Image Modified

...

Introduction

Aanleiding

Background

Informal and poor neighbourhoods in cities in developing countries, are often located in flood prone areas. This is particularly the case for Informele en arme wijken in steden in ontwikkelingslanden liggen vaak in overstromingsgevoelige gebieden. Dit is ook het geval in Dar Es Salaam – Tanzania. Hier is wel een vrijwilligersnetwerk (van het Tanzaniaanse rode kruis), maar er is geen tijdige informatie over overstromingsgevaar en geen communicatiemiddelen om ernaar te handelen. Voorspellingssystemen inclusief monitoring zijn te duur en te ingewikkeld om lokaal onderhouden en gebruikt te worden. FloodTags maakt methodes om sociale media te gebruiken als middel om wateroverlast situaties te monitoren, en actief (en passief) de lokale gemeenschap te betrekken bij het voorspellingsproces. Het resultaat wordt via het web uitgeserveerd. In eerdere projecten (o.a. met TKI bijdrage) heeft Deltares een koppeling tussen early warning systemen met Delft-FEWS, en deze social media informatie gemaakt (o.a. in Jakarta), het resultaat is een methode om real-time overstromingskaarten te maken. Deze methode wordt nu ook succesvol toegepast in de Filippijnen. Zo wordt informatievoorziening tijdens wateroverlast behapbaar, community-based en kosten efficiënt.

Deze innovaties maken het mogelijk om een lokale gemeenschap (community) veel meer te betrekken in het voorspellingsproces, en te activeren in de respons, mits er goede kwaliteit modellen beschikbaar zijn om voorspellingen mee te maken. Dit laatste is vaak de beperkende factor omdat er geen historische gegevens beschikbaar zijn om fysieke voorspellingsmodellen van overstromingen en hun impacts op te baseren en te kalibreren, iets wat typisch een langjarige reeks van meteorologische (o.a. neerslag), hydrologische (waterstanden c.q. afvoer) en impact informatie (hoeveel huizen waren beschadigd, hoeveel schade, hoeveel slachtoffers). Het nauwkeurig simuleren van impacts is erg afhankelijk van lokale kwaliteit van overstromingsmodellen en data over wegen, huizen en aantallen mensen in de potentieel getroffen gebieden. 

FloodTags en Deltares hebben samen methodes ontwikkeld om social media te gebruiken in voorspellingsomgevingen. 

Doel

In dit project genereren we en passen we toe, een impact-based voorspellingsmodel voor Tanzania, en specifiek Dar Es Salaam, gebaseerd op data-mined relaties tussen gemonitorde events en impacts in publieke en sociale media, en grote hoeveelheden mondiaal dekkende hydro-meteorologische Earth Observation datasets door Deltares bewerkstelligd in het Earth2Observe project. De innovatie in dit project is dus (anders dan voorgaand TKI) niet gericht op operationele monitoring en flood mapping, maar op het afleiden van impact modellen op basis van historische media en hydro-meteorologische data, die vervolgens in een operationele setting weer ingezet kunnen worden voor voorspellingen. Tijdens het project beoordelen we op welke schaal dit soort impact-relatiemodellen gemaakt kunnen worden (provinciaal op landniveau, versus wijken binnen Dar Es Salaam). Door het gebruik van sociale en publieke media (ipv impactmodellen met daarvoor benodigde grote hoeveelheden gegevens over blootgestelde objecten) en mondiaal dekkende hydro-meteorologische datasets kan in principe overal op aarde een eerste orde impact-forecasting systeem gemaakt worden, dat inzicht geeft hoeveel mensen mogelijk getroffen gaan worden in de komende tijd (uren/dagen). Het resultaat hiervan is dus een schaalbare, en innovatieve methode om impactmodellen in data schaarse gebieden te bewerkstelligen uit BIG data. Dit contrasteert met traditionele meer fysisch gebaseerde modellen, die erg nauwkeurige data nodig hebben over onder meer het drainage systeem, terrein, en blootgestelde bevolking, huizen, wegen en andere objecten.

