Residents’ Perception of Flood Prediction Products: The Study of NASA’s Satellite Enhanced Snowmelt Flood Prediction

Abstract

In the context of emergency management, individual or household decisions to engage in risk mitigation behaviors are widely recognized to be influenced by a benefit–cost perception (perceived applied value (PAV) vs. perceived economic value (PEV), respectively). To better understand how such decisions are made, we conducted a mail survey (N = 211) of households living in the Red River of the North Basin, North Dakota, in 2018. The survey is aimed at understanding the overall experience of households with flooding and their behavior toward advanced protective strategies against future floods by analyzing household PEV—their willingness to pay for the National Aeronautics and Space Administration’s (NASA) Satellite Enhanced Snowmelt Flood Prediction system. This paper presents a mediation model in which various predictors (flood risk, experience, flood knowledge, flood risk perception, flood preparedness, flood mitigation, and flood insurance) are analyzed in relation to the PAV of the new Satellite Enhanced Snowmelt Flood Predictions in the Red River of the North Basin, which, in turn, may shape the PEV of this product. We discuss the potential implications for both the emergency management research community and professionals regarding the application of advanced risk mitigation technologies to help protect and sustain communities across the country from floods and other natural disasters. This paper provides a greater understanding of the economic and social aspects of sustainability in the context of emergency management and community development.

1. Introduction

In the United States, floods are one of the most common [1] and costliest [1,2] types of natural hazards. The use of water resources and water systems for transportation and urbanization, along with other human activities on floodplains, are among some of the factors that have contributed to flood-related damage [3]. The Red River Basin (RRB) in the north-central region of the country is a prime example of a flood-prone area in the U.S. [4,5]. The RRB lies within Canada (20%) and the United States (80%) [6], with parts of it, including the Red River of the North at Grand Forks, experiencing frequent floods [7]. Due to the distribution of waterways in the basin, the region significantly benefits from agriculture production [8], making it critical to investigate proper flood mitigation and preparedness strategies to reduce environmental vulnerabilities in this area.

Currently, the RRB’s flood forecasts are provided by the National Oceanic and Atmospheric Administration (NOAA) National Weather Service’s Office of Water Prediction (OWP). During periods with potential snowmelt, the OWP’s SNOw Data Assimilation System (SNODAS) [9] uses a computer model to provide daily snow estimates. While SNODAS can use ground, airplane, and satellite snow observations, it does not currently use passive microwave observations that indicate how much water is in the snow. For the RRB, there is considerable variability in snow due to the flat landscape and strong winds that redistribute snow and limited ground observations available to support the SNODAS model. Other approaches such as airborne gamma radiation snow surveys cover larger areas but typically only occur about two times each winter due to the cost and availability of aircraft. Because there are few snow observations in the RRB [10], flood forecasting is problematic [11]. For example, in 2013, NOAA over-predicted peak snowmelt flow in the RRB by 70% and proactive protective measures costing USD ~2 million were needlessly implemented.

Satellite passive microwave observations provide daily snow observations at ~25 km spatial resolutions globally including the RRB. The snow information can be provided day and night independent of cloud cover, small-scale snow variability, or hazardous weather conditions. Tuttle et al. [12] and Schroeder et al. [13] compared the satellite passive microwave snow observations in the study region. Their findings indicate that the satellite snow data provide a consistent and reliable measure of snow in the RRB that NOAA forecasters can use to improve snowmelt flood estimates.

Given the frequency and consequences of floods in the United States, as highlighted by existing research, this paper focuses on the application of a particular flood prediction product developed from National Aeronautics and Space Administration (NASA) satellite observations. The flood prediction product provides estimates of the snowpack water and snowmelt from the satellite passive microwave, Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), which are used to enhance NOAA’s operational snowmelt flood forecasts in the RRB [12,13].

In this paper, we sought the opinions of RRB residents about the economic value of this NASA flood prediction product. Since residents do not directly purchase this product but benefit from its outcomes, this paper utilized the willingness-to-pay approach to assess how residents value this product using the contingent valuation (CV) method [14]. The CV method is used to evaluate how individuals value products, such as public goods, that do not have a marketed value [15]. This method is consistent with the approach taken by prior research conducted under this funded project (see [4]).

The findings of this study contribute to the economic and social pillars of sustainability. First, by understanding how RRB residents value the NASA product, we can gain insight into the level of financial support for the product, ultimately helping communities prepare for future floods through this product and reduce the response and recovery costs in their communities (economic pillar). Secondly, the support provided for such flood risk reduction tools (i.e., NASA product) can strengthen communities’ power in accurately predicting flooding and develop better action plans for at-risk regions (social pillar).

