Valuing the Economic Impact of River Floods and Early Flood Warning for Households in Bangladesh

Flood early warning systems have the potential to mitigate damages for vulnerable populations that experience river flooding in Bangladesh. We conducted a 2,247 household survey and series of focus groups to estimate the economic damages from 2016 river floods and the hypothetical savings of a 3-and 8-day warning for households living in the Jamuna River floodplain. Households were identified using geo-sampling, a novel geographic information system (GIS)–based sampling methodology that facilitates probability-based sampling where data are insufficient. Total damages for the entire flood plain in 2016 totaled to $1.3 billion, or 25% of household income and assets. Respondents estimated avoided damages from a hypothetical 3-and 8-day warning to be $73m and $85m, respectively, reflecting diminishing returns to additional days of early warning. With the hypothetical early warning, respondents derived the greatest savings from protecting their land, household/dwelling, and livestock. The greatest savings to households receiving a hypothetical additional 5 days of warning (from a 3-to an 8-day warning) would be realized in protecting agricultural production. Selling assets/livestock and employing protective sandbags were the preventative actions with the highest benefit–cost ratios that households said they would undertake. Importantly, only 11% of households received any early warning at all during the 2016 flood season, suggesting that the greatest benefits moving forward would be achieved by communicating existing or improved warnings more effectively to households in the floodplain.


Introduction
Bangladesh is one of the most vulnerable countries in the world to floods.The country experienced five large floods from 1987-2007 (Mirza 2011), and Bangladesh's Ministry of Disaster Management and Relief reported that monsoon-induced flooding of 2016 affected 3.7 million people.(ReliefWeb 2016).According to the international disasters database, there was an average of two floods, 236 deaths, and nearly 500,000 people affected per year from river floods in Bangladesh from 1990 through 2016 (EM-DAT: The Emergency Events Database-Université catholique de Louvain [UCL] -CRED, 2019).Furthermore, flooding patterns point to an increase in the frequency of flooding in Bangladesh (Ghatak et al 2012;Mirza 2011;Parvin et al. 2016;Dasgupta et al. 2011), suggesting that the impacts of flood events may worsen over time.In 2017, monsoon-induced river floods in Bangladesh were the worst in four decades (ReliefWeb (2017)).
The frequency and severity of floods in Bangladesh negatively affect the livelihoods of populations living in or near affected areas, which translates into significant impacts on human health and economic damages to households (Mirza 2011).Although the extent of the damage varies by year and region, floods affect households across a range of material assets and income-generating activities (Sultana and Rayhan 2012;Parvin et al. 2016).Floods disrupt the daily lives of populations living in or near affected areas, affecting their social and economic activities; damaging property, crops, agricultural lands, and livestock; and endangering people's lives.
Early flood warnings, which aim to give households time to prepare for oncoming floods, hold the promise of deterring significant economic losses from flooding in Bangladesh.These messages typically communicate information about flood water height at a specific location and are shared across multiple formal and informal mediums, including by government-issued short message service (SMS), newspaper, and word of mouth (Shah et al. 2012;Flood Forecasting and Warning Center 2016).In Bangladesh, the Food Forecasting and Warning Centre within the Bangladesh Water Development Board formally issues warnings when statistical flood forecast models indicate a reasonable certainty of flood risk.
The FFWC issues warnings as early as 3 to 5 days before upstream river floods (Flood Forecasting and Warning Center 2016).The FFWC uses flood forecasting models that draw on multiple data sources as inputs, including real-time river height and rainfall data measured at different points along the river.More recently, FFWC has used data from NASA's JASON-2 Earth satellite as an experimental technology to remotely monitor water height at "virtual stations" along upstream rivers, providing them with additional days of early warning time.As satellite technologies continue to improve, they have the potential to increase the amount of warning time provided to households for riverine floods in Bangladesh.
Flood warning messages are issued to motivate households at risk of flooding to prepare for a flood event by undertaking activities to mitigate damage.These activities may include moving property to higher land; protecting property and infrastructure with sandbags or other nonporous materials; stocking water, medicine, and food; and leaving the area.Time and cost are critical to completing these damage-mitigating activities and flood warnings are issued as early as possible.
Generally, an increase in flood early warning time is assumed to motivate mitigation activities, thereby decreasing net economic losses experienced by households.Hallegatte et al. (2012) estimated that the potential benefits from upgrading hydrometeorological and warning services to increase preparedness time in developing countries are significant: between 300 million and 2 billion USD per year in savings or avoided losses because of natural disasters, an average of 23,000 saved lives per year, and between 3 and 30 billion USD per year in additional economic benefits.
This study quantifies the economic value of increased flood warning time to households in the Jamuna River floodplain of Bangladesh.To do so, we estimated three distinct values related to households living in the river flood plains: the economic impact of 2016 flooding, the economic impact incurred by flood preparedness activities that were undertaken in 2016, and the economic impact of increased flood warning time to conduct preparedness activities in two hypothetical scenarios in which households received 3-and 8-day warnings.We surveyed 2,247 households in the Jamuna River floodplain in Bangladesh in 2017 and analyzed results to find the extent of damage of the 2016 river floods on households and potential savings from a 3-day and an 8-day warning.Our findings contribute to the evidence base on the economic severity of flood damages in Bangladesh and the potential benefits of improved early warning.