We ontwikkelen dit, en passen dit toe in Tanzania. De gevonden relaties vormen de grondslag voor een eerste impact-based forecasting systeem en dashboard, gericht op gebruik door de Disaster Management Department van de Tanzania Red Cross Society (TRCS). De in-kind / in-cash contributie van FloodTags wordt onder andere aangewend om TRCS alsmede de Tanzania Meteorological Agency te betrekken bij dit project, en het dashboard aan te laten sluiten bij hun behoeften. De gevonden relaties kunnen ook gebruikt worden om meer fysisch gebaseerde modellen te valideren in een later stadium (dit vormt onderdeel van een Partners voor Water aanvraag in het verlengde van dit TKI project). Door middel van dit project sluiten we aan bij de uitdaging in het Water en ICT KIC (grote hoeveelheden gegevens van verschillende kwaliteit en vertalen naar bruikbare informatie ten behoeve van besluitvorming voor complexe vraagstukken met veel stakeholders. Bovendien sluiten we aan bij het aandachtspunt dat Earth Observation data beter benut moet worden. 

Inceptie

Informatievraag

Concept

Resultaten tot nu toe

Overzicht Activiteiten

2015:

• gedetailleerde analyse van gegenereerde kaarten Jakarta door validatie met nieuw beschikbare overstromingskaarten (o.a. uit recent uitgevoerde hydrodynamische modelberekeningen), ten behoeve van aanscherping van de methode
• inventarisatie van methoden om waarnemingen te valideren door vergelijking (in ruimte en tijd) met andere waarnemingen
• verbetering van de herkenning van relevante berichten in Social Media door verbeterde semantische filtering
• uitbreiding van de locatiebepalingsfunctionaliteit door combinatie met open data sources zoals OpenStreetMaps.

2016:

• Onderzoeken gebruikersvraag
• Ontwikkelen concept
• verbeteren van de methode van het genereren van de overstromingskaarten door ook waarnemingen zonder referentie van een waterdiepte op te nemen: hierdoor neemt de hoeveelheid voor de methode beschikbare waarnemingen sterk toe
• Ontwikkeling van een methodiek voor toekenning van betrouwbaarheidsindicatie van berichten door her- en toekenning van “champions”: betrouwbare gebruikers
• ontwikkelen van een methode om de betrouwbaarheidsindicatie van de waarneming te integreren in de methode voor het genereren van overstromingswaarschijnlijkheidskaarten.
• toepassing van de methode in twee andere gebieden waar zich recent overstromingen voor hebben gedaan.
• validatie van de gegenereerde overstromingskansenkaarten en aanscherpen van de methode

Contact en deelname

The Tanzanian Red Cross Society has a volunteering network for relief actions, but there is no timely information about flood hazards and no means of communication to act upon them. Forecasting systems and monitoring are to expensive to maintain and use locally. FloodTags develops software and methods to use social media as a means to monitor floods as they appear, and to actively and passively involve the local community in the forecasting process. The result is being served to the audience through the web. In past projects (in particular TKI project DEL018), Deltares made a connection between early warning systems using Delft-FEWS, en this social media harvested information, and to convert flood locations into probabilistic flood maps. The result is a method to prepare real-time flood maps from social media. This method is succesfully being applied in the Phillippines (see TKI project DEL018). In this way, provision of information during floods becomes feasible, community-based and cost effective, which may lead to rapid and tailored relief actions after a flood.

These innovations make it feasible to strongly involve local communities in the forecast process, and activate them for response. The question is: can the developed technology also be used to activate them before an event? A prerequisite is that good quality models must be available to predict and event's impact, before the event strikes. Good quality models, predicting flood genesis and impacts are usually not available, because these require long historical records of meteorological conditions (rainfall and other parameters), hydrological data (water levels and rated flows) and impact information (amount of affected houses/households, damage, casualties). Furthermore, a cascade of hydrological, hydraulic, flood iniundation and impact models is likely to result in a cascade of uncertainties as well. Hence, the accurate simulation of impacts is very much dependent on local quality of flood models, quality of hydrometeorological forcing, as well as exposure data on presence of infrastructure, houses, and population in potentially affected areas. 