1.1. Theoretical and Applied Frameworks

At the individual level, mitigation has long been described as a process involving several steps ranging from risk assessment, evaluation of the array of mitigating behaviors and strategies, estimation of the potential impact of such behaviors and strategies, and final decision about their adoption [16]. In a similar fashion, many scholars have referred to this decision-making process for the adoption of hazard adjustments as a benefit–cost analysis involving many complex and interrelated aspects, such as the efficacy and versatility (utility for other purposes) of the adjustments (which relates to the potential benefit represented by the PAV in this study) and its economic cost, time requirements, or required knowledge and skills (which involves the cost represented by the PEV in this study) [17,18]. Benefit–cost analysis is commonly used by government agencies and researchers to evaluate disaster mitigation projects. For example, recent research by Douthat et al. looked at the intersection of equity and benefit–cost analysis methods with respect to flood mitigation projects [19]. Government agencies, such as the Federal Emergency Management Agency (FEMA), have utilized benefit–cost analyses for mitigation projects to assess whether these projects should be funded [20].

In this research, we investigate several dimensions of the mitigation process, particularly examining the connection between the process of the individual risk assessment and previous risk behavior and the benefit–cost analysis of the protective actions. In much of the recent literature, these components of the disaster risk reduction process have been investigated individually (i.e., the effect of the individual risk assessment in the willingness to mitigate or the benefit–cost analysis and its connection with actual mitigation decision making) (e.g., [21]). In this contribution we look at the mitigation process from an integrative and comprehensive standpoint and evaluate how individual risk assessment and previous risk-related behaviors shape the benefit–cost analysis of a public mitigation strategy to reduce flood risk such as the NASA’s Satellite Enhanced Snowmelt Flood Predictions in the Red River of the North Basin.

The Protective Action Decision Model (PADM) has been widely adopted to analyze mediation effects in the context of understanding how individuals respond to disasters [18]. PADM emphasizes the pre-decision processes that precede protective actions. It also highlights the role of psychological and experiential factors, such as hazard knowledge, disaster experience, risk perception, and protective behavior perception in mediating the relationship between risk information and protective behaviors. For this study, we employed a modified PADM to examine the mediating effect of PAV on the relationship between risk-related variables and willingness to pay for the NASA flood prediction product (i.e., PEV). We, therefore, hypothesize (Model 1, H1) that the relationship between people’s flood risk assessment and their previous behavior (flood risk experience, flood risk, flood risk knowledge, flood risk perception, application of preparedness or mitigation strategies, and having flood insurance) and the perceived economic value (PEV) of the satellite-enhanced snowmelt flood prediction developed by NASA can be understood as a case of mediation [22,23]. We believe, based on previous literature revealing that the decision-making process to adopt risk reduction behaviors derives from a previous benefit–cost analysis [17,18], that the PEV of the product (expressed through WTP) is mediated by the subjective evaluation of the potential applicability of the product. In other words, with this study we test the hypothesis that people with higher degrees of self-evaluated risk and who have already undertaken flood mitigation and preparedness strategies would be willing to pay (more) for the product, mediated by their subjective evaluation of its quality and applicability. In addition, we hypothesize (Model 2, H2) that this benefit–cost analysis and its two components—(1) perceived applied value (PAV, benefit) and (2) PEV (cost)—will determine the overall support of citizens for the application of the NASA product. The conceptual model is depicted in Figure 1.

1.2. Risk Assessment (Risk-Related Variables)

This section discusses the theoretical foundation for the variables included in the conceptual framework for understanding residents’ perception of the NASA product. We use a set of risk assessment variables (defined below) that can explain how individuals assess the benefit–cost of the NASA product and add to the existing research on the economic value of improved flood prediction systems (see [24]).

Flood Risk. This variable focuses on the actual risk of flood hazards, free from any human bias. Existing research has studied individuals’ proximity to risk areas to assess individuals’ flood risk [25]. Oftentimes, floodplain locations are used as the flood risk determinant [26], specifically since floodplain developments have been influencing flood risk [27]. In this study, this variable is measured by asking respondents if they know their residence is in a Federal Emergency Management Agency (FEMA) flood zone.

Flood Risk Perception. This variable assesses whether individuals believe they are at risk of flooding. Many studies (such as [28,29,30]) use this variable to gain a deeper understanding of complex factors influencing how individuals perceive flood risk. Previous research in the context of hurricane evacuations suggests that individuals who perceive a higher risk of flooding are more likely to evacuate during hurricanes [31]. We further extend the findings of Lazo et al. [31] to see if there is any correlation between flood risk perception and perception of the NASA product. In our study, the variable flood risk perception is depicted by individuals’ perception of the potential disruptive impact of flooding events (e.g., damage to homes or injury).

