Building a detailed flood model with community mapped drainage data
A Research and Innovation Story
To walk with dry feet on the streets is taken for granted in some parts of the world. But did you know that each and every street in a city must be engineered and designed in a way so that it conveys water into the drainage system, or else the street may get flooded when it rains? And did you know that engineering designs of drainage strongly rely on detailed geometric data about the existing drainage system and terrain data? Many countries are lacking the resources to collect this geometric information and to maintain large databases. How can they then design and maintain a functional drainage system, if they do not have the prerequisite data? The Ramani Huria community mapping project in Dar es Salaam tried to address this issue through collecting drainage data with smartphones using a community mapping approach in the years 2017–2019. The Resilience Academy partnered with Delft University of Technology to investigate if this data is suitable for urban flood modelling through the development of a flood model with community mapped drainage data.
By Louise Petersson, Hessel Winsemius, Iddy Chazua, Elia Dominic
Flood modelling is used to understand where flooding occurs in a city, and what measures you can take to mitigate them. This requires a lot of geometric data, including topography, land use and records of rainfall events. To mitigate flood problems on a local scale, on street level, you need to understand the interaction between surface runoff (rainfall water flowing over ground) and the drainage system. In this blog, we explain how a flood model was developed using local resources and devices in Dar es Salaam, Tanzania. The Humanitarian OpenStreetMap Team with its local partner Open Map Development Tanzania was contracted by the Tanzania Urban Resilience Program, the World Bank, to collect large amounts of risk identification data of Dar es Salaam. Research Institute Deltares assisted in setting up technical surveys, and the drainage system of Ilala, Ubungo and Kinondoni municipalities was surveyed by university students between 2017 and 2019.
The development of smartphones is a game changer when it comes to data acquisition across the globe. As smartphones are equipped with GPS-sensors, cameras and internet connection, they are in fact surveying instruments that are available to an increasing share of the global population, also in low-income countries. These surveying instruments, combined with open source applications and data analysis programs, makes it possible for anyone with access to a smartphone to collect geometric data and make maps. With this, the accuracy and quality of data can now be achieved through an ecosystem of accessible components: smartphone applications with clever surveys, trained mappers that know how to map the specific attributes in question in the field, and an extensive and creative (but low-cost!) data validation protocol. An important component of successful data collection with smartphones is conditional surveys, that facilitates that the mapper can only record the data that is relevant for the features in question. For example, round culverts should have a diameter recorded, not width and depth — and a diameter entry should be in a number format, not in text. All such requirements must be correct in the survey to collect a useful data set.
But how accurate is really the data? Is it accurate enough to replace traditional surveys? This was our research question, so we used a systematic approach to investigate the quality of community mapped drainage data and subsequently implemented it in a hydrodynamic modelling software. A full scientific report about the study is found here.
We used a subset of the Ramani Huria drainage data that covered the subwards Mpakani A, Kijitonyama, Bwawani, and Alimaua in Kinondoni municipality. In the figure below, the locations and types of drains in the study area are shown. The drainage elements are divided into four classes: culverts, ditches, drains and decommissioned drains, which in turn can be open or closed and have different cross sections. Decommissioned drains are poorly maintained, e.g., full of vegetation, building materials or other blockages.
The Open Data Kit Collect Android application (ODK) was used for data collection, a free survey application for smartphones that can add geopoints and geotraces to each data entry, making it suitable for the record of drainage segments. OpenStreetMap was used as a platform to store data, making it openly and freely available. The data collection started with the design of an ODK survey to collect drainage data using a community mapping approach. This involved identifying which data was to be collected, how this should be done, but also identifying the affordable tools and open source platforms that could be used in the process. Defining the questions in the survey required several iterations in the field to find the most common drainage elements in Dar es Salaam and their characteristics, including dimensions.
To make the mapping comprehensive and complete, a special drainage data collection team was formed, consisting of students that were instructed how to measure the different attributes and in what order to observe the drainage channel so that no drains would be overseen. The “geotrace” function of ODK makes it possible to track the reach of a drainage segment by walking along the drain while recording the walked path by using the GPS sensor in the smartphone. This method is well suited for open channel networks, which is the most common drain type in Dar es Salaam. As local devices, cross sections of drainage segments were measured with locally made measuring sticks and tape rulers and the dimension values and material attributes were recorded in the ODK app. The vertical distance between the drain bottom and the nearest road was measured at the upstream end of each segment, to derive the absolute elevation of the drain bed levels in relation to a digital terrain model, as altitude cannot yet be recorded by smartphones with sufficient accuracy.
