Towards impact-based flood forecasting and warning in Bangladesh : a case study at the local level in Sirajganj district

Impact-based forecasting and warning services aim to bridge the gap between producers and users of warning information by connecting and increasing synergies between the components of effective early warning systems. We tested qualitatively whether a warning message based on colour codes is understandable and useful to trigger risk mitigation 15 actions at the local level in the flood-exposed communities of Rajapur and Ghorjan unions in Sirajganj district, Bangladesh. With a community-based approach for different groups of users (i.e. sectors), flood-impact scenarios were determined from past events and related to colour codes. These were developed into impact-based forecasting and warnings that can connect water levels, through the colour code, to localised guidance information tailored to sectors‟ needs on how to respond to the expected flood. This approach was tested through a limited number of focus group discussions and interviews at the 20 community level. Overall, the colour coded impact-based warnings were found to be an easy and understandable way to link water level forecasts to the necessary risk mitigation actions, however, further investigation is needed to validate these findings under real-time conditions. IBFW has huge potential in Bangladesh but its integration requires significant institutional changes, such as an inter-facing agency (long term) or team (short term), adjusted policy frameworks (standing orders on disasters), and new resource allocations for skills development and technological innovation from national to local 25 levels. Overall, this paper aims to offer a first insight into impact-based forecasting and warning services in Bangladesh to trigger further research and project developments.


Introduction
Bangladesh is one of the most vulnerable countries to floods, often resulting in high socio-economic impacts (Haque and Zaman, 1993).The reason lies in the combination of the poor socio-economic features with the unique geophysical location in the vast floodplain at the confluence of three major rivers forming the Ganges -Brahmaputra -Meghna basins (GMB) Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.(Islam et al., 2010).Particularly, Sirajganj district, located on the bank of Brahmaputra, has experienced high damages caused by severe flood events in 1988, 1998, 2004, 2007and 2010(Bhuiyan and Al Baky, 2014).The Jamuna River, the main branch of the Brahmaputra River bounding the eastern side of the district, is a braided river forming the so-called chars (islands in Bengali), which are unstable and flood prone morphological formations.Nevertheless, chars are home to some of the poorest and most vulnerable people in Bangladesh (Paul and Islam, 2015).Traditional embankments are ineffective risk mitigation measure (Hossain Md. Z. et al., 2008) because of the dynamic shape of chars.Therefore, policy interventions targeted to cope with the yearly floods are necessary to reduce adverse impacts (Haque and Jahan, 2015).
Early Warning Systems (EWS) are risk mitigation measures targeted to deliver information on an emerging hazardous event, to enable actions in advance that reduce the risks involved (Basher, 2006).A complete and effective EWS comprises of four inter-related elements: a) risk knowledge, b) monitoring and warning service, c) dissemination and communication and d) response capability.A failure in one of these components determines the failure of the whole system (UNISDR, 2006).Risk knowledge refers to the combination of hazards and vulnerability at specific locations through systematic data collection and analysis targeted to understand the dynamic nature of risk.The monitoring and warning service includes the scientific aspects required to timely forecast the hazard and prepare accurate warnings in space and time.Dissemination and communication ensure warnings reach all those at risk, and provide clear, simple and useful information through different channels.Response capability makes sure that warnings are received and understood activating all the actions needed to reduce risk (UNISDR, 2006).The World Meteorological Organization (WMO), conveying the outcomes of the Third UN World Conference held in Sendai (UNISDR, 2015), has highlighted the need for impact-based forecasting and warnings (IBFW) services to bridge the gaps hindering effective EWSs (WMO, 2015).The term "impact-based" aims to translate the hydro-meteorological forecast by shifting the paradigm towards end users, which is forecasting the expected consequences of hazards for different sectors of interest.Although recently introduced, best practices can be found in national meteorological services, like the UK Meteorological Office (UK MetOffice, 2017) and the United States National Weather Service (US NWS, 2017), as well as in international programs leaded by WMO through dedicated workshops (WMO, 2017).
In order to steer EWS towards impact-based services, Table 1 gathers the indicators mostly suitable for this purpose.The indicators were selected from the "Early warning checklist" (UNISDR, 2006) and supported by further literature.

