, Africa, and Oceania. Disasters, 43, 752–770, https://doi.org/10.1111/disa.12404.
Baker, E. J., 1991: Hurricane evacuation behavior. Int. J. Mass Emerg. Disasters, 9, 287–310.
Bannister, T., and Coauthors, 2021: A pilot forecasting system for epidemic thunderstorm asthma in southeastern Australia. Bull. Amer. Meteor. Soc., 102, E399–E420, https://doi.org/10.1175/BAMS-D-19-0140.1.
Basher, R., 2006: Global early warning systems for natural hazards: Systematic and people-centered. Philos. Trans. Roy. Soc., A364, 2167–2182, https://doi.org/10.1098/rsta.2006.1819.
Bauer, P., A. Thorpe, and G. Brunet, 2015: The quiet revolution of numerical weather prediction. Nature, 525, 47–55, https://doi.org/10.1038/nature14956.
Bergin, M., 2011: NVivo 8 and consistency in data analysis: Reflecting on the use of a qualitative data analysis program. Nurse Res., 18, 6–12, https://doi.org/10.7748/nr2011.04.18.3.6.c8457.
Blumer, H., 1969: Symbolic Interactionism: Perspective and Method.1st ed. University of California Press, 224 pp.
Brotzge, J. A., S. E. Nelson, R. L. Thompson, and B. T. Smith, 2013: Tornado probability of detection and lead time as a function of convective mode and environmental parameters. Wea. Forecasting, 28, 1261–1276, https://doi.org/10.1175/WAF-D-12-00119.1.
Casteel, M., 2016: Communicating increased risk: An empirical investigation of the national weather service’s impact-based warnings. Wea. Climate Soc., 8, 219–232, https://doi.org/10.1175/WCAS-D-15-0044.1.
Casteel, M., 2018: An empirical assessment of impact based tornado warnings on shelter in place decisions. Int. J. Disaster Risk Reduct., 30, 25–33, https://doi.org/10.1016/j.ijdrr.2018.01.036.
Charmaz, K., 2006: Constructing Grounded Theory.SAGE, 208 pp.
Ching, P. K., V. Carr de los Reyes, M. N. Sucaldito, and E. Tayag, 2015: An assessment of disaster-related mortality post-Haiyan in Tacloban City. West. Pac. Surveill. Response J., 6 (Suppl. 1), 34–38, https://doi.org/10.5365/wpsar.2015.6.2.HYN_005.
Chun Tie, Y., M. Birks, and K. Francis, 2019: Grounded theory research: A design framework for novice researchers. SAGE Open Med., 7, 1–8, https://doi.org/10.1177/2050312118822927.
Cisneros, R., and D. W. Schweizer, 2018: The efficacy of news releases, news reports, and public nuisance complaints for determining smoke impacts to air quality from wildland fire. Air Qual. Atmos. Health, 11, 423–429, https://doi.org/10.1007/s11869-018-0545-9.
Cole, S. J., and Coauthors, 2016a: Natural Hazards Partnership surface water flooding hazard impact model: Phase 2 final report. Natural Hazards Partnership Rep., 19 pp., http://www.naturalhazardspartnership.org.uk/wp-content/uploads/2016/10/NHP-HIM-Surface-Water-Flooding-Phase-2-Final-Report.pdf.
Cole, S. J., R. J. Moore, S. C. Wells, and P. S. Mattingley, 2016b: Real-time forecasts of flood hazard and impact: Some UK experiences. FLOODrisk 2016—Third European Conf. on Flood Risk Management, Lyon, France, IRSTEA, Deltares, HR Wallingford, and Samui, 18015, https://doi.org/10.1051/e3sconf/20160718015.
Crawford, M. H., K. Crowley, S. H. Potter, W. S. A. Saunders, and D. M. Johnston, 2018: Risk modelling as a tool to support natural hazard risk management in New Zealand local government. Int. J. Disaster Risk Reduct., 28, 610–619, https://doi.org/10.1016/j.ijdrr.2018.01.011.