Flood Experience. The variable flood experience in this study reflects on residents’ previous flooding experience (e.g., flood in the community and living in a home damaged by a previous flood). Existing research has conceptualized that prior hazard experience is correlated with risk perception [25]. Thus, it is assumed that people with previous flood experience have higher risk perceptions and are more likely to invest in flood prediction strategies.

Flood Knowledge. Previous research shows that the lack of knowledge about the potential impacts of floodplain on human lives or property can influence individuals’ ability to mitigate properly [26]. We add this variable to our framework to measure residents’ overall knowledge about the different aspects of a flooding event (i.e., likelihood, potential damage, and preventive actions). Survey questions that measured flood knowledge in our study were the following: “the chances of being impacted by a flood”, “the types of damage your home may sustain from a flood”, “what you may do to prevent the flooding damage”, “how often do you think about flood events?”, and “how often do you think about flood safety issues?”.

Flood Preparedness. This variable refers to efforts (e.g., stockpiling food and water and creating an emergency plan) to prepare for a potential flood [25]. Flood preparedness, which, in part, can be influenced by previous flood experiences [32], offers insight into the extent to which individuals are willing to adopt preparedness or mitigation measures. In our study, this variable is measured through preparedness activities (e.g., sandbagging and storing sufficient food and water) undertaken by individuals to prepare for a flooding event [33].

Flood Mitigation. Flood mitigation refers to strategies that can reduce the potential risk of future flooding events. Flood mitigation can be categorized into the following two general categories: structural flood mitigation and non-structural flood mitigation [30]. Non-structural mitigation refers to strategies for reducing the flood risk without changes to the physical environment. These measures can include “policies and laws, raising public awareness, flood forecasting and warnings, evacuation, training and education, land use adjustment, regulations and insurance, funding and subsidies, spatial and flood management plans” [34] (p. 4). Structural mitigation is a type of flood hazard adjustment adoption that aims to create a more resilient home structure [25]. In this study, we looked at both structural and non-structural mitigation strategies. The information about structural mitigation was collected by asking individuals about structural upgrades to their residences (e.g., elevated floors) to prevent major flood damage. Non-structural mitigation data were collected by asking participants about their flood insurance purchases (the following variable). Mitigation was measured through the following survey questions: “do you have flood insurance?”, “is the place where you live elevated?”, “is the place where you live dry flood-proofed?”, and “is the place where you live wet-proofed?”.

Flood Insurance. Flood insurance is a popular non-structural mitigation [25] and disaster risk strategy [35] to reduce the potential financial liability in the aftermath of a flooding event. The Federal Emergency Management Agency’s (FEMA) National Flood Insurance Program (NFIP) is one of the successful federal-level insurance policies that can help property owners and businesses recover from the flood damage [36]. A study by Landry and Turner suggests that factors such as being in a flood zone or how individuals perceive damage can determine residents’ decision to purchase flood insurance [37]. It is noteworthy that this study views “flood insurance” as a variable distinguishable from flood mitigation. This decision was made to better understand how individuals’ economic decisions and behaviors can contribute to their willingness to pay for the NASA product. It can be argued that residents that purchase flood insurance are likely to pay for or support another flood risk reduction product that can reduce their level of financial loss. We measure this variable by directly asking respondents if their homes are protected by flood insurance (i.e., “do you have flood insurance”?), which was one of the four survey questions that were used to operationalize “flood mitigation’.

We also included socioeconomic characteristics, such as gender, age, educational attainment, income, and city of residence, as control variables, as previous literature research reported significant effects of these correlates in the likelihood of engaging in risk reduction strategies [38,39].

1.3. Empirical Models

The empirical development of Model 1 is depicted in Figure 2. Model 1 is composed of a set of four equations, estimated separately. Specifically, the four regression equations, where $X_i$ represents the predictor (independent) variables, $PA{V}_i$ accounts for the mediating variable, ${\epsilon }_i$ represents the residual, and $PE{V}_i$ is the dependent variable, are shown below:

$$\mathrm{Path} \mathrm{C} PE{V}_i=f(X_i, {\epsilon }_i)$$

(1)

$$\mathrm{Path} \mathrm{A} PA{V}_i=f(X_i, {\epsilon }_i)$$

(2)

$$\mathrm{Path} \mathrm{B} PE{V}_i=f({PAV}_i, {\epsilon }_i)$$

(3)

$$\mathrm{Path} \mathrm{C}\prime PE{V}_i=f(X_i,{PAV}_i, {\epsilon }_i)$$

(4)

The first equation looks at the direct relationship between the predictor variables (risk-related variables and socioeconomic variables) and the PEV of the NASA product. The second looks at the relationship between the predictor variables and the mediating variable (PAV of the NASA product). The third equation relates the mediator (PAV) and the outcome variable (PEV). Finally, the fourth presents the relationship between the predictors and the outcome variable (PEV), controlling for the effect of the mediating variable. The mediation model implies the following specific hypotheses [37]:

H1.1. 