Vulnerability mapping
Vulnerable elements and critical infrastructures need to be mapped, documented un updated periodically.

Timely and accurate forecast
Good quality data have to be collected and processed in a real or quasi-real time to produce meaningful, timely and accurate forecasts.
Impact-based thresholds Warnings must be prepared and issued based on expected impacts severity, enabling end user to take appropriate risk mitigation actions.

Geographic specific warnings
A dense monitoring network ensures a good coverage of the forecast being more specific, issuing warnings to localized targets.Cumiskey et al., 2015;Oktari et al., 2014;Shah et al., 2012.Sector specific warning Warning content have be prepared and issued to different end users clusters with same needs.

Robust standing operating procedure
Government policy establishes the warning dissemination pathway and the roles for defining specific impact-based warnings.Oktari et al., 2014;Rahman et al., 2012;Scolobig et al., 2012.

Complete and timely dissemination
Warnings must reach the entire exposed communities, including those in remote areas, in time before the hazardous event occurs.Oktari et al., 2014;Rahman et al., 2012.

Multiple dissemination channels
Exposed communities must be warned via different media according to their ability/possibility of using them.Oktari et al., 2014.

Information on impacts and advisories
Messages must contain information on the expected impacts and the advisories on how to implement risk mitigation actions.Shah, 2012;Basher, 2006.

Community and volunteers education
Volunteers must ensure the coordination at the local level, helping communities to effectively respond to alerts.Cumiskey et al., 2015;Scolobig et al., 2012.

Preparedness and contingency plans
Hazard and vulnerability maps are used as a management tool for improving response and coordinating emergencies.Cools et al., 2016;van den Homberg et al., 2016;Dewan et al., 2014;Scolobig et al., 2012.

Local community participation
End users can actively contribute to all the four components of EWS.Cools et al., 2016;Fakhruddin et al., 2015;Maidl and Buchecker, 2015;Dewan, 2014;Rahman et al., 2012;Basher, 2006.In the Bangladesh context, the Department of Disaster Management (DDM) is responsible to collect and maintain risk The FFWC provides warnings with respect to "danger levels" (DL), fixed by BWDB for each station.A DL at a river location is defined as the level above which it is likely that flood may cause damages to nearby crops and homesteads (FFWC, 2016).DLs are representative for large areas surrounding the forecasting stations, therefore floods might occur either for lower or higher levels at different locations.FFWC associates a green band for WL up to 50 cm below the DL, then yellow up until the DL, magenta for up to 1 m beyond DL and red for WLs more than 1 m above DL., 2010).This network typically receives warnings from the DMIC communication system, via emails, phone calls and SMS.Furthermore, the information can be accessed by anyone via Interactive Voice Response service (IVRservice on request by calling 10941), FFWC website and in some cases on TV and radio.Union Digital Centres (UDCs) provide accessibility to such facilities also in rural areas (A2I, 2017).
For a schematic diagram of the institutional network for warning dissemination see Shah et al. (2012).
Despite the progresses in the forecasting technologies, major challenges remain to realise the potential benefits of an EWS (Cools et al., 2016).Limiting factors have been identified for Bangladesh and listed below:  Hazard, exposure and vulnerability of the risk environment are not sufficiently evaluated through a comprehensive and shared approach among stakeholders and institutions (Fakhruddin et al., 2015).In Bangladesh, flood risk is highly dynamic as river morphology, population and society constantly change.location-specific information on expected impacts and actions for different users (van den Homberg et al., 2016;Rahman et al., 2012;Shah et al., 2012). Although advances in forecasting and communication technologies are improving access and timeliness of warning, there is still room to improve the design and layout of warning messages.Currently flood forecast bulletins or warning messages simply contain the forecasting station name and its expected rise or decline in WL in the coming period.Not all the exposed population is warned by forecast bulletins, especially remote locations where the existing dissemination means may not be effective. Warning lacks meaningful guidance information on what water levels mean and how risk mitigation actions can be supported.Improvements can be achieved by giving evidence on expected impacts and possible advices, which would require strong collaboration between FFWC, DDM and other relevant stakeholders (Shah et al., 2012).
As consequences, warning thresholds might be misrepresenting the actual flood levels for which impacts occur, warnings can be imprecise for specific areas, warnings lose meaning concerning their purpose of reducing risk as risk is not specified into the message, warnings might not be implemented into actions.Improving access to IBFW services in Bangladesh could offer a mean to tackle these limitations by connecting information on the potential impacts by developing impact-based thresholds to produce more targeted and meaningful warnings that could trigger risk mitigation actions.
This research presents the potential of this approach with the aim of triggering further research and developments on IBFW in Bangladesh.