Literature Review
Although several studies quantify flood damages in Bangladesh, flood damage assessment is understudied despite its importance for assessing vulnerability and determining appropriate risk reduction and recovery measures (Merz et al. 2010).Few studies focused on Bangladesh characterize household-level flood damages, and even fewer focus on upstream river flooding.We found no specific estimates of damages caused by river floods in Bangladesh disaggregated by household wealth categories, which is an important omission considering the contributions this information makes to the flood response community to better understand and prioritize damage mitigation activities.
Studies that have quantified flood impacts in Bangladesh have estimated damages in the hundreds of millions to several billion dollars.Mirza (2011) found that the economic loss caused by all floods in Bangladesh in a typical flood year is about US$175 million, and in extreme cases, the damage can be over 2 billion dollars.A study conducted by the government of Bangladesh estimated that the 2007 floods resulted in US$1.1 billion in losses (Subbiah et al. 2008).Haque and Jahan (2015) modeled the impacts of the 2004 and 2007 floods on output, income, and employment in six regions of Bangladesh and attributed 249,611 million Bangladeshi Taka (BDT) (US$3.5 billion) and 148,408 million BDT (US$2.1 billion) in direct damages, respectively.Parvin et al. (2016) studied the impact floods have on the livelihoods of the rural poor in Bangladesh, finding that almost all households in their sample experienced a reduction in their income because of floods yet the authors do not detail the income sources or any loss of assets.
Other authors have characterized household flood damages outside of Bangladesh, providing context for the scope of damages incurred in Bangladesh.Navrud et al. (2012) found that households in Quang Nam province in Central Vietnam incurred an average of 3,816,105 Vietnamese Dong (US$167) during the 2007 floods.Wijayanti et al. (2017) valued total economic damages per household in Jakarta, Indonesia, at $308 for the 2013 flood event.
Early flood warning systems present information on which households and governments can act to reduce flood damages.Effective flood forecasts allow governments and civil society to better support flood-prone communities (Rogers and Tsirkunov 2011) and can guide households to take preventative actions.Where implemented, advance warnings can substantially decrease deaths and other impacts of severe weather events (Pappenberger et al. 2015;Hallegatte et al. 2012;Habib et al. 2012), translating into significant economic savings.
Although several studies detail the actual or potential savings to affected populations as a result of early flood warning systems in high-income countries (Pappenberger et al. 2015;Carsell et al. 2004), there is scant evidence on the extent of these savings in lowand middle-income countries in general and Bangladesh in particular.Subbiah et al. (2008) estimated savings from probabilistic forecasting for river flooding in Bangladesh of US$166.3 million and categorized these savings by sector (i.e., livestock, agriculture, fisheries) at a country level.Cumiskey et al. (2015) found that average savings from a 5-day warning among households affected by river flood damages in Bangladesh are US$472 per household.
Household damage mitigation behaviors, or flood coping strategies, are discussed widely in the context of Bangladesh (Brouwer et al. 2007;Chanda Shimi et al. 2010;Shah et al. 2012) and beyond (Shah et al. 2017;Poussin et al. 2014); however, few correlate these behaviors with economic savings in relation to flood warning time.This may be because, as we found, flood warnings do not reach most households living in the river flood plain of Bangladesh.Therefore, it is difficult to retrospectively value the impact of this information and the activities it motivates.Nevertheless, these data are important because the dearth of quantitative estimates of the costs and benefits of warnings and subsequent responses results in difficulty convincing governments of the economic and social value of early warning systems for disaster damage reduction (Rogers and Tsirkunov 2011) and does not promote optimal and cost-efficient decision-making for risk management (Merz et al. 2010).
This study adds to the existing evidence base on impacts of river flooding in Bangladesh flood early warning by quantifying the direct household-level economic impacts of the 2016 floods in the Jamuna River flood plain and estimating hypothetical savings of a 5-day increase in early flood warning time (Fig. 1).Furthermore, we characterized these damages and savings by household wealth category to demonstrate how households' assets and incomes are being affected.We additionally explored the time and value of mitigation actions available to and taken by flood-prone households to determine the cost efficiency of these actions.