There are statistically significant relationships between the predictors and the PEV of the NASA product (Equation (1), Path C in Figure 2).

H1.2. 

There are statistically significant relationships between the predictors and the mediating variable (PAV) (Equation (2), Path A in Figure 2).

H1.3. 

There is a statistically significant relationship between the mediating variable (PAV) and the outcome variable (PEV) (Equation (3), Path B in Figure 2).

H1.4. 

There is a statistically significant relationship between the mediating variable (PAV) and the outcome (PEV) and not a statistically significant relationship between the predictors and the outcome (PEV) (Equation (4), Path C′ in Figure 2).

If all of these hypotheses are held, the PAV may be said to completely mediate the relationship between the predictors and willingness to pay for the NASA product (i.e., PEV). However, if the last of these hypotheses is not met, but the magnitude of relationships between the predictors and PEV in Equation (4) (i.e., when the mediator is included) is smaller than the relationships observed in Equation (1), partial mediation is indicated.

Model 2 follows a classic regression equation (Equation (5)), where $Si$ is the dependent variable (support of the NASA product), $X_i$ represents the independent variables (PAV and PEV), and ${\epsilon }_i$ represents the residual. The equation is shown below:

$$S_i=f\left(X_i,{\epsilon }_i\right).$$

(5)

2. Methodology

Using the variables in Model 1 and Model 2, a mail survey was designed and conducted in the two most populous metropolitan areas within the RRB, including West Fargo–Fargo–Moorhead and Grand Forks–East Grand Forks. As noted previously, “PAV” was conceptualized as benefits, and “PEV” was viewed as costs. Following the standard mail survey procedure [40], a self-administered mail survey was distributed by Survey Systems in four waves to 1500 randomly selected households from September 2018 to October 2018. The survey was sent out with a cover letter explaining the project goals. A total of 211 completed surveys were received, representing a response rate of 14%.

2.1. Dependent Variables

The dependent variable in Model 1 is the PEV of the NASA product. With the final objective of estimating the economic value of the new forecast product, the respondents were provided with a comparison of the new and old flood forecasts and presented with a series of price categories that they would be willing to pay in the hypothetical scenario that this product was marketed. These categories were “Not willing to pay”, “$1 to $2 per year”, “$3 to $5 per year”, “$6 to $10 per year”, “$11 to $15 per year”, “$16 to $20 per year”, “$21 to $50 per year”, and “$51 or more per year”. Due to the low number of respondents who chose the two latter categories, we combined them into a “$21 or more per year” category.

The dependent variable of Model 2 is the support of the use of the new NASA product. To collect this information, the survey presented 5 categories (ranging from strongly disagree to strongly agree) to measure the extent of the agreement of the respondent with the following statement: “I would support the use of the new NASA product by the National Weather Service for the Red River Spring flood forecast for my city/county”.

2.2. Independent Variables

The PAV of the NASA product acts as both the mediating variable of Model 1 and an independent variable in Model 2. In order to estimate the perceived usefulness of the product, the survey included questions to determine to what extent the respondent perceives that he/she would receive the benefit of the NASA product in several aspects of their lives. Thus, the survey asks the respondent to rate from “Not at all” to “Very great extent” (5 categories) the potential benefit of the product to (a) reduce the likelihood of flood damage to his/her home, (b) reduce the likelihood of injury or death of a member of the household, (c) reduce the likelihood of disruption on his/her job that prevents him/her from working, and (d) reduce the likelihood of disruption of electrical, telephone, or other basic services. In addition, the survey also invites the respondent to rate in these same 5 categories the extent of the improvement in the local flooding warning system by adopting the improved flood forecast product. The PAV variable is produced by averaging the responses to these five questions.

3. Analyses and Results

3.1. Model 1: Mediation Model Between Risk-Related Variables and PEV

First, correlations were run to determine if there was any evidence of collinearity between the independent variables (see