Methodology
The study follows a qualitative approach and was set up in four steps.
1.The study participants at the case study locations were identified and clustered in groups, representing the key stakeholders from different end-user groups (i.e.sectors).2. From past events, flood impact scenarios were determined investigating flood consequences and related flood peak WLs (reference floods).Then, impact-scenario thresholds (expressed in WLs) could be determined too and impact mitigation actions for the flood-exposed groups were identified for each scenario.Impact-scenario maps were produced by transposing the reference floods WLs at the case study locations, where detailed geographical spatial data were gathered from existing sources.3. Colour codes were associated to each impact-scenario, thus creating impact-based and colour-coded bands.Each band represents warning ranges related to WLs by knowing impact-scenario thresholds.Colour-coded warnings were then developed: while warning messages can address main information (i.e.location, time, expected water level, etc.), the correlated colour code acts as a visual link to local guidance-information like maps, expected impacts and suggested risk mitigation actions for facing the expected consequences (Figure 1).Example of such guidance information were tested at the local level together with dissemination channels and message content required to trigger effective risk mitigation actions.4. Meetings with key authorities involved in the EWS service at both national and local level were performed in order to discuss the requirements for making IBFW an operational service in Bangladesh.

Community-Based Approach
A community-based approach (CBA) was chosen as data collection method for step 2 and step 3 of methodology.CBA was preferred to conventional risk assessment tools (i.e.water depth-damage curves) as water depth-damage curves at the case study areas were not available for the considered sectors, although studies for rice crops damage assessment were performed (Kwak et al., 2015).For the study"s purpose, CBA had the advantage of providing site-specific local data, collected directly from the community level at the case study locations.However, data are representative for small samples, therefore highly subjective and difficult to validate.Nevertheless CBAs methods were successfully applied in previous studies, for example, to plot risk maps based on past flood events and collecting vulnerability data (Smith et al., 2016;Fakhruddin, 2015;Bahauddin and Uddin, 2012).
Data were collected during the fieldwork which took place in May 2016.Focus Group Discussions (FGDs) were used to: gather multiple responses, deepen data through dynamic discussions with participants and to test warning messages.Six FGDs, for 40 participants, were conducted.Each discussion group represented a sector of interest (i.e.farming, education and "disaster management") in two case study areas (see Table 2).FGDs were led with the help of a checklist to pursue the objectives of step 2 (i.e.reference floods, impact scenarios, mitigation actions) and step 3 (test of a possible impact-based warning message) as introduced in the methodology.Semi-structured interviews (SSIs) were conducted to complete step 4 (requirements for effective IBFW in Bangladesh).SSIs, with the help of a checklist, were conducted with 13 governmental officials and knowledgeable experts.(low preference) to 5 (high preference) and later reported in pie charts for analysis (Likert, 1932).

Case study locations and participant selection (step 1)
This study focused on the flood-prone unions of Rajapur (Belkuchi sub-district) and Ghorjan (Chawhali sub-district), both located in Sirajganj district, Rajshahi division (Figure 2).Rajapur lies on the right bank of Jamuna River, protected for one third of its extension by a large embankment on the western side.Ghorjan mainly lies on chars, where no embankments are built.The closest monitoring station for which WLs are forecasted is located near Sirajganj City, the district"s headquarters, roughly 15 km and 35 km upstream of Rajapur and Ghorjan respectively (Figure 2).At the case study locations, the "farming sector" was selected as it would mostly benefit from early warnings (Fakhruddin et al., 2015).The farming sectors" main activities are agriculture and domestic-scale farming (cattle, goats and vegetable gardening) which suffer from flooding every year (Fakhruddin et al., 2015).Major crops grown in the study area include rice drivers and handloom industries were also identified during the SSIs as sectors that could benefit from EWS, however they were excluded due to time limitation during the fieldwork.Table 2 summarises the groups involved in FGDs and the number of participants interviewed.The sample size limits this study, both for the number of participants and for the number of FGDs held for each sector.