Data
We used data from a survey of 2,247 flood-prone households in the Jamuna River flood plain and conducted supplemental qualitative fieldwork that informed the design of the survey instrument.Respondents were male and female inhabitants of households living in flooded regions of the Jamuna River basin across six districts (Bogra, Gaibandha, Jamalpur, Kurigram, Sirajganj, and Tangail).The survey collected information on perceptions of floods, past behaviors during floods, a detailed economic accounting of a household's wealth (assets and income), and damages from flooding.We also collected additional household information, including demographics, socioeconomic information, attitudes, and experiences toward flood warnings, and data on the precise height of water at a household's dwelling.
The survey was designed to account for real and hypothetical changes to a household's wealth as a result of flood events across a range of household income and asset categories in the year following the flood events.To track the components that make up a household's wealth systematically, the survey used a detailed economic inventory over three key time periods.First, it captured a household's "baseline" wealth before the 2016 floods began.Second, the survey used the same inventory to collect economic losses accrued in 1 year from flooding in 2016 according to the wealth sources identified in the baseline.Third, the survey used a hypothetical scenario to capture savings from actions a household would have taken if they received a 3-and 8-day flood warning in advance of the 2016 floods.
We conducted a series of focus group discussions in representative communities during the instrument's design to inform the survey development, ensuring construct validity and the salience of context-specific details, including development of a hypothetical scenario for valuing flood information.Additional groups further provided important contextual detail to help characterize actions that could realistically be undertaken by households and the size and scope of damages that households had experienced previously.The survey instrument was designed by RTI in collaboration with a Bangladeshi partner, the South Asian Network on Economic Modeling (SANEM).SANEM also facilitated focus group discussions, translated the survey, programmed it in CSPRO, pretested the survey, and implemented the survey in the field.

Household Identification and Selection
We employed a geographic information system (GIS)-based sampling methodology called geo-sampling to draw the sampling frame from which households were randomly selected.RTI International developed geo-sampling for use in situations where a multistage probability-based sample is required but there are insufficient census data available to create a frame of households and an individual listing of households is too costly and time-consuming to create (Cajka et al. 2018).The basis of geo-sampling is the availability of gridded population estimates from data suppliers such as LandScan (from Oak Ridge National Laboratory) and WorldPop (a project of the GeoData Institute and the University of Southampton).One-km 2 grid cells and their estimated population sizes were used as primary grid cells (PGCs) at the first stage of selection.We limited the survey to areas that were inundated by water during the peak of the 2016 flood, during August 3 and 4, 2016, within six districts (Bogra, Gaibandha, Jamalpur, Kurigram, Sirajganj, and Tangail).The peak flood data were obtained from NASA's MODIS Near Real-Time Global Flood Mapping Project.We limited the flood zone to contiguous areas of flooding.We also limited the PGCs eligible for selection to those where at least 75% of the PGC was within the flooded area and the estimated population of the PGC was over 250 residents, eliminating areas that were protected by flood control embankments or had very small populations.
Once PGCs were selected, they were divided using GIS software into a grid of secondary grid cells (SGCs).These SGCs can be set to an appropriate size based on the density of the population in the PGC and the desired sample size for the survey.Once a PGC is selected by a probability proportional to its population size, the SGCs are selected with equal probability of selection.In this application of geo-sampling, up to four primary SGCs were selected from a 6 × 6 tessellation of the PGC.After PGCs and SGCs were selected, digital maps were created, allowing interviewers to navigate to their assigned PGCs and SGCs using the global positioning system capability of the tablet or smartphone used to collect data (Fig. 2 shows the PGCs in the flood zone where households were sampled).
The household survey was implemented between April 10 and May 22, 2017, before the onset of the 2017 flood season.The survey was conducted by a trained team of 16 interviewers and four supervisors.Communities where survey training, focus group discussions, and pretesting were conducted were excluded from the study's sampling frame.

