Mapping Reduced Accessibility to WASH Facilities in Rohingya Refugee Camps With Sub-Meter Imagery

TL;DR

The paper tackles inequitable access to WASH in densely populated Rohingya refugee camps by marrying high-resolution remote sensing with a SAM-informed semi-supervised segmentation framework to map refugee shelters and then applying an enhanced 2SFCA to quantify WASH accessibility at fine spatial scales. It introduces a two-stage pipeline: (i) detect shelters from sub-meter imagery using a teacher–student model with SAM-based alignment and pseudo-label refinement, and (ii) compute network-distance-based WASH accessibility, including gender-disaggregated analyses where data permit. Key findings show declining access over time (fewer facilities relative to growing population) and notable gender disparities under safety constraints, emphasizing the need for demand-responsive, equity-focused facility placement. The framework demonstrates how high-resolution sensing and structure-guided learning can inform resource planning in protracted humanitarian settings and is adaptable to other camps and service networks.

Abstract

Access to Water, Sanitation, and Hygiene (WASH) services remains a major public health concern in refugee camps. This study introduces a remote sensing-driven framework to quantify WASH accessibility-specifically to water pumps, latrines, and bathing cubicles-in the Rohingya camps of Cox's Bazar, one of the world's most densely populated displacement settings. Detecting refugee shelters in such emergent camps presents substantial challenges, primarily due to their dense spatial configuration and irregular geometric patterns. Using sub-meter satellite images, we develop a semi-supervised segmentation framework that achieves an F1-score of 76.4% in detecting individual refugee shelters. Applying the framework across multi-year data reveals declining WASH accessibility, driven by rapid refugee population growth and reduced facility availability, rising from 25 people per facility in 2022 to 29.4 in 2025. Gender-disaggregated analysis further shows that women and girls experience reduced accessibility, in scenarios with inadequate safety-related segregation in WASH facilities. These findings suggest the importance of demand-responsive allocation strategies that can identify areas with under-served populations-such as women and girls-and ensure that limited infrastructure serves the greatest number of people in settings with fixed or shrinking budgets. We also discuss the value of high-resolution remote sensing and machine learning to detect inequality and inform equitable resource planning in complex humanitarian environments.

Figures (13)

• Figure 1: Representative segmentation results for high-density (#1), mid-density (#2), and low-density (#3) camps, comparing our method with baselines. Compared to other baselines, our model produces results that more closely resemble the ground-truth (GT) label, effectively handling the complex visual challenges of dense tree canopy noise, congested arrangements, and irregular shelter shapes.
• Figure 2: (A-D) Detected refugee shelters of Cox's Bazar in December 2017 (black), October 2018 (blue), and January 2025 (red). (E) Blue bars displays the total area of refugee shelters across seven time points, as estimated by our model (2017, 2018, 2022, and 2025) and ground-truth data (2020 and 2021). A black line indicates the living space per person (m2 per person).
• Figure 3: Accessibility to WASH facilities based on network distances in 2022 (A) and 2025 (B), and their change (C). Accessibility scores represent the mean accessibility of water pumps, latrines, and bathing cubicles. Changes are summarized at the camp block administrative level. Blue indicates areas where accessibility improved between 2022 and 2025, while red indicates areas where accessibility declined.
• Figure 4: Gender-disaggregated accessibility analysis: average in three facilities (A, D), latrines (B, E), and bathing cubicles (C, F). Blue indicates relatively higher accessibility for women and girls, while red indicates relatively higher accessibility for men and boys. Scenario 1 (A–C) assumes that females are not reluctant to use all-gender facilities. Scenario 2 (D–F) assumes that 25% of females are reluctant to use all-gender facilities, based on thresholds derived from survey data Rohingya_WASH_dashboard.
• Figure 5: Outlines of the 33 Rohingya refugee camps (shown in green) and the extent of UAV/satellite imagery used for the segmentation framework (indicated by red borders). The camps are located in the Ukhiya (A) and Teknaf (B) Upazilas (sub‑districts) within Cox’s Bazar, southeastern Bangladesh. The names of the refugee camps are labeled above the outlines. Camp boundary data, compiled on May 19, 2024, were sourced from the Inter Sector Coordination Group (ISCG). UAV imagery from Camp 21 in Chakmarkul (C) provides an example of a Rohingya refugee shelter layouts. Zoomed‑in views in (D and E) illustrate examples of input imagery and corresponding ground‑truth labels of shelter outlines for the segmentation model.