Goal

In this project, we will establish and apply an impact-based forecast model for Tanzania, and specifically Dar Es Salaam. Other than traditional approaches, we will circumvent the requirement of a hydrology-hydraulic-flood-impact (4 models in total) model cascade, but instead use data-mined relationships between monitored events and impacts in public and social media, and large amounts of globally available hydro-meteorological Earth Observation data, established in the Deltares-led Earth2Observe project. Hence, the innovation in this project is not operational monitoring and flood mapping, but derivation of impact models using historical media records and data on hydro-meteorology, and applying these in an operational setting for predictions and early warning. During the project we will assess at what scale (provincial/country/neighbourhood) we can establish and apply such impact models.

By using social and public media (instead of a physical impact model cascade, requiring enormous amounts of local and accurate data) and globally covering hydro-meteorological data, we can in principle establish a first-order impact-forecasting system anywhere in the world, providing insight in how many people and assets may be affected by floods in the forthcoming time (hours/days). Hence when this project succeeds, the result will be a scalable and innovative method to establish impact models in data poor areas. This is in stark contrast with traditional, more physically based models, requiring very accurate data on the drainage system, terrain and exposed people, houses, roads and other objects. 

We will develop and apply this for Tanzania. The found relationships will form the basis for a first impact-based forecasting system and dashboard, aimed to improve the forecast capabilities of the Tanzania Meteorological Agency, and to further support the Disaster Management Department of the Tanzania Red Cross Society (TRCS). The relationships found can also be used to establish and validate more physically based models at a later stage. This is part of an ongoing application for Partners for Water. With this project, we connect to the challenge in the Water and ICT KIC (translating large amounts of data of different quality to useful information aimed for decision-making for complex questions with many stakeholders). Besides this we contribute to the challenge that earth observation data should be used more appropriately. 

Inception

Requirements

The inception phase of the project included a design workshop, to identify the needs and product requirements for two products, the media monitoring dashboard (this part is co-funded by GFDRR-DFID Challenge Fund) and the impact-forecasting product, envisaged in this TKI project. The design workshop was used. The consortium visited the Tanzania Meteorological Agency (contact persoon Hellen Msemo), the Tanzania Red Cross Society (contact persons Nyambiri Kimacha and Renatus Mkaruka) and the local World Bank office (contact person Edward Anderson). We established a flow of information and then prepared mockups to support these.

Flow of events

The most important consideration in the flow of information is that we have to be aware where the mandate for forecasting and early warning lies, and where the mandate for response. Early warning (whether related to meteorology, severe weather, floods or impacts, expected from floods) is done by TMA, and early warnings issued by any other agency or actor is an illegal act. Therefore, any early warning made should either be issued by TMA, or it should be kept completely internal with the Red Cross and only used for internal considerations.

There are 4 actors involved in the process of using the impact forecast table. These are:

• TMA: The Tanzania Meteorological Agency, responsible for delivering severe weather warnings

• TRCS HQ: The Tanzania Red Cross Society Head Quarters, responsible for alerting the TRCS branch offices and coordinating relief actions

• TRCS BO: the Tanzania Red Cross Society Branch Office of a specific district, responsible for coordination of the volunteering network on the ground in the given district

• TRCS Volunteers: the district level volunteers that operationally provide and distribute relief on the ground

Below, the basic flow of information, foreseen with the product is described. There is a relationship with OP1, the social media dashboard.

1. TMA runs their numerical weather prediction model operationally. Rainfall amounts are accumulated per district and per day.

2. If expected rainfall in the forthcoming 24 hours exceeds 50 mm, a severe warning is sent to DMD, Red Cross and communicated through media. These communications consist of the severe weather warning, as well as a recommendation to individuals living in the region (such as “individuals living near flood zones are recommended to move to high grounds”). If expected rainfall in the forthcoming 24 hours exceeds only 20 mm such severe warnings are sent for those regions that have been dry or are likely already saturated.