Methodology
Different methods are appropriate for valuing economic damages and hypothetical savings from early warning in different situations.Two common approaches for assessing the value of early warning systems are contingent valuation and cost avoidance (Fritz et al. 2008).The former is an economic technique that asks respondents about their willingness to pay for early warning, but these questions can lead to systematic bias, especially in low-income settings where people may not always accurately value the service (Brouwer et al. 2009).Cost-avoidance calculations involve using cost estimations to determine avoided damages from early warning systems, a preferred approach when input data are available (Klaft et al. 2011).
In this study, we employed a cost-avoidance approach to estimating damages and hypothetical savings.To do so, we captured an economic inventory of assets (stocks) and income (flows) of each household at different periods: before and after the 2016 floods and from hypothetical damage mitigation actions that would be undertaken with both a 3-day and an 8-day warning.Although hypothetical answers introduce uncertainty by requiring a respondent to speculate about potential actions, we systematically provided detailed information about the hypothetical scenario to each respondent to reduce assumptions and bias to the estimates provided. 1Because river floods occur annually and therefore flood impacts may be difficult to remember, we tried to minimize 1 Respondents were read the following script to characterize the hypothetical scenario: I am going to describe a hypothetical scenario-that is related to a technology that is being developed, that would allow households in your community to receive accurate warnings of an upcoming flood 8 days in advance of the flood.Can you remind me how many days of advance notice you received last year?(Enumerator wait for respondent to answer.)Now, I want you to imagine that you will receive accurate information eight days in advance.I want you to imagine that this flood warning will tell you the amount that the water level will those errors by relying on households' experiences of the most recent floods, which had happened less than 1 year before and were among the worst in recent history according to preparatory qualitative research.
Table 1 demonstrates the division of economic stocks and flows and the time frames in which values were captured in the survey.Respondents were asked to list only health impacts directly resulting from the 2016 floods, so no household member health inventory (pre-flood) was included.Health damages incurred as a result of 2016 seasonal floods were measured in terms of missed days of work, school, and caretaking in addition to direct medical expenditures.The survey did not capture loss of life or households that were destroyed; these data are officially recorded in Bangladesh and available from external sources.
As part of the in-person survey, households were asked about the specific damages caused by the 2016 flood and what mitigation activities they could have potentially undertaken with advanced flood warning.Table 2 lists the types of stocks/flows impacts and the potential mitigation actions they indicated they potentially could have undertaken with advanced flood warning.The mitigation activities listed in Table 2 were not provide to the households during the survey so as to not bias their response.Table 2  We summed the value of itemized responses in the survey to calculate total household wealth pre-2016 floods and damages post-2016 floods.We calculated the hypothetical savings in two steps.First, we summed the values of each specific item that respondents reported as a hypothetical saving with a warning (e.g., livestock) that was associated with specific actions that a respondent said they could take to reduce damages (e.g., move livestock to safe place).If a household selected more than one action to save one item (e.g., bed), then the value saved (i.e., remaining value after damages) was divided between these actions equally.We then subtracted the cost incurred by a household to complete a specific action from this value, resulting in the final value of the action: Value of action t = (∑ value saved of item i associated with action t ) -cost to take action t .
Second, the values of the specific actions themselves were summed to create the total household hypothetical savings if the damage mitigation action(s) associated with the item could have been achieved with either 3 or 8 days of warning time, respectively.The 3-or 8-day time values were chosen because an experimental early warning tool based on the JASON-2 satellite altimetry has achieved accuracy with up to 8 days of early warning, and other satellite-based warning tools are under development that could achieve similar lead times.The 3-and 8-day warning time frames assumed no rise, specific to your community, so you will be able to understand how high the water will be in relation to your dwelling and to the land that you own everyday for the next eight days.The message will be delivered by a local volunteer in your community or through a cell phone will deliver this message to your household 8 days ahead of time, so that you can act as soon as you receive the message, if you choose to do so.The information that you receive is highly accurate and can be trusted.Now, I want you to think about the floods that occurred last year and the activities you undertook and the damages that occurred related to those floods.Imagine that instead of receiving the warning / no warning that you received last year, instead, you would receive this accurate flood warning from a local volunteer 8 days in advance of a large flood in your community.Now we'll ask you about the additional activities that you would have taken to reduce damage to your household if you had this extra time.
Footnote 1 (continued) communication lag for a warning; rather, they assumed that households receive a warning and have a full 3 or 8 days to respond.