Table 1. Correlation matrix of the study variables.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)	(17)	(18)
(1) Support	1
(2) PEV	0.31	1
(3) PAV	0.34	0.34	1
(4) Flood Risk	0.01	0.05	0.16	1
(5) Flood Knowledge	0.08	−0.01	−0.05	−0.11	1
(6) Flood Experience	0.02	−0.01	0.09	0.06	0.30	1
(7) Flood Risk Perception	0.18	0.18	0.42	0.19	0.06	0.11	1
(8) Preparedness	0.04	−0.09	0.02	0.06	0.26	0.18	0.14	1
(9) Mitigation	−0.04	−0.08	−0.05	0.01	0.10	0.14	−0.04	0.20	1
(10) Insurance	0.00	0.15	0.15	0.48	0.01	0.03	0.18	0.07	−0.05	1
(11) Age	−0.14	−0.07	−0.11	0.07	−0.02	0.12	−0.08	0.06	0.08	−0.00	1
(12) Gender	0.01	0.08	0.06	−0.07	−0.03	0.01	0.04	−0.01	0.00	−0.01	−0.07	1
(13) Education	0.11	0.19	0.04	−0.01	0.11	−0.09	0.17	−0.00	−0.12	0.01	−0.14	−0.04	1
(14) Income	0.06	0.15	0.02	−0.03	0.10	−0.00	0.05	0.02	−0.04	0.05	−0.20	−0.15	0.30	1
(15) West Fargo	−0.09	−0.08	−0.09	−0.09	−0.01	−0.21	−0.10	−0.05	−0.17	−0.07	−0.07	0.02	0.02	0.09	1
(16) Moorhead	0.14	0.06	−0.04	−0.12	0.03	−0.02	−0.03	0.03	0.04	−0.07	0.00	−0.03	−0.09	0.01	0.16	1
(17) Grand Forks	−0.00	−0.13	0.10	0.01	0.10	0.39	0.04	0.08	0.06	−0.00	0.14	0.02	−0.09	−0.11	−0.16	−0.21	1
(18) East Grand Forks	−0.03	−0.14	0.03	−0.02	−0.03	0.04	0.00	−0.17	−0.00	0.01	−0.08	−0.08	−0.10	−0.05	−0.05	−0.07	−0.07	1

for the correlations). Then, we run the set of regressions on the total and mediation models (Equations (1)–(4)). It is evident from both the correlations and the regression analyses that collinearity is not an issue. According to Greene [41], the correlations presented in Table 1 are considered to be low to moderate.

Table 2. PEV as a function of risk-related and socioeconomic variables (Path C) (N = 211).

	Variables	Ordered Logistic Regression		Interval Regression
		Coef.	p > z	Coef.	p > z
Risk-Related Variables
	Flood Risk	−0.1804854	0.586	−0.1399995	0.894
	Flood Knowledge	−0.0630362	0.678	0.0231417	0.962
	Flood Experience	0.4648496	0.407	1.101521	0.530
	Flood Risk Perception	0.7786564	0.008 ***	1.361854	0.148
	Preparedness	−1.303235	0.035 **	−3.865661	0.043 **
	Mitigation	−0.2091166	0.666	−0.816733	0.602
	Insurance	0.8020452	0.040 **	1.963369	0.114
Socioeconomic Variables
	Age	−0.078388	0.656	0.0688766	0.901
	Gender	0.4700356	0.130	1.06356	0.285
	Education	0.1933892	0.220	0.8409217	0.091 *
	Income	0.1379197	0.192	0.4861878	0.137
	West Fargo	−0.6475869	0.140	−2.010479	0.138
	Moorhead	0.157751	0.656	0.6928277	0.539
	Grand Forks	−0.6699317	0.071 *	−2.423807	0.043 **
	East Grand Forks	−2.806116	0.016 **	−5.586335	0.041 **
	Cons			0.1388899	0.894
	Ordered Logistic Regression			Interval Regression
	${LR \chi }^2$: 41.42 *** Akaike’s information criterion (AIC): 725.671			${LR \chi }^2$: 30.67 ***
	Schwarz’s Bayesian information criterion (BIC): 796.060			Mean predicted PEV: $5.029

Coefficients are from an ordered logistic regression and interval regression. Significance: *** p < 0.01, ** p < 0.05, * p < 0.1.

,

Table 3. PAV as a function of risk-related and socioeconomic variables (Path A) (N = 211).

Variables		OLS Regression
		Coef.	Std. Err.	p > t
Risk-Related Variables
	Flood Risk	0.1246805	0.1329033	0.349
	Flood Knowledge	−0.0736689	0.0610064	0.229
	Flood Experience	0.1381029	0.2221467	0.535
	Flood Risk Perception	0.6767283	0.1194154	0.000 ***
	Preparedness	−0.076446	0.2424903	0.753
	Mitigation	−0.0862088	0.1988598	0.665
	Insurance	0.0926968	0.1572196	0.556
Socioeconomic Variables
	Age	−0.1063598	0.0703202	0.132
	Gender	0.0799444	0.1261931	0.132
	Education	−0.0143187	0.0630566	0.821
	Income	0.0090785	0.0414207	0.827
	West Fargo	−0.104762	0.1722813	0.544
	Moorhead	−0.0057688	0.142905	0.968
	Grand Forks	0.1768769	0.1522515	0.247
	East Grand Forks	0.1081676	0.3491642	0.757
	Cons	1.248779	0.6267125	0.048 **
R2	0.2175
Adjusted R2	0.1573

Coefficients are from multiple regression. Significance: *** p < 0.01, ** p < 0.05, * p < 0.1.