Cullen, S. M., K. Christensen, and C. Price, 2017: Rangitāiki river scheme review – April 2017 flood event. RRSR Final Rep., 162 pp., https://www.boprc.govt.nz/media/773858/rangitaiki-river-scheme-review-april-2017-flood-event-final-report-as-supplied-to-by-of-plenty-regional-council.pdf.
Farnell, C., T. Rigo, and N. Pineda, 2017: Lightning jump as a nowcast predictor: Application to severe weather events in Catalonia. Atmos. Res., 183, 130–141, https://doi.org/10.1016/j.atmosres.2016.08.021.
Fernandez, W. D., 2005: The grounded theory method and case study data in IS research: Issues and design. Information Systems Foundations: Constructing and Criticizing, D. N. Hart and S. D. Gregor, Eds., ANU Press, 43–59, https://doi.org/10.22459/ISF.06.2005.05.
Flowerdew, J., K. Mylne, C. Jones, and H. Titley, 2013: Extending the forecast range of the UK storm surge ensemble. Quart. J. Roy. Meteor. Soc., 139, 184–197, https://doi.org/10.1002/qj.1950.
Gallina, V., S. Torresan, A. Critto, A. Sperotto, T. Glade, and A. Marcomini, 2016: A review of multi-risk methodologies for natural hazards: Consequences and challenges for a climate change impact assessment. J. Environ. Manage., 168, 123–132, https://doi.org/10.1016/j.jenvman.2015.11.011.
Gao, M., S. K. Guttikunda, G. R. Carmichael, Y. Wang, Z. Liu, C. O. Stanier, P. E. Saide, and M. Yu, 2015: Health impacts and economic losses assessment of the 2013 severe haze event in Beijing area. Sci. Total Environ., 511, 553–561, https://doi.org/10.1016/j.scitotenv.2015.01.005.
Garcia, C., and C. J. Fearnley, 2012: Evaluating critical links in early warning systems for natural hazards. Environ. Hazards, 11, 123–137, https://doi.org/10.1080/17477891.2011.609877.
Glaser, B., and A. Strauss, 1967: The Discovery of Grounded Theory: Strategies for Qualitative Research.Aldine Publishing Company, 271 pp.
Global Facility for Disaster Reduction and Recovery, 2016: Implementing multi-hazard impact-based forecast and warning services. WMO and GFDRR Rep., 76 pp., https://www.gfdrr.org/sites/default/files/publication/Report%20on%20Implementing%20Multi%20Hazard%20Impact%20Based%20Forecast%20and%20Warning%20Services%2017-02-10%20Final.pdf.
Gneiting, T., and A. E. Raftery, 2005: Weather forecasting with ensemble methods. Science, 310, 248–249, https://doi.org/10.1126/science.1115255.
Golding, B., M. Mittermaier, C. Ross, B. Ebert, S. Panchuk, A. Scolobig, and D. M. Johnston, 2019: A value chain approach to optimising early warning systems. United Nations Office for Disaster Risk Reduction Rep., 30 pp., https://www.preventionweb.net/files/65828_f212goldingetalvaluechain.pdf.
Golnaraghi, M., 2012: An overview: Building a global knowledge base of lessons learned from good practices in multi-hazard early warning systems. Institutional Partnerships in Multi-Hazard Early Warning Systems: A Compilation of Seven National Good Practices and Guiding Principles, M. Golnaraghi, Ed., Springer, 1–8.
Harrison, J., C. McCoy, K. Bunting-Howarth, H. Sorensen, K. Williams, and C. Ellis, 2014: Evaluation of the National Weather Service impact-based warning tool. Great Lakes Sea Grant Social Science Network Rep. WISCU-S-14-001, 37 pp., https://publications.aqua.wisc.edu/product/evaluation-of-the-national-weather-service-impact-based-warning-tool/.
Harrison, S. E., and P. A. Johnson, 2016: Crowdsourcing the disaster management cycle. Int. J. Inf. Syst. Crisis Response Manage., 8, 17–40, https://doi.org/10.4018/IJISCRAM.2016100102.