Reference floods, impact scenarios and mitigation actions (step 2).
The data gathered was as follows: a) Identification of reference floods leading to minor, significant and severe impacts; b) Flood impacts related to the experienced events for each sector; c) Mitigation actions required for reducing the identified impacts.
As currently is practiced by different countries (i.e.UK MetOffice, 2017; Meteoalarm, 2017;Pagasa, 2017;Protezione Civile, 2017;IMD, 2018), four colour-coded classes were considered in this study (according to the international standards of ISO 22324:2015).Each class represents the magnitude of flood impacts: given the normal condition as a no-impact scenario (green), minor impacts is the first level (yellow), followed by significant impacts (orange) and severe impacts (red).
Therefore, neglecting the no-impact condition, three impact-scenario classes had to be determined.During FGDs at the local level, questions were asked to investigate previous flood events according to the experienced impacts.Although it was recognized that the time of flooding influences the magnitude of the related flood impacts, an increasing linear relation with WLs was assumed for this study.By knowing the year of the events, the maximum flood peaks (reference floods) were determined on the time series of observed WLs at Sirajganj station in the period 1980-2015 (Source: FFWC).Community impact-based thresholds were identified as the average WL between the consecutive reference floods for each scenario.In case of multiple events for the same class, the one having the lowest WL (more frequent) was considered as the reference.WLs at the case study locations were then transposed from Sirajganj station using correlation equations, assumed to be linear in this particular case, determined using measurements from staff gauges installed during a previous project at the same locations (RIMES, 2015).Only the formula for Rajapur was validated with further data and hence used in this study.Hazard maps could be plotted by subtracting the classes" reference WL from the digital elevation model (DEM) built over surveyed points (RIMES, 2015).Including vulnerability data (i.e.infrastructures, buildings, land-use, vulnerable elements, etc.; RIMES, 2015), flood risk scenarios maps linked to the appropriate colour-coded class were also plotted.

Impact-based warning message (step 3)
The findings gathered from step 2 were implemented in a warning message sample proposed during FGDs and SSIs.The message was kept short to reflect the space and time limitations of IVR and SMS.In these formats colour codes cannot be visualised.On the other hand, bulletins and announcements via TV/radio can more clearly indicate the colour codes.For these reasons, the example shown was prepared including colour codes also in written and spoken form (Figure 3).The example relied upon the site-specific voice call message developed under a pilot project at the case study locations (Cordaid, 2016).Note that warning does not have impact information, the use of the orange colour will convey the consequent impacts consultable from the guidance information.
"Welcome to the FFWC of BWDB.The information gathered for testing impact-based warning messages was as follows: a) the sequence of colours for expressing impact severity comparing the best practices model (green, yellow, orange and red) and the practice currently used at FFWC (green, yellow, magenta, red); b) test for understanding the acceptance of including colour codes into warning; c) the most appropriate dissemination channels; d) the overall grade of appreciation for delivering impact-based information (Figure 1).

2.5
Operational requirements for IBFW service in Bangladesh (Step 4) The mechanisms through which the different institutions could distribute responsibilities to generate an IBFW was explored.
An IBFW service should be performed on a daily basis.Such tasks would include translating the WLs forecasted by FFWC

