What Does It Take to Map a Country? Scaling OpenStreetMap Mapping for Accurate Health Accessibility Modelling in Madagascar

This study demonstrates that while scaling exhaustive OpenStreetMap mapping to achieve household-level health accessibility modeling across Madagascar is feasible, it is highly resource-intensive, requiring an estimated 220 to 350 person-years to generate the granular geospatial data necessary for equitable health system planning.

The Gist

Imagine you are trying to organize a massive delivery service for a country the size of France, but you don't have a map. You don't know where the houses are, you don't know which paths are dirt trails and which are paved roads, and you certainly don't know how long it takes to walk from a remote village to the nearest clinic.

This is the reality for many countries in Africa, including Madagascar. Without a good map, it's impossible to know if a sick child can reach a doctor in time, or if a health worker can deliver medicine to a remote family.

This paper is the story of a team that decided to draw the map from scratch to solve this problem, and then calculated how much effort it would take to do this for the whole country.

Here is the breakdown of their journey:

1. The Problem: The "Blank Page"

Think of the digital map of the world (called OpenStreetMap, or OSM) like a giant, collaborative Wikipedia for geography. In rich countries, this map is detailed down to the last mailbox. But in rural Madagascar, it's like a blank page with a few scribbles.

• The Issue: If you try to calculate how long it takes to get to a hospital using a blank map, you might guess "1 hour." But in reality, because there are no roads, you have to walk through rice fields and up steep hills, making it a 4-hour trek.
• The Goal: The team wanted to fill in every single blank spot so they could calculate real travel times for every single household.

2. The Pilot: Training Wheels

First, they tested their method in one small district (Ifanadiana). They treated it like a "training run."

• The Method: They didn't just guess. They hired a team of "digital cartographers" (people who draw maps on computers).
• The Tool: They used high-resolution satellite photos (like looking at the earth from a very high-flying drone) and drew every single building, footpath, and rice field onto the digital map.
• The Twist: They didn't just draw lines; they asked local health workers to come in and name every village on the map. It was like a community "show and tell" where locals pointed to their homes on a paper map and said, "This is where we live."

3. The Scale-Up: Building the Engine

Once they proved it worked in one district, they expanded to seven more districts, covering an area the size of Belgium.

• The Team: They hired full-time professional mappers (10 at a time) to work non-stop for over a year.
• The Result: They added 1.5 million buildings and 176,000 kilometers of footpaths to the map. That's like adding every single house and walking trail in a small country to the internet.
• The Payoff: With this new, hyper-detailed map, they could calculate exactly how long it takes to get to a health center.
  • Old Map: "Everyone is close to a clinic."
  • New Map: "Actually, 60% of people in some areas are more than an hour away from a primary clinic, even though they live only a few miles away as the crow flies."

4. The Big Question: Can We Do This for the Whole Country?

The team asked: "If we can do this for 8 districts, how hard would it be to do it for all of Madagascar?"

They used two different ways to guess the answer:

1. The "By the Numbers" Method: They looked at how long it took to map different types of areas (busy towns vs. empty fields) and multiplied it by the size of the whole country.
2. The "Statistical Crystal Ball" Method: They built a computer model that predicted how long it would take based on how many buildings and roads were in each square mile.

The Verdict:

• Time: It would take between 220 and 350 "person-years" to map the whole country.
  • Analogy: If you hired 100 people to work full-time, it would take them about 2 to 3.5 years to finish the job.
• Cost: Roughly $1 million USD (plus coordination costs).
• Feasibility: It is doable, but it requires serious money and commitment. It's not something that happens by accident or with just a few volunteers.

5. The AI Factor: The "Robot Assistant"

The team also tested if Artificial Intelligence (AI) could help. Tech giants like Microsoft and Facebook have AI that can automatically spot buildings in satellite photos.

• The Hope: Maybe the robots can do the work for us!
• The Reality: The robots were okay at spotting big buildings in cities, but they were terrible at finding small dirt paths, footbridges, or houses in rural villages. They often confused shadows for houses or missed tiny trails entirely.
• The Lesson: AI is a helpful assistant, but it can't replace the human eye yet. You still need a human to check the work, especially in complex, rural landscapes.

Why Does This Matter?

Think of this mapping project as building the foundation for a house. You can