Harrison, S. E., and P. A. Johnson, 2019: Challenges in the adoption of crisis crowdsourcing and social media in Canadian emergency management. Gov. Inf. Quart., 36, 501–509, https://doi.org/10.1016/j.giq.2019.04.002.
Harrison, S. E., A. Silver, and B. Doberstein, 2015: Post-storm damage surveys of tornado hazards in Canada: Implications for mitigation and policy. Int. J. Disaster Risk Reduct., 13, 427–440, https://doi.org/10.1016/j.ijdrr.2015.08.005.
Harrison, S. E., S. H. Potter, R. Prasanna, E. E. H. Doyle, and D. Johnston, 2020: Volunteered Geographic Information for people-centred severe weather early warning: A literature review. Australas. J. Disaster Trauma Stud., 24, 3–21.
Harrison, S. E., S. H. Potter, R. Prasanna, E. E. H. Doyle, and D. Johnston, 2021: ‘Where oh where is the data?’: Identifying data sources for hydrometeorological impact forecasts and warnings in Aotearoa New Zealand. Int. J. Disaster Risk Reduct., 66, 102619, https://doi.org/10.1016/j.ijdrr.2021.102619.
Hemingway, R., and J. Robbins, 2020: Developing a hazard-impact model to support impact-based forecasts and warnings: The Vehicle OverTurning (VOT) model. Meteor. Appl., 27, e1819, https://doi.org/10.1002/met.1819.
Hew, M., M. F. Sutherland, F. Thien, and R. O’Hehir, 2017: The Melbourne thunderstorm asthma event: Can we avert another strike? Intern. Med. J., 47, 485–487, https://doi.org/10.1111/imj.13413.
Horita, F. E. A., J. Porto de Albuquerque, L. C. Degrossi, E. M. Mendiondo, and J. Ueyama, 2015: Development of a spatial decision support system for flood risk management in Brazil that combines volunteered geographic information with wireless sensor networks. Comput. Geosci., 80, 84–94, https://doi.org/10.1016/j.cageo.2015.04.001.
Hosterman, H. R., J. K. Lazo, J. M. Sprague-Hilderbrand, and J. E. Adkins, 2019: Using the National Weather Service’s impact-based decision support services to prepare for extreme winter storms. J. Emerg. Manage., 17, 455–467, https://doi.org/10.5055/jem.2019.0439.
Hughes, J., and S. Jones, 2003: Reflections on the use of grounded theory in interpretive information systems research. Proc. 11th European Conf. on Information Systems, Naples, Italy, Association for Information Systems Council, 833–845, http://aisel.aisnet.org/ecis2003/62.
IFRC and Met Office, 2020: The future of forecasts: Impact-based forecasting for early action. International Federation of Red Cross and Red Crescent Societies Rep., 82 pp., https://www.forecast-based-financing.org/wp-content/uploads/2020/09/Impact-based-forecasting-guide-2020.pdf.
International Science Council, 2021: Unleashing science: Delivering missions for sustainability. ISC Rep., 49 pp., https://council.science/wp-content/uploads/2020/06/202108_Unleashing-Science_Final.pdf.
Kaltenberger, R., A. Schaffhauser, and M. Staudinger, 2020: “What the weather will do”—Results of a survey on impact-oriented and impact-based warnings in European NMHSs. Adv. Sci. Res., 17, 29–38, https://doi.org/10.5194/asr-17-29-2020.
Kelle, U., 2005: “Emergence” vs. “forcing” of empirical data? A crucial problem of “grounded theory” reconsidered. Soc. Res., 6, 133–156, https://doi.org/10.17169/fqs-6.2.467.
Kotsuki, S., K. Kurosawa, S. Otsuka, K. Terasaki, and T. Miyoshi, 2019: Global precipitation forecasts by merging extrapolation-based nowcast and numerical weather prediction with locally optimized weights. Wea. Forecasting, 34, 701–714, https://doi.org/10.1175/WAF-D-18-0164.1.