Reference floods, impact scenarios and mitigation actions (step 2)
Table 3 shows the past flood events identified as reference floods (section 2.3objective a)).Looking at Table 3, there is a good match within the same geographical area across sectors.However, "minor" and "significant" between Rajapur and Ghorjan were identified differently.This aspect highlights the importance of having more geographically specific forecasting as the hydraulic conditions may differ significantly along the river, leading to imprecise warnings issued according to wrong thresholds.In Rajapur three events were identified as belonging to the "severe" scenario.
On the other hand, only the 2007 flood was mentioned as severe event from Ghorjan"s FGDs.It might be either that the previous events was not experienced by Ghorjan inhabitants or people remembered only the most recent.Perhaps, that char was not formed yet (Paul and Islam, 2015).
Table 4 shows the reference floods" WL for each scenario, transposed from Sirajganj station to Rajapur union using existing correlation formulas available (described in Step 2 of Methodology) for two gauges placed within the union boundary.These were used to plot hazard maps (Figure 4) and thus delimiting the inundated areas.Figure 5 represents the peak WLsrecorded at Sirajganj station -of the events mentioned in Table 3 and the sectorwise impact-based warning bands (impact scenarios) determined with the CBA, plotted for both unions.This result shows that CBA findings can be different in different locations.The differences may also be attributed to small sample sizes, and we hope in the future the presented methodology will be explored further with larger sample sizes at diverse locations.The colour bands identify the colour-coded warning representing both impacts and the flood severity (hazard) leading to these.
Colour codes can then link to guidance information for each sector (e.g.maps, expected impacts and suggested risk mitigation actions).4).The yellow warning band starts 50 cm before the danger level, as per the current application at FFWC.

5
For each scenarios, experienced impacts and necessary risk mitigation actions were collected through FGDs and reported in Table 5 and Table 6.Tables like these might be important for managing risk and the post-event aftermaths, additionally they provide information available for increasing preparedness before the event.The CBA has limitations.First, the FGD participants were selected among the members involved in a project working in the same locations (Cordaid, 2016).However, Rahman et al. (2012) showed that flood risk awareness is remarkably high also for individuals living in remote communities but not participating in similar programmes.Second, the CBA in this study was performed over a small number of interviewees, all belonging to the same village, where both flood categories and impacts may not be representative for other settlements in the same union.Other impact-assessment methods, like water depthdamage curves (DDCs), are determined over adequate sample sizes and hence representative for larger communities.
However, this strength might turns into weakness as specific needs and differences could be underestimated.Furthermore, DDCs are usually determined for few land uses (i.e.agriculture and settlements), not considering other stakeholders thus excluded from impact assessment.On the other hand, CBA ensures that the local needs are considered and can be representative for other sectors of interest.For the purposes of this study, flood categories could be clearly determined with a CBA and hence used to tailor IBFW thresholds.Although the data collected through the CBA proved to be useful for the purpose of this research, future applications would require more in-depth data collection to validate results.This may not be feasible in many applications presenting a challenge for using the approach.

Impact-based warning message (step 3)
The following list reports the results gathered according to the points presented in step 3 of the methodology.
a) The entire FGDs sample (40 individuals) expressed the green-yellow-orange-red colour code sequence as more intuitive than the sequence currently shown on FFWC website.Participants demonstrated interest in colour-coded warnings as it is easier to interpret flood severity.Bulletins, currently not reporting colours codes, can benefit from the linkage of severity classes and colour codes.
b) The example reported in Figure 3 was shown to a sample of 48 individuals (FGDs and SSIs).Eighty-three percent of them strongly agreed with the inclusion of colour codes into the warning message as it is a more intuitive framing of risk and consequent risk mitigation actions, 11% acknowledged the benefits, while 4% and 2% respectively remained neutral and did not agree.This result shows that colour codes can already provide an information for taking risk mitigation actions without the need of interpreting numbers (i.e.WLs) (Fakhruddin et al., 2015).However some end users might favour the existing warning message (i.e.voice call without implementation of colour codes) as colours can be misunderstood.c) Guidance information can be shown either online on web-based applications or on printouts/boards made available at key locations at the local level (i.e.governmental buildings, schools, bazaars, mosques, etc.).Volunteer based warning dissemination was agreed to be an effective solution to reach the very rural areas where current technologies are either not available or people"s ability to understand warning messages is limited.During FGDs it was recognised that multiple reinforcing communication mediums are needed to reach everyone at risk (Oktari et al., 2014).Despite the poor conditions of the local communities, the possibility of using mobile phones for dissemination was widely accepted (Cumiskey et al., 2015).However, respondents indicated that remote areas could experience shortage of electricity for long periods, making those devices less reliable.
d) Seventy-six percent of the sample, composed of 49 interviewees (FGDs and SSIs), strongly agreed that linking colour-coded warnings with guidance information at the local level would benefit the exposed population towards specific risk mitigation actions.Twenty-two percent acknowledged the utility of it, while only 2% disagreed.
Interviewees mentioned institutional buildings, UDCs, bazaars, schools and mosques as locations where to consult guidance information materials.Across the sectors, many stated that the maps" resolution should be focused at the smallest scale, namely villages.Having this level of detail would permit the inclusions of vulnerable elements and critical infrastructure, useful for coordinating relief efforts and used as tool for fostering agencies collaboration (van den Homberg et al., 2016;Dewan et al., 2014;Scolobig et al., 2012).Although this guidance information can be provided, further efforts are required to train and educate end users on how to interpret warning messages, for example with documentaries, drills and media held in the local language (Cumiskey et al., 2015;Scolobig et al., 2012).
It is recognised that these results are very positive and realistically this could vary significantly if tested during a real time situation.Further testing and applied research is needed to define the most suitable IBFW.