Value of hypothetical savings 8 day =
∑ value of actions achievable in 8 days  Thus, the hypothetical savings are time conditioned in that a household's savings are based on their ability to complete an action for which respondents provided time estimates.Data quality checks instituted during and after data collection ensured that actions could realistically be completed within the 3-and 8-day time windows.

Extrapolation of Survey Sample Values to Full Population of Flood Plain
We used household survey data to extrapolate totals for the flood-affected areas of the six regions included in the survey.To accomplish this, we derived multipliers that would allow us to scale up mean PSU-level estimates to reflect the total flood-affected area.Central to creating the multipliers was the use of the WorldPop data, combined with average district household size (based on our survey results) to estimate the number of households within the sampled PGCs.We estimated there to be 395,041 households within the affected flood plain of the Jamuna River.Multipliers were calculated based on the product of the inverse sampling probability and the estimated total number of households in each PGC.
We calculated the average value of each variable of interest (e.g., total value of assets) for each PGC and multiplied this value by each PGC's multiplier, described previously.These values were then summed across all PGCs to get to final extrapolated values for each variable of interest.

Results
This section presents results from both our survey sample and those that have been extrapolated to all households in the Jamuna River floodplain and are noted accordingly.We first offer an overview of the sample's demographic characteristics (Tables 3 and 4) and then Value of hypothetical savings 3 day = ∑ value of actions achievable in 3 days present the baseline income and wealth of households before the 2016 floods, the damages households sustained because of the floods, and the hypothetical savings households could have realized if they had a 3-day and an 8-day warning (Tables 6, 7, 8, 9 and 10).Throughout, we present the percentage change that each asset type or income activity had across our categories (pre-flood, post-flood, savings) to identify where impacts are proportionally the greatest.
Our sample consisted of 2,247 randomly selected households for which respondent selection was also randomized between heads of household and primary caretakers.Table 3 summarizes the individual characteristics of these household members.Among them, 47% identified as heads of households while 52% identified as primary caretakers.Nearly all heads of households identified as male, and nearly all primary caretakers identified as women, suggesting that the sample represented male and female respondents nearly equally.Most heads of households and primary caretakers in our sample reflected low literacy and educational attainment; over half noted they had not completed any formal education.Most commonly, heads of households worked on farms as cultivators, followed by daily wage laborers in construction, factory, or petty work.In stark contrast, 87% of primary caretakers said they worked in the home as housewives.Table 4 summarizes key characteristics of surveyed households.On average, households comprised 4.2 members and most often had between one and three rooms in the household dwelling.Electricity was not widely available, and many households relied on electric (grid), kerosene lamps, and solar power sources for light.Finally, most households identified as Muslim.
The survey examined household economic well-being through both monetary and non-monetary indicators.The average daily wage for each household member was $1.48 during the dry season, which decreased to $1.05 during the flood season.Wage estimates do not include income from household members listed as students, housewives, retirees, and those not currently working and therefore do not account for any indirect or non-income-generating labor they participate in.
Respondents also reported low perceived economic status relative to others in their local political district.Respondents were asked to imagine a stairway with six steps-with the poorest people of the union being on Step 1 and the richest people on Step 6.Over 80% placed themselves on Steps 1 and 2, and over 95% placed themselves below Step 4.

