SDG 6: A Rohingya Refugee Camp Analysis

Utilizing the power of deep learning, hydrology & network analysis to empower site facilities overview in the Rohingya refugee camps

    SDG 6

    Ensure availability and sustainable management of water and sanitation for all

    Targets


    Study Area

    The camps are located in the Cox's Bazar district in Bangladesh. For our analysis, we chose to focus on just a subset of that area which crossed over several camp boundaries in an attempt to get a more representative sample.

    Area of Interest in Cox's Bazar District, Bangladesh


    Analysis

    Questions

    1. What percent of the refugee camp do not have access to a wash room within a 2.5 minute walk time based on network analysis?
    2. How does a coverage analysis based on DEM and Cost Surface compare to network-based analysis?
    3. What percent of the population lives in the most critical areas of flooding?
    4. How many wash rooms and latrines are located within the most critical areas of flooding?

    Findings

    Question 1

    • Our study area consists of 1.8% of the population that do not have access to a wash room within a 2.5 minute walk time
    • Our study area consists of 24% of the population that are greater than a 1 minute walk time from the nearest wash room facility

    Question 2

    • Coverage analysis using the distance accumulation tool provides a comparable result as analysis based on network analysis. This approach can help augment an analysis where no road network dataset is available.

    Questions 3 & 4

    • Approximately, 27% of the population are located within the most critical areas of flooding
    • There are 104 latrines and 45 wash rooms located within the most critical areas of flooding

    Data

    Finding the Population

    To start any analysis, you need to identify what datasets are available and which of those will be most useful in helping you address your questions.

    Our questions center on understanding the population. We chose to derive this information using drone imagery and an available deep learning package hosted by the Living Atlas.

    When it comes to Rohingya refugee camps, our first milestone was to get the footprints of tents and structures built in these camps. To do so we utilized drone imagery shared by  International Organization for Migration  (IOM).

    For this workflow we utilized tools available with  ArcGIS Image Analyst for ArcGIS Pro .

    With the drone imagery and the deep learning package, we can begin by running the  Detect Object Using Deep Learning  tool in ArcGIS Pro.

    The initial results, as expected, were not relevant at all and most of the tents footprints were missing.

    Therefore, the next step was to go through a retraining workflow. We first digitized a training dataset and then used the  Export Training Data for Deep Learning  tool in ArcGIS Pro.

    Export Training Data for Deep Learning

    We were then ready to re-train the model and extract our footprints!

    We utilized  Train Deep Learning Model  ArcGIS Pro geoprocessing tool to retrain the model.

    Re-train deep learning model

    Re-training progress

    Creating a training dataset was well worth our time.

    The results were inferenced with  Detect Objects Using Deep Learning  ArcGIS Pro geoprocessing tool

    Extracting tents features

    After Regularizing the extracted features. We were able to locate Tent usage information  Camp Facilities  layer from IOM.

    We then spatially joined the facilities layer with extracted tents to derive the next map.

    Here is the final output of the model, classified by "Tent Use".

    It is accurate to almost 85%

    To estimate population, we took an average of 1 person living per 8 square meters (from UNHCR) and the total area of the tent.

    ArcGIS Pro Notebooks were utilized to help with this process.

    Estimating Population

    Washroom capacity was also derived from a metric shared by UNHCR. Capacity averages around 7 households per day per 6 square meters washrooms.

    ArcGIS Pro Notebooks were also utilized in this process.

    Washrooms Capacity

    Creating the Network

    Now that we have an idea of where refugees are living within the camp, we wanted to understand how they move around. As our area of interest is in a remote area, there was not a readily available network dataset to drop in. However, one source we found was OpenStreetMap. The coverage from OSM in Bangladesh was able to provide us the necessary data to create a network dataset.

    Although a lengthy process, we ended up with a traversable network

    For travel mode across this network, a walk-time was established as length over time. In our case, we used 83 meters per minute as described in the  Sphere Minimum Standards and Indicators for Humanitarian Response .

    Adding in our walk time in the network dataset settings

    We also needed to validate the network Dataset with sample points.

    The network dataset was ready to be plugged into our analysis.

    ModelBuilder Workflow

    Running and disseminating results for the Service Area and Origin Destination Matrix analysis through geoprocessing tool built in ModelBuilder.

    Location Allocation Analysis

    Augmenting the Network with Coverage Analysis

    In order to augment and validate the network-based analysis, we performed the coverage analysis by incorporating  Tobler's Hiking Function  based on the slope of the terrain. The analysis used Distance Accumulation to generate a walk-time surface based on a 2.5 minute cut-off time.

    To get as detailed as possible, we were able to locate a dataset of 1 meter contour lines for our Area of Interest. Using the Topo to Raster tool in ArcGIS Pro, a DEM was generated as the base dataset.

    Left: Contour Lines | Right: DEM

    Flood Risk

    In order to address questions three and four, we needed to create a flood risk layer for our area of interest. To do this, we used an existing ArcGIS Learn Lesson,  Predict floods with unit hydrographs .

    ArcGIS Learn Lesson

    Using the same DEM as mentioned above, we walked through the entire lesson and ended up with an isochrone map, showing us the speed at which water will travel throughout the watershed.

    Below is our output watershed for our AOI. The darkest blue area is telling us that water will spend the shortest amount of time in that area increasing the chances of flooding.

    Output Watershed and Time to Travel (in seconds) for our AOI

    Hydrograph Modelbuilder


    Conclusion

    Question 1

    What percent of the refugee camp do not have access to a wash room within a 2.5 minute walk time based on network analysis?

    Additionally, we wanted to determine the nearest wash room facility to each refugee tent.

    Finally, we ran a location allocation maximize capacitated coverage analysis model to determine and assign the resource allocation of the total population and tents to the most optimal washroom location.

    ArcGIS Dashboards

    Question 2

    How does a coverage analysis based on DEM and Cost Surface compare to network-based analysis?

    Difference in coverage between DEM only (Left) vs DEM + Paths (Right)

    Questions 3 & 4

    What percent of the population lives in the most critical areas of flooding?

    How many wash rooms and latrines are located within the most critical areas of flooding?

    Limitations

    • The detailed imagery we had only covered a small area across a few camps. Having finer resolution imagery for the entire area would be ideal.
    • Higher resolution DEMs for the entire area could also improve the Coverage Analysis
    • More time would allow for a more thorough examination of the risk of flooding across the entire camp
    • The most up-to-date camp facility dataset would be preferred
    • Working knowledge is required for Network Analysis implementation in Model Builder
    • Network Analysis field mapping in Model Builder environment is challenging

    Outlook

    • Make network dataset more robust an refined with junctions and linear features connecting
    • Run this analysis across all facility types across multiple refugee camps and determine what facility type tends to be most inaccessible. Use that finding to focus resources during a refugee camp’s establishment. 

    Resources

     

    Although a lengthy process, we ended up with a traversable network

    Adding in our walk time in the network dataset settings

    ArcGIS Learn Lesson

    Export Training Data for Deep Learning

    Re-train deep learning model

    Re-training progress

    Extracting tents features

    Estimating Population

    Washrooms Capacity