Requirements for IBFW service in Bangladesh (Step 4)
Implementing IBFW in Bangladesh will not happen overnight.It requires a long process of upgrading technology, building capacity and enabling both institutional and policy change.We propose some guidance on possible steps that could be taken in the short and long term to move towards IBFW in Bangladesh.
a) Regarding the institutional requirements, in the short-term, the FFWC can take simple steps to integrate area-wise colour coded warnings in the current FFWC website and to set up an interagency team to decide on the guidance information at the national/ district level, whilst continuing to improve the FFWC water level gauge network and modelling capabilities.Such a team would consist of the critical agencies including FFWC, DDM, BMD and if possible some sector specific agencies e.g.health, agriculture.In this case, all the forecasting would be done at the national level but support could be provided on how to interpret these colour coded warnings at local level to provide guidance.Experimentation with probabilistic forecasting and communication and local pilot projects should continue especially research on the behavioural response of recipients to impact/ colour coded messages.In the long run, it is suggested to develop a neutral interfacing agency that joins these institutions both at national level (as above) and at the sub-district level to connect the Upazila government office with the district level BWDB office and local government institutions and other locally relevant people (e.g.UDC entrepreneurs, local champions).The sub-district level was found to be a more appropriate level for performing this task on a daily basis given their manpower and potential access to resources.Such an agency would be responsible to translate the forecasts from  et al., 2015;Fakhruddin et al., 2015;Koks et al. 2015;Bhuiyan and Al Baky, 2014).However, the rapid morphological alterations of the main watercourses might create a challenge requiring periodical updates.Similarly, vulnerability data should be periodically surveyed, mapped and shared for coordinating emergency plans at the local level (Dewan et al., 2014;Scolobig et al., 2012).c) For capacity building requirements, workforce needs expertise of the forecasting system and technical tools for warning preparation (i.e.digital services, maps, etc.).Volunteering activity requires training/drills and coordination with governmental and non-governmental organizations (Cumiskey et al., 2015;Scolobig et al., 2012).
Volunteering activity showed to be an effective means for message delivery, IBFW can trigger its inclusion as institutional practice as per other hazards in Bangladesh (i.e.storms).Education campaigns need to be planned and periodically reiterated, ensuring a high proactivity of the community level, thus creating a direct communication channel for being updated with end users" needs (i.e. for tuning IBFW thresholds in future) (Cumiskey et al., 2015;Fakhruddin et al., 2015).