Flood Experience in 2016
The floods of 2016 represented a recent and highly damaging flood season for many surveyed households, relative to previous years of flooding.Over 90% of respondents said water had reached their dwelling during the floods of 2016, and about two-thirds had flood water reach above the floor level.Of those who experienced water above the floor of their dwelling, 30% said it had reached the height of their bed or mattress (Table 5).
Approximately 90% of all respondents felt that the floods of 2016 lasted longer, brought higher water, and incurred worse damages than typical flood seasons they had experienced in the same area.However, only a small fraction (4% to 6%) of respondents thought 2016 floods were the worst they could remember in terms of duration, water height, and damages.

Household Economic Value (pre-2016 floods)
Respondents provided an inventory of their household assets, livestock, and income before the 2016 floods.Table 6 shows the aggregate sum of all households and average household valuations for each asset and income stream.The largest component of household wealth is derived from owned land and the household dwelling.However, only about half of our sample owned land.The two next greatest sources of households' wealth were livestock and nonagricultural income.The majority of households owned livestock (87%), and although agricultural income represented a smaller share of a household's total wealth, over half of the sample produced agricultural goods for consumption or sale.

Experience with Flood Information and Flood Warning Systems
Notably, only 11% of respondents, all of whom resided in households in the flood plain, received a warning (of any content or via any medium) that a flood was coming (Table 7).
Of those households that did receive a warning, many received different messages.Most commonly, respondents received a message that communicated that river water would rise near their community soon.However, over 93% of households that did receive a warning of any kind believed the warning's content.Among the small number of households that did not believe the message, the most common reason cited was not understanding the message (N = 7).The most common source of warning came from family members, friends, or neighbors; although we cannot track the medium through which they received the warning, our findings indicate that fewer than 11% of our survey sample received warning messages through a formal message channel.

Damages Incurred During 2016 Flood Season
We measured the economic impact of the 2016 floods by household damages directly resulting from the flood reported by respondents.On average, households lost 25% of their income and assets because of the floods.Overall, the largest damages were incurred to land owned by a household (through sedimentation or erosion). 2Agriculture and livestock constituted the next largest monetary impacts from the floods on surveyed households (see Table 8).Proportional to what households owned or earned before the floods began, the largest impact on households was on savings (because of damage repairs) and agricultural income.
In this analysis, we characterized health impacts as treatments costs and foregone wages because of sickness or caring for household members who were sick because of the floods.However, health impacts are underestimated, because they do not account for loss of life or other health impacts that are harder to monetize, such as psychological distress.In addition, we did not include household dwellings that were destroyed by the 2016 floods in our total damage estimates because in most instances the families had left the area and were not available to be interviewed.ReliefWeb estimated that 16,770 household dwellings were entirely destroyed, and 106 people died between July 25 and August 9, 2016 (ReliefWeb 2016).