,

Table 4. PEV as a function of PAV (Path B) (N = 211).

Variables	Ordered Logistic Regression		Interval Regression
	Coef.	p > z	Coef.	p > z
PAV	0.8926713	0.000 ***	2.171059	0.000 ***
Cons			−0.0007	1.000
Ordered Logistic Regression ${LR \chi }^2$: 31.87 *** Akaike’s information criterion (AIC): 707.213 Schwarz’s Bayesian information criterion (BIC): 730.676			Interval Regression ${LR \chi }^2$: 17.02 *** Mean predicted PEV: $5.031

Coefficients are from an ordered logistic regression and interval regression. Significance: *** p < 0.01, ** p < 0.05, * p < 0.1.

and

Table 5. PEV as a function of PAV, risk-related, and socioeconomic variables (Path C′) (N = 211).

Variables		Ordered Logistic Regression		Interval Regression
		Coef.	p > z	Coef.	p > z
PAV		0.907561	0.000 ***	2.169246	0.000 ***
Risk-Related Variables
	Flood Risk	−0.4038577	0.234	−0.41162	0.684
	Flood Knowledge	−0.0128678	0.934	0.1806315	0.697
	Flood Experience	0.4072686	0.479	0.8129701	0.630
	Flood Risk Perception	0.2290778	0.469	−0.1098682	0.910
	Preparedness	−1.292235	0.038 **	−3.704483	0.044 **
	Mitigation	−0.1284437	0.793	−0.6310746	0.675
	Insurance	0.7974848	0.044 **	1.759469	0.142
Socioeconomic Variables
	Age	−0.0025225	0.989	0.3031428	0.572
	Gender	0.3923441	0.206	0.8906143	0.353
	Education	0.2263433	0.156	0.8716287	0.069 *
	Income	0.1379159	0.194	0.4634811	0.141
	West Fargo	−0.6515055	0.148	−1.778784	0.173
	Moorhead	0.0917734	0.799	0.6969605	0.521
	Grand Forks	−0.9222587	0.016 **	−2.809063	0.015 **
	East Grand Forks	−3.223654	0.008 ***	−5.826651	0.027 **
	Cons			−2.553762	0.595
	Ordered Logistic Regression ${LR \chi }^2$= 66.39 *** Akaike’s information criterion (AIC): 702.692 Schwarz’s Bayesian information criterion (BIC): 776.433			Interval Regression ${LR \chi }^2$: 46.03 *** Mean predicted PEV: $5.029

Coefficients are from an ordered logistic regression and interval regression. Significance: *** p < 0.01, ** p < 0.05, * p < 0.1.

present regression results. It is noteworthy that the survey response options for the variable “PEV of the product” were not only ordered but also measured as a range (e.g., $1 to $2 per year, $3 to $5 per year). While the use of ordered logistic regression is well justified in most subfields of the social science discipline, interval regression has also been applied in the economic literature for variables with a range of response options (see [4,42]). Given the different results obtained from ordered logistic regression and interval regression models, we discuss both results for paths with PEV as the dependent variable (i.e., Paths C, B, and C′).

We find evidence supporting the first of the hypothesis of the mediation model (H1.1), as follows: several risk-related variables are related to PEV (Table 2). The likelihood ratio (LR) ${\chi }^2$ test statistics indicate the overall model fit at the 1% significance level for both the ordered logistic and interval regression models, suggesting robust regression results. Table 2 represents the results for the ordered logistic regression and interval regression models for PEV when risk-related and socioeconomic variables are included in the model (Path C). In the ordered logistic regression results, variables of flood risk perception and insurance are positively correlated with PEV, indicating that residents who perceive being at the risk of flood and residents with insurance are willing to pay (or pay more) for the NASA product. The results from both ordered logistic regression and interval regression models indicate a negative significant relationship between the independent variable of preparedness and the PEV of the NASA product, suggesting that residents who are more prepared for floods are less likely to pay for the NASA product. Such a finding, to some extent, may seem counterintuitive but may be explained by the perceived marginal benefit of the NASA product, i.e., the public forecast information, relative to the perceived marginal benefit of a personal action. Homeowners who are proactive in flood preparedness may view the public good as less essential and, therefore, have a lower PEV and a lower demand for it. In other words, the homeowners substitute private actions (i.e., preparedness) for the public good. The remaining risk-related variables in the model (flood risk, flood experience, flood knowledge, and flood mitigation) were not significant, and, therefore, they do not explain PEV. In the same model, socioeconomic variables were also non-significant, except for the variable education, in the interval regression output. Additionally, both the ordered logistic regression and interval regression results show that the cities of Grand Forks and East Grand Forks were negatively correlated with PEV. This finding would indicate that residents from these two cities are significantly less likely to be willing to pay for NASA products or would choose to pay less than residents from West Fargo or Moorhead.