Kox, T., H. Kempf, C. Lüder, R. Hagedorn, and L. Gerhold, 2018a: Towards user-orientated weather warnings. Int. J. Disaster Risk Reduct., 30, 74–80, https://doi.org/10.1016/j.ijdrr.2018.02.033.
Kox, T., C. Lüder, and L. Gerhold, 2018b: Anticipation and response: Emergency services in severe weather situations in Germany. Int. J. Disaster Risk Sci., 9, 116–128, https://doi.org/10.1007/s13753-018-0163-z.
Krennert, T., R. Kaltenberger, G. Pistotnik, A. M. Holzer, F. Zeiler, and M. Stampfl, 2018a: Trusted spotter network Austria—A new standard to utilize crowdsourced weather and impact observations. Adv. Sci. Res., 15, 77–80, https://doi.org/10.5194/asr-15-77-2018.
Krennert, T., G. Pistotnik, R. Kaltenberger, and C. Csekits, 2018b: Crowdsourcing of weather observations at national meteorological and hydrological services in Europe. Adv. Sci. Res., 15, 71–76, https://doi.org/10.5194/asr-15-71-2018.
Lamb, S., D. Walton, K. Mora, and J. Thomas, 2012: Effect of authoritative information and message characteristics on evacuation and shadow evacuation in a simulated flood event. Nat. Hazards Rev., 13, 272–282, https://doi.org/10.1061/(ASCE)NH.1527-6996.0000070.
LeClerc, J., and S. Joslyn, 2015: The cry wolf effect and weather-related decision making. Risk Anal., 35, 385–395, https://doi.org/10.1111/risa.12336.
Lehmann, H., 2010: Research method: Grounded theory for descriptive and exploratory case studies. The Dynamics of International Information Systems, Springer Science+Business Media, 53–65, https://doi.org/10.1007/978-1-4419-5750-4_5.
Lindell, M. K., C. S. Prater, and R. W. Perry, 2006: Hazard, vulnerability, and risk analysis. Fundamentals of Emergency Management, Emergency Management Institute, 153–191.
Mack, N., C. Woodsong, K. M. MacQueen, G. Guest, and E. Namey, 2005: Qualitative Research Methods: A Data Collector’s Field Guide. Family Health International, 120 pp.
Mackie, B., 2013: Warning fatigue: Insights from the Australian Bushfire Context. Ph.D. thesis, Media and Communication, University of Canterbury, 294 pp., https://ir.canterbury.ac.nz/handle/10092/9029.
Majumdar, S. J., and Coauthors, 2021: Multiscale forecasting of high-impact weather: Current status and future challenges. Bull. Amer. Meteor. Soc., 102, E635–E659, https://doi.org/10.1175/BAMS-D-20-0111.1.
Matavire, R., and I. Brown, 2013: Profiling grounded theory approaches in information systems research. Eur. J. Inf. Syst., 22, 119–129, https://doi.org/10.1057/ejis.2011.35.
Merz, B., and Coauthors, 2020: Impact forecasting to support emergency management of natural hazards. Rev. Geophys., 58, e2020RG000704, https://doi.org/10.1029/2020RG000704.
MetService, 2019: Changes to the official severe weather warnings. MetService Doc., 4 pp., https://about.metservice.com/assets/Watches-and-Warnings/PR-Changes-to-the-official-Severe-Weather-Warnings-140519.pdf.
MetService, 2021: Warnings. Accessed 9 June 2021, https://www.metservice.com/warnings/home#:∼:text=MetService%20will%20issue%20a% 20Heavy,period%20of%2014s%20or%20more.
Miles, M. B., and A. M. Huberman, 1994: Qualitative Data Analysis: An Expanded Sourcebook. 2nd ed. Sage, 338 pp.
Morss, R. E., J. L. Demuth, J. K. Lazo, K. Dickinson, H. Lazrus, and B. H. Morrow, 2016: Understanding public hurricane evacuation decisions and responses to forecast and warning messages. Wea. Forecasting, 31, 395–417, https://doi.org/10.1175/WAF-D-15-0066.1.