Avoided Damages in Hypothetical Scenarios
Table 9 shows total household hypothetical savings for a 3-day and an 8-day warning across household income and asset categories extrapolated to the population of households living in the Jamuna flood plain.As expected, households were able to save more with an 8-day warning than a 3-day warning, suggesting that increased flood warning time translates to reduced economic impacts from flooding but at a decreasing return to additional days of warning.The greatest total hypothetical savings from damage mitigation actions derived from protecting land and household dwellings ($45.6m for a 3-day warning and $53.3mfor an 8-day warning).Relative to damages, the two categories that households said they could have saved the most with a 3-day and an 8-day warning were livestock (39% and 42%) and household assets (24% and 25%), suggesting that households are more likely to save these two assets in any hypothetical scenario.However, the highest proportional monetary savings from an additional 5 days of warning is for agricultural income, which reflects a 44% increase in savings, suggesting that households most value an extra 5 days of warning for this category.Health impacts, however, remain static across the incremental increase in warning time, suggesting that households could take the same actions to protect their health with 3 or 8 days in advance.
Table 10 shows a benefit-cost ratio of the actions households said they would take if they had an 8-day warning.This ratio is a comparison between the average amount a household would have saved by taking preventative actions with an 8-day warning and the average monetary costs a household would expend to complete the action. 3Selling assets or livestock before the flood has the highest benefit-cost ratio among the actions evaluated, suggesting that although other actions may have greater benefit, this action may be To value the damages to livestock that were sick because of the floods, we took the difference between the pre-flood livestock value and the amount for which it was sold.For livestock that became sick because of the floods, we valued damages using the amount spent on medicine.To value damages to livestock that died because of the floods, we took the total pre-flood value of the livestock c To value damages to household assets that were repairable, we valued the repair costs.For assets that were not repairable, we valued damages at the total pre-flood asset value the most cost-efficient for households to take. 4 In addition, Table 10 displays the average number of days that each action would take.It shows that certain actions require less time to complete, such as selling assets/livestock, having other household members return home to help prepare, or delay seed preparation whereas others are more time-consuming such as conducting an early harvest, digging trenches around fields/dwelling/livestock, or raising the household dwelling with rods/pillars.

Discussion
This study quantified and detailed the vast damages that households in the Jamuna River floodplain in Bangladesh endured during the 2016 floods, valued at US$1.3 billion or 25% of households' pre-flood valuation.Of all household assets and income sources, land and household dwellings (structures) experienced the highest total damages from the 2016 floods and were cited as having the largest hypothetical savings from advanced warning. 4 Our survey did not explicitly account for changes to market prices for selling assets earlier because of impending floods; although respondents reported selling values they received "before the flood" (i.e., lower prices should have been considered), we did not measure behavior relative to an exact change.Thus, this may be a conservative estimate.Another important consideration is that some of these actions could be considered "one-time actions"; for example, placing a protective fence around a dwelling, although potentially expensive, could last for more than one flood event.Although we did not weight potential "one-time actions" differently when calculating the benefit-cost ratio, this is important to consider when evaluating the potential long-term benefits of a specific action.
These high-value assets are difficult to protect and thus endure high economic damages during flood season.The mitigation actions typically associated with protecting land or household dwellings from damage (e.g., place protective sandbags, dig trenches, raise dwelling) may be expensive and time-consuming, presenting constraints for low-income households.As such, stakeholders may be able to harness resources to help households carry out these actions to protect their most economically valuable assets.Among household wealth categories, livestock was the single category with the highest savings from a hypothetical 3-day or 8-day warning.This finding is in agreement with the benefit-cost ratio of mitigation actions (Table 10); in these results, selling assets/livestock and moving assets/livestock to a safe place were among the three most cost-beneficial actions, largely because they require a relatively short amount of time to complete.Given that all households own at least one household asset and that over 85% of our sample owned livestock, the breadth and volume of savings that an improved early warning system could introduce is significant and could diminish damages for a majority of households in the floodplain.
Agricultural income was the category with the highest damage reduction potential from 3 to 8 days; this additional 5 days of warning, represents a 44% increase in hypothetical agricultural savings.Conducting an early harvest has a high benefit-cost ratio, representing a high economic gain from a large time investment.This and other efforts to reduce damages to agricultural crops are more time-consuming than others and thus harder to carry out with little to no advance warning.Furthermore, with limited warning households were more likely to conduct actions that salvaged personal assets or protected personal wellbeing (e.g., buy/store medicine and food) over actions to protect agricultural income.
Damage estimates by household wealth category point to where proportional losses may be highest, guiding stakeholders to prioritize damage mitigation activities that may maximize returns to a household.Agricultural crops (39%) and household savings (53%) were the categories of household wealth that had the highest percent damage by floods compared to the baseline.These two categories have important implications.First, households lost about half of the savings they had before the floods on average, because of repairs or other needs after the disaster.In a community that is largely at or below the poverty line, an annual event that depletes household savings may severely affect the economic and social well-being of families and affect their long-term resiliency, including their ability to respond to and recover from floods in the following year.Second, over half of our survey sample produced agricultural goods for consumption or sale.Many heads of households work as farm cultivators, suggesting that these damages have widespread effects on households in the floodplain.