Conclusions
IBFW services require site-specific forecasting and warnings issued in an understandable way able to trigger risk mitigation actions by informing end users about the expected impacts.This research presented some challenges of IBFW and proposed an approach for implementation in Bangladesh.We investigated the use of colour-codes to relate flood warnings with expected impacts, impact maps and suggested risk mitigation actions, grouped in three scenarios.Each scenario represents a range of water levels identified by a colour band that might lead to similar impacts as experienced in the past.Impact-based warnings can be produced for those forecasted water levels falling into the identified scenarios.Overall, by improving the existing pilot projects outcomes carried out at the case study locations, the colour coded impact-based warnings were found to be an easy and understandable way to link water level forecasts to the necessary risk mitigation actions in comparison to the existing FFWC bulletin.However, further applied research is needed to validate these findings under real-time conditions and to properly investigate the linkage of colour codes with guidance information.Furthermore, studies have to be performed to identify the appropriate medium for communicating such messages according to the behavioural responses to these.IBFW has huge potential in Bangladesh but its integration requires significant institutional changes, such as an inter-facing agency (long term) or team (short term), adjusted policy frameworks (standing orders on disasters), and new Furthermore, research needs to be carried out for correlating the dynamic nature of varying flood impacts in space and time on a large scale that can be still representative of the local context.
information and assessments.Recently, DDM improved their risk assessment through the Multi-Hazard Risk and Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.Vulnerability Assessment Mapping and Modelling (MRVA) project.For flood risk, hazard was computed at national scale for different return periods (25, 50, 100 and 150 years) while vulnerability was defined as the combination of population, housing, livelihoods, critical facilities and infrastructures (DDM, 2016).The FFWC, under the Bangladesh Water Development Board (BWDB), is the agency entitled to monitor and forecast water levels (WLs).Currently the deterministic forecast is available up to 5 days in advance at 54 representative stations on 21 rivers.Precipitation data, necessary to run the forecast, is collected from the Bangladesh Meteorological Department (BMD) and from Indian Meteorological Department (IMD) for the portion of the GMB basin lying in India (BWDB, 2014).
FFWC issues flood forecast bulletins to government and non-government organisations (NGOs), media groups and other concerned parties.Bulletins contain observed rainfall, observed WLs and forecasted WLs for all the stations, emphasizing the expected trends.Warning dissemination from the national to the local level is established and coordinated by the DDM through the Disaster Management Information Centre (DMIC).The main components of this framework are the Disaster Management Committees (DMCs) at different levels: national (National DMC -NDMC), district (District DMC -DDMC), sub-district (Upazila DMC -UzDMC) and local (Union DMC -UDMC).These committees are composed of both government and stakeholder representatives.Their responsibilities include communicating warnings from the FFWC through the dissemination network and supporting the response actions (MoFDM Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 1 .
Figure 1.Use of colour codes to link impact-based forecasting and specific warnings to trigger response at the local level in combination with guidance information.
For both FGDs and SSIs the answers were collected qualitatively (open answers) and quantitatively (closed answers).Data collected from FGDs and SSIs were analysed.Local interpreters translated the response to English when needed.The answers were noted during meetings and later reported as written text.These were then analysed and used to develop the Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.raw data into thresholds, flood impacts, risk mitigation actions and impact-based warnings (from FGDs); institutional needs, technical arrangements and training/educational resources (from SSIs).Closed answers were related to a rating scale from 1

Figure 2 .
Figure 2. The map at the top right corner shows the location of Sirajganj district.Rajapur union, located in Belkuchi upazila, and Ghorjan union, located in Chawhali upazila, are shown in the larger map.The model's WL forecasting node, coinciding with the WL monitoring station, is located on the Jamuna River in Sirajganj city identified by a green dot.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.(boro and aman qualities), sugarcane, jute, tomato, potato, onion, beans etc. Severe floods of 2004 and 2007 caused agricultural damages for more than 17,000 million BDT together (roughly 200 million USD) in Rajshahi division (Haque and Jahan 2015).The "educational sector" was included as it"s likely to be relevant for the study purposes.Students compose a large part of the local community susceptible to flood impacts as they are exposed to waterborne diseases, disruption of the educational service and risk of drowning.Teachers can play a key role in enhancing preparedness by disseminating warnings, cooperating with NGOs and volunteers.A third sector was considered to represent the general interests of the society during emergency situation: the "disaster management sector" including members of the local Union Disaster Management Committees (UDMCs) and the volunteers taking part in warning dissemination.Health, boat owners, rickshaw Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 3 .
Figure 3. Warning messages shown during FGDs and SSIs as an example of a possible impact-based warning.The existing voice call warning was improved by adding the bolded text and shading the text with the colour code.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.at the forecasting stations into a more location-specific, impact-based and colour-coded warning.SSIs" interviewees were asked what organization would be the most suitable for fulfilling this task and what geographical size IBFWs should be prepared for.More generally the investigation aimed to identify the requirements in terms of: a) institutional requirements (i.e.responsibilities and operational framework); b) technical requirements (i.e.forecasting network, WL gauges); c) capacity building requirements (i.e.workforce, volunteering activity, awareness and drills).