The overall survey was designed so that the individual records, when combined, lead to a topologically correct and connected drainage dataset with a complete set of attributes. The data model is found here.
After recording the drainage system in the field, the next step was to undertake data cleaning and quality assurance of the collected data. Within Ramani Huria, a team of data cleaners was appointed to quality check the results before uploading the data to OpenStreetMap. This involved aligning inaccurate geotraces with drainage segments seen on satellite and drone imagery in GIS, checking the completeness and validity of the attributes and checking the topology and connectivity of the network. Automated quality checks were used for both attributes and topology. The scripts for quality assurance are openly shared here.
If errors or missing elements were encountered, the mapping team was instructed to record these again in the field. The data cleaning was taking place typically at the end of each field campaign day. The components of Ramani Hurias mapping methodology is shown in the figure below.
When the data later was implemented in hydrodynamic software, we found remaining quality issues with the data despite the data cleaning efforts of Ramani Huria. However, only 12 of 532 (2.25%) drainage segments had positional errors and they could be easily corrected manually. The positional errors are shown in the table below.
Table 1. Analysis of positional errors
Some drainage segments were missing information on construction material, amounting to 5.3% of the dataset. We made assumptions about these drainage segments, but for future surveys, it is important to make the material attribute strictly conditional — if construction material is not recorded, the mapper cannot save the record.
The connectivity of the network was analyzed in a hydrodynamic software, which revealed that 36 out of 532 segments, i.e., 6.8%, were hydraulically disconnected from adjacent drainage segments. Upon further inspection in QGIS, it was found that this problem was caused by so-called “T-junctions.” A T-junction occurs when a drainage side-branch connects to a main branch, while the main branch is recorded as a continuous geotrace. A connection from the side-branch to the main branch is not recognized automatically and the side-branch remains unconnected in the modeling software. This problem was solved manually by splitting the segments at T-junctions in QGIS.
To assess the quality of the drainage data, we ran two model simulations of a heavy rainfall event that fell in the study area in the early morning of 3 March, 2019. In the first simulation, we did not include the drainage system in the model, and in the second simulation we implemented the drainage data recorded by Ramani Huria as line elements with water conveying capacities following the surveyed dimensions and properties. To compare the performance of the two models, and see what the difference was when adding the community mapped drainage data, the modelled water depth (the model output) was compared by flood depth data collected by the drainage mapping team. They went into the study area the morning after the rainfall event and asked community members about the depth of water on the street outside of their house the same morning. The community members referred their responses to a person’s height, i.e. if the water was ankle deep, knee deep etc. Their responses were recorded with ODK and were categorized as shown in the table below.
Table 2. Water depth classes used in the model validation.
When comparing the maximum simulated water depth against the flood depth that was reported by community members for the two model scenarios, we see that the simulated water depth corresponds better with the reported flood depth data when taking the community mapped drainage data into account. When running a model of the area without the drainage data (which in practice means that you assume that the drainage system simply is not constructed in the city), the simulated water depths fall below the reported class at 12 locations, within correct reported class at 19 locations, and above reported class at 13 locations (figure below, lower left). When introducing the community mapped drainage data, simulated water depths fall in the correct reported class for 30 out of a total of 44 locations (figure below, lower right). The simulated water depths are below reported class at 10 locations and above reported class at 4 locations. This is a considerable improvement compared with the model run without community mapped drainage data.
Table 3. Summary of model development results. A number of observation points below/in/above reported water depth class (and percentage).
The results show that drainage data collected with a community mapping approach allows for reliable simulations and a detailed understanding of flood patterns on street level. Thanks to this, it is possible to run reliable flood model simulations that can inform urban planning, investments, and upgrading in flood-prone areas, also in environments where extensive drainage data is usually missing.
In 2020, the drainage mapping initiative was expanded to Mwanza at the shore of Lake Victoria. In the next blog post, we outline how the mapping methodology was improved to allow for even more accurate drainage data, acquired with locally available technologies.