Conclusion
This study contributes to the literature on flood damage and mitigation activities in Bangladesh and delivers actionable insights to practitioners for making cost-effective decisions for mitigating household flood damage.Results from our comprehensive survey shed light on detailed household-level impacts and the experiences of households in the Jamuna River floodplain in Bangladesh.In addition, this study explores the potential economic savings that households may have with early warning systems and highlights some areas where practitioners, stakeholders, and policy makers can increase impact through improved flood early warning efforts.
The hypothetical savings in our results and the actions households identify as the most valuable also point to important implications for policy makers and stakeholders involved in developing and implementing early warning systems in Bangladesh.Overall, hypothetical savings from an early warning increase with additional days of warning, but there are diminishing returns to additional days of lead time.When presented with a hypothetical advanced warning of 8 days versus 3 days, households estimated that they could save 6% and 5.5% of their 2016 damages, respectively.This finding implies that most of these benefits-nearly 84%-could be saved with the 3-day warning.Thus, even a relatively short amount of warning time can avoid significant economic losses to households.
A related key finding is that most households in the Jamuna River are not receiving an early warning at all.Communication networks for populations living in the Jamuna River flood plain are largely informal and difficult to penetrate.It is also difficult to develop early warning messages that are actionable for geographically dispersed communities.Thus, priority should be given to combined efforts to improve early warning technologies that can predict floods with those that mobilize flood damage mitigation efforts in vulnerable areas.It could be that the greatest advantage of additional days of early warning is to more effectively disseminate early warning messages to households in the floodplain.If more households received even a minimum of a 3-day warning, this might reduce damages the most because the majority of the early warning benefits are realized in this time frame.Additional days of early warning allow for greater time to communicate warnings to stakeholders on the ground and to translate these into actionable information for households located in different communities.Examples of these early warning dissemination efforts are well-documented (Golnaraghi 2012), and we can learn lessons to improve preparedness actions from them.
Additional research that would help practitioner communities includes the following: (i) Identification of communication channels most appropriate and effective for river communities such as those surveyed in this study.(ii) The cost of communicating early warnings more effectively compared with the potential benefits quantified in this and other studies.(iii) How to improve perceptions and confidence in information that is received.(iv) Determining the impact of "false positives" in future warning systems in the region and the tradeoff between accuracy and level of advanced warning.

Declarations
Competing Interests The authors declare no competing interests.

Fig. 2
Fig. 2 Survey Locations in the Flood Zone

Table 8
Valuation of Damages to Assets, Income, Health (post-flood, 2016) a a These values have been extrapolated to the entire population affected by the flood in our study area b is a composite of what households independently identified as what was possible.

Table 1
Overview of Survey Data Time Frames and Stock/Flow Categories

Table 2
Overview of Survey Data Time Frames and Stock/Flow Categories

Table 3
Head of Household and Primary Caretaker Characteristics (%)

Table 5
Flood Water Height (flood season 2016) a,b a These values come from the survey sample b Standard error in parenthesis

Table 6
Valuation of Assets and Annual Income (pre-2016 floods) a,b a These values have been extrapolated to the entire population affected by the flood in our study area b Asset value at one point in time and income are presented annually bRespondents were allowed to choose all sources from a list of 19 messages that were identified during pre-survey focus groups.We understood "realized by oneself" to mean that households viewed the water levels rising and considered this "advanced warning."Allpercentages are based on the total N

Table 10
Benefit-Cost Ratio of Preventative Actions a a We excluded all actions directly associated with health impacts from the benefit-cost analysis based on our survey's approach to valuing health impacts from a monetary perspective alone.Health impacts in this study comprise the direct monetary costs associated with (1) lost wages from taking care of a sick household member, (2) lost wages from being sick oneself, and (3) expenditures for medicines/treatments associated with an illness caused by floods.Alternative approaches, such as a value of a statistical life methodology, might identify a number that incorporates different economic and quality-adjusted costs, making no single number a reliable indicator of this value The authors of this study have no competing interest associate with this paper's research or presentation of results.Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.