The second hypothesis (H1.2) of the mediation model suggested that there must be a significant relationship between the risk-related variables and the mediating variable (PAV). Focusing only on those variables that were significant in relation to PEV, only flood risk perception was significantly associated with PEV (Table 3). This relationship is positive, similar to the relationship between this variable and PEV in the ordered logistic and interval regression outputs in Table 2, meaning that those who believe themselves to be more at risk of flooding perceive the NASA product to be more valuable from an operative application standpoint (usefulness). Flood preparedness, flood insurance, education, and the Grand Forks and East Grand Forks variables were not statistically significant, which discards the possible mediation effect of PAV in their relationship with PEV.

Table 4 shows the results of Equation (3), which tests the relationship between the mediator (PAV) and the outcome variable (PEV) (H1.3). There is a strong positive relationship between PAV and PEV, indicating the expected assumption that those who believe the NASA product has valuable applications and that it will somehow reduce their flooding risk are those who are more willing to pay and, thus, confer a higher economic value to the product.

Finally, Table 5 shows the results of Path C′ (Equation (4)). In this equation we include in the model all the independent variables and the mediating variable, expecting that the association of those risk-related variables that both showed a significant association with the mediator and the outcome decrease (partial mediation) or disappear (complete mediation). In our case, only the flood risk perception variable passed the previous filters. As we can see in Table 5, flood risk perception is no longer significant, confirming a complete mediation by PAV in its effect on PEV. The rest of the variables who presented significant associations in Path C (flood preparedness, flood insurance, Education, Grand Forks, and East Grand Forks) remain unaffected.

With these results, we can, therefore, confirm that our H1 hypothesis, expecting a mediation effect of PAV in all risk-related variables, is only partially met (Figure 3). Model 1 validated that four of the risk-related variables included do not relate at all with PEV while three had explanatory capacity. Among these three, the model rejects the mediation effect of PAV in the relation of flood preparedness and flood insurance with PEV, therefore being a direct relationship. However, the model confirms the complete mediation of flood risk perception. The regression equations in Table 2, Table 4, and Table 5 estimates the mean PEV per household to be about USD 5.03 per year. This amount translates into an estimated total PEV of approximately USD 570,367 for the households residing in the five cities (according to the 5-year America Community Survey, there was an estimate of 113,393 households residing in the cities of Fargo, Moorhead, West Fargo, Grand Forks and East Grand Forks in 2018.). Finally, out of the five socioeconomic variables included, only education and the city of residence resulted significant, in particular the cities of Grand Forks and East Grand Forks.

3.2. Model 2: Support of NASA Product

Our second hypothesis (H2) stated that the support of the new NASA’s Satellite Enhanced Snowmelt Flood Predictions in the Red River is determined by the subjective benefit–cost analysis made by the surveyed residents. Therefore, the PAV and perceived economic value of the product will condition the overall acceptance of this mitigation strategy among the population.

Table 6. Support as a function of PEV and PAV (N = 211).

Variables	Coef.	Std. Err.	p > z
PEV	0.2842218	0.0882719	0.001 ***
PAV	0.792018	0.1894495	0.000 ***
Pseudo R2: 0.0791 Akaike’s information criterion (AIC): 515.802 Schwarz’s Bayesian information criterion (BIC): 535.914

Coefficients are from an ordered logistic regression. Significance: *** p < 0.01, ** p < 0.05, * p < 0.1.

confirms this hypothesis, as we can see how PAV and PEV have a strong positive significant association with the dependent variable (Support). Thus, we can confirm that those who conferred a higher degree of applicability to the product and those who gave it a higher economic value are more likely to support the use of this mitigation strategy put forward by NASA.

4. Discussion

This study focused on the application of NASA’s Satellite Enhanced Snowmelt Flood Predictions product in the Red River of the North Basin. We examined a range of factors related to flood risk reduction behaviors to assess if and to what extent residents were willing to pay for and support the NASA product.