3. Both TMA and TRCS HQ have access to the digital dashboard and are able to insert a given forecast rainfall amount for a given province, resulting in a highlighted expected impact for that province region. TMA can use the expected impacts to improve their severe weather warning. For the time being we will stick to the same thresholds as used by TMA (20 mm and 50mm, as well ass dry, moderate and wet conditions) to ensure that our method connects the most to the currently used practices.

4. When this occurs, TMA sends a report in PDF form to TRCS HQ, containing the severe weather warning as well as recommended actions for individuals. With the impact warning table, TMA can also add the expected impacts to the PDF.

5. TRCS HQ takes the impact table for the affected district and looks up what the expected impact will be given the forecast rainfall amount. Either the digital version or the lookup table can be used. At the moment, the information is deterministic, and therefore our impact table also assumes a deterministic impact forecast only. In the future, when TMA considers using probabilistic forecasts, we could adapt this to accommodate probabilistic forecasts as well.

6. Two options are possible

1. If the impact table indicates that there is no impact to be expected; do nothing

2. If the impact table indicates an impact (low, medium or severe) is expected; contact the affected branch office actively by phone.

7. The affected branch office receives the TMA weather warning from HQ and also reads the impact table and decides upon an action. What the level of impact is per severity level, and which preparatory actions should be connected to the expected impacts should be defined with TRCS during the preparation of the impact tables. The actions are likely to be of a preparatory nature. An example is provided below:

1. low: send a whatsapp message to the volunteers to be on the alert, ensure their phones are powered enough to ensure communication channels are open upon an event.

2. medium: do the same as low and start monitoring the dashboard actively (if internet is available)

3. severe: start to take preventive actions. In particular ensure relief places (schools, churches, hospitals, police stations) are aware of the potential forthcoming situation, and ensure these are protected as much as possible. Alert the neighbourhood ward manager.

These are example actions and will be extended in interaction with TRCS while developing the impact tables.

8.Branch officers stay in touch with HQ about the situation. When impacts are experienced on the ground, HQ starts monitoring the dashboard and where necessary coordinate relief actions using their available materials (inception report: cars, mattresses, blankets, mosquito nets, buckets).

Concept

The requirements led to a mockup of the impact table. We are currently pursuing both a "paper" form impact table and a very simple to understand dashboard. The paper table can be read without a computer by any field officer or regional red cross office. 

Image Added

The dashboard is more advanced and can be read by the Tanzania Meteorological Agency. At the moment we are implementing this using GFS forecasts as a placeholder, but we plan to ensure that the forecasts that TMA produces will be used instead. This would make the dashboard directly connecting to TMA's already ongoing practices. 

Image Added

Activities

2017:

• Design inception mission Dar Es Salaam with the parties involved, these include the Tanzania Meteorological Agency and Tanzania Red Cross Society.
• Set up of Delft-FEWS environment, standard data feeds and media monitoring dashboard / SMS warnings
• Review of methods and datasets in Earth2Observe and other reanalysis projects - result: list of historical meteorological/hydrological time series potentially useful for impact forecasting
• Review of online media data availability, setup of queries and access to repositories. The result is an online media dataset of public media based flood conditions and impacts in Tanzania / Dar Es Salaam (including geolocations at provincial, city or neighbourhood scale)

2018:

• Development of a correlation method to relate hydrometeorological thresholds in Earth2Observe data to events, found in historical media-based database. 
• Applying method in Dar Es Salaam and 1 other region in Tanzania, selected with visited groups during the inception mission
• Setup of data feeds for impacts models in Delft-FEWS
• Monitoring and testing period (monitoring real events)

Contact

Questions about this project, possible applications elsewhere and related new applications, can be directed toVoor vragen over dit project, mogelijke toepassingen in andere gebieden en nieuwe toepassingen in het verlengde kunt u contact opnemen met:

• Hessel Winsemius (hessel.winsemius@deltares.nl)
• Jurjen Wagemaker (wagemaker@floodtags.com)

...