Figure 5 .
Figure 5. Site-specific warning thresholds determined with the CBA for each union with respect to Sirajganj station.On the right, the colour bands as per at FFWC for Sirajganj station.On the left, the stage reports the classes' reference WLs and the year they occurred.Thresholds between bands were obtained as average between two reference WLs (see table4).The yellow warning band starts 50 cm before the danger level, as per the current application at FFWC.

Figure 6
Figure 6 schematizes the approach proposed for delivering site-specific impact-based warnings in the short and long term based on the considerations drawn above.

Figure 6 .
Figure 6.Proposed approach for delivering site-specific impact-based warnings at the local level.The framework shows both short-term and long-term solutions proposed in this study for delivering more tailored and understandable impact-based warning to different groups of users.Bolded frames represent the main actors identified for these tasks in the short term.
Nat. Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci. Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License.resource allocations for skills development and technological innovation from national to local levels.More specifically, the spatial domain of the forecasting model, the density of WL gauges and their correlations, and high resolution risk assessments, need to be improved.Resources are required to develop technical skills, increase manpower and coordinate volunteering activities at the local level.Future studies can deepen these findings by investigating up-scaled applications by increasing samples for different sectors, comparing the results with other impact assessment methods (i.e.water depthdamage curves) and doing deeper analysis on the behavioural responses to IBFW warnings in real-time at the local level.
Periodical updates can provide a good know-how of risk at the local level, thus requiring stakeholders, governmental organizations and NGOs to cooperate together with the community (i.e.Community Risk Assessment -CRA) . The FFWC warnings are based on WLs that do not represent the possible impacts at the local level.Instead of being distributed over large and heterogeneous areas connected to one forecast point, FFWC warnings should include Nat.Hazards Earth Syst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26Manuscript under review for journal Nat.Hazards Earth Syst.Sci.

Orange code. Flood may occur in 4 days in Ghorjan union"
Today Friday 29th July 2016.As per the observations of 6 AM this morning Jamuna river at Sirajganj is flowing 10 cm below Danger Level.According to the latest flood forecast water may rise 29 centimetres in Rajapur union, in next 5 days.Yellow code.Flood may occur in 4 days in Rajapur union" "Welcome to the FFWC of BWDB.Today Friday 29th July 2016.As per the observations of 6 AM this morning Jamuna river at Sirajganj is flowing 10 cm below Danger Level.According to the latest flood forecast water may rise 30 centimetres in Ghorjan union, in next 5 days.

Table 5 . Impacts experienced for previous flood events by the sectors involved in the research in Rajapur community. WL is referred at Sirajganj station in mPWD.
Shah et al. (2012)4)yst.Sci.Discuss., https://doi.org/10.5194/nhess-2018-26ManuscriptunderreviewforjournalNat.Hazards Earth Syst.Sci.Discussion started: 19 February 2018 c Author(s) 2018.CC BY 4.0 License. the national level to the local context and provide meaningful guidance information for different colour coded warnings.These warnings can then be further communicated to local volunteers via mobile services, accessible via the UDC entrepreneurs via internet and maps, and shared via mobile and in-person to individuals.This arrangement would require significant investment and resource allocations for skills and technology development within existing agencies, newly developed agencies and local volunteers, to monitor water levels, produce localised forecasts, and communicate them.Furthermore, legislative changes in the Standing Orders on Disasters are required to allocate new roles and responsibilities for IBFW and engrave the institutional changes.b)On the technical requirement" side, specific warnings can be prepared only if WLs can be predicted at smaller geographical areas, namely at the union level.It is now possible in the short term by correlating WLs of the forecasting stations with hydrometric stages placed at the locations where specific warnings have to be issued to.In the long term, a dense forecasting network should be developed by FFWC/BWDB, ensuring a good flood forecasting coverage as identified byOktari et al. (2014)andShah et al. (2012).High resolution DEMs are needed for mapping (forecasted) inundation areas (Kwak