A notable contribution of this paper is related to the methodology used for quantifying and measuring the benefits of a risk reduction product like NASA’s Satellite Enhanced Snowmelt Flood Predictions product. The cost-benefit analysis method is a widely studied method in the context of flood risk (see [43,44,45]). To evaluate the costs and benefits of the NASA product, our paper utilized the CV method to assess residents’ willingness to pay. The study included several risk-related variables such as knowledge, mitigation, and insurance to better understand residents’ awareness of and approach to flood preparedness. Our initial assumption was that these risk-related variables could influence residents’ perceived applicability or usefulness of the NASA product (i.e., PAV variable). Inspired by the PADM model examining mediation effects from psychological and experiential factors, we further hypothesized that PAV functions as a mediating variable between risk-related variables and perceived economic value (i.e., PEV variable or willingness to pay). Variable PAV (conceptualized as benefits) was measured by our survey using five items asking about residents’ perception of how the NASA product would impact “damage to a home, risk of injury or death, disruption of job, lifelines, and warning systems”. While only one risk-related variable was significant in its relationship to PAV (partial mediation), the measurement of the PAV construct in our study can be used by future studies to measure the potential benefits of flood prediction products or other “non-marketed” disaster products. Such an approach to the operationalization of the benefits of a disaster product can be helpful for scholars or practitioners who aim to conduct a preliminary assessment of the potential benefits associated with disaster-related products that lack an established measurement tool. From a theoretical contribution standpoint, the study has applied a modified PADM to an interdisciplinary design that incorporates a benefit–cost analysis in a disaster risk reduction study. Future disaster studies with a focus on economic analysis can benefit from replicating the research design of this empirical research.

Another implication of this study stems from the finding that risk-related variables included in this study could not all collectively predict changes in the benefits (PAV) or cost (PEV) of the NASA product (see Table 2, Path C; Table 3, Path A). Future studies should include other risk-related variables (e.g., the duration of living in a flood zone, the year the home was built) as potential alternative variables that can affect residents’ perception of supporting and paying for a flood prediction model. Regardless of these shortcomings, one of the critical findings of this study can be attributed to one of the risk-related variables, Flood Insurance, and its relationship to PEV. We previously noted that Flood Insurance could be considered a non-structural mitigation strategy [25] and that we treated it as a variable separate from Mitigation to gain a better understanding of how residents’ financial investments in a mitigation strategy shape their willingness to pay for the NASA product. The findings indicated that Flood Insurance, measured separately from Mitigation, was significant in Path C and Patch C′ using ordered logistic regression outputs. Mitigation, as a general category, on the other hand, showed no significant relationship with any of the statistical models we tested. This finding highlights the importance of insurance programs, such as the National Flood Insurance Program (NFIP), in flood-prone areas and suggests that residents who are already paying for a non-structural mitigation strategy (i.e., flood insurance) are willing to pay (or pay more) for yet another non-structural mitigation strategy such as the NASA product. While such a result speaks to the power of paid non-structural mitigation strategies, we encourage future research studies to test this finding in other settings and for other disasters to cross-validate our results.

Lastly, an important methodological contribution of this study is the collective use of ordered logistic regression and interval regression for examining ranges in our dependent variable (i.e., PEV) that is measured using a Likert scale. The application of interval regression as a complementary method strengthened our understanding of the survey results. Remarkably, flood risk perception, as the only variable whose relationship was mediated by PAV, had a positive relationship with residents’ level of support for the NASA product. This indicates the importance of individual perception of PAV, or benefits, of the NASA product and the PEV that ensued. This is a critical finding of this paper, as it suggests that one’s perception of flood risk can shape the understanding of the usefulness of the NASA product and, in turn, their willingness to pay for the product. Given that the study participants were randomly selected, such findings could imply that RRB residents’ flood perception determines PAV and, ultimately, PEV for the NASA product. While understanding how individual perceptions are formed and influenced is beyond the scope of the study, we suggest future studies investigate how local news outlets and social media platforms (such as Nextdoor) and community factors (e.g., apartments communities and close-knit community ties) can impact residents’ perception of a flood risk reduction tool. Notably, our findings should be studied with a critical assumption in mind, as follows: risk-related variables (e.g., flood risk perception) deal with complex interpersonal factors that are not easily measured in disaster studies. Although previous research has discussed cognitive characteristics in disaster contexts [46], many existing studies mainly view residents’ disaster-related behaviors in the context of how risk information is perceived by them. This paper, however, incorporated the effects of personal characteristics and past experience to better understand how residents value the NASA product in terms of costs and benefits. Such an approach contributes to existing emergency management research by highlighting the interplay between individual characteristics and support for non-marketed risk reduction goods and services. We suggest future studies use and build on our model to further explore the relationship among cognitive, personal, and risk-related factors with mitigation and preparedness strategies.

It is, however, noteworthy that the survey had a relatively low response rate (14%) and was only distributed to residents in a particular region (RRB). While the study contributes to the existing research on flood prediction products and residents’ behavior, the findings and interpretations of this study are limited to the region (RRB) and hazard type (flood), making it challenging to generalize the findings of this study to the entire RRB area, other regions, or hazards. Additionally, our study relied on the CV method to assess residents’ level of support for the NASA product. However, since the NASA product is not a market good, the survey presented a hypothetical situation in which residents could pay for the product, making the responses subject to biases and not reflecting the actual level of support for the product.