Morss, R. E., C. L. Cuite, J. L. Demuth, W. K. Hallman, and R. L. Shwom, 2018: Is storm surge scary? The influence of hazard, impact, and fear-based messages and individual differences on responses to hurricane risks in the USA. Int. J. Disaster Risk Reduct., 30, 44–58, https://doi.org/10.1016/j.ijdrr.2018.01.023.
National Research Council, 2006: Completing the Forecast: Characterizing and Communicating Uncertainty for Better Decisions Using Weather and Climate Forecasts.National Academies Press, 124 pp., https://doi.org/10.17226/11699.
Neal, R. A., P. Boyle, N. Grahame, K. Mylne, and M. Sharpe, 2014: Ensemble based first guess support towards a risk-based severe weather warning service. Meteor. Appl., 21, 563–577, https://doi.org/10.1002/met.1377.
New Zealand Government, 1991: Resource Management Act 1991. Ministry for the Environment Doc., 842 pp., https://www.legislation.govt.nz/act/public/1991/0069/latest/DLM230265.html.
Obermeier, H., and M. R. Anderson, 2014: Verification and analysis of impact-based tornado warnings in the central region of the National Weather Service. Electron. J. Severe Storms Meteor., 10 (1), https://ejssm.org/archives/wp-content/uploads/2021/09/vol10-1.pdf.
Pascal, M., and Coauthors, 2006: France’s heat health watch warning system. Int. J. Biometeor., 50, 144–153, https://doi.org/10.1007/s00484-005-0003-x.
Patton, M. Q., 2002: Qualitative Research and Evaluation Methods.3rd ed. Sage, 598 pp.
Pita, G. L., and M. L. A. de Schwarzkopf, 2016: Urban downburst vulnerability and damage assessment from a case study in Argentina. Nat. Hazards, 83, 445–463, https://doi.org/10.1007/s11069-016-2323-z.
Poolman, E. R., 2014: A probabilistic impact-focussed early warning system for flash floods in support of disaster management in South Africa. Ph.D. thesis, University of Pretoria, 182 pp., http://hdl.handle.net/2263/44331.
Potter, S. H., P. V. Kreft, P. Milojev, C. Noble, B. Montz, A. Dhellemmes, R. J. Woods, and S. Gauden-Ing, 2018: The influence of impact-based severe weather warnings on risk perceptions and intended protective actions. Int. J. Disaster Risk Reduct., 30, 34–43, https://doi.org/10.1016/j.ijdrr.2018.03.031.
Potter, S. H., S. E. Harrison, and P. Kreft, 2021: The benefits and challenges of implementing impact-based severe weather warning systems: Perspectives of weather, flood, and emergency management personnel. Wea. Climate Soc., 13, 303–314, https://doi.org/10.1175/WCAS-D-20-0110.1.
Ripberger, J. T., C. L. Silva, H. C. Jenkins-Smith, and M. James, 2015: The influence of consequence-based messages on public responses to tornado warnings. Bull. Amer. Meteor. Soc., 96, 577–590, https://doi.org/10.1175/BAMS-D-13-00213.1.
Ross, J., D. LaDue, and J. Correia, 2015: Motivators and important factors influencing tornado decisions in Oklahoma during May 2013 . 10th Symp. on Societal Applications: Policy, Research and Practice, Phoenix, AZ, Amer. Meteor. Soc., 7B.4, https://ams.confex.com/ams/95Annual/webprogram/Manuscript/Paper262354/For%20PDF%20Julia%20Ross%27%20REU%20Project%20Paper.pdf.
Rouse, H., 2011: Flood risk management research in New Zealand: Where are we, and where are we going? GNS Science Rep. 2012/04, 77 pp.
Sabih, A., C. Russell, and C. L. Chang, 2020: Thunderstorm-related asthma can occur in New Zealand. Respirol. Case Rep., 8, e00655, https://doi.org/10.1002/rcr2.655.
Saima, H., J. Jaafar, S. Belhaouari, and T. A. Jillani, 2011: Intelligent methods for weather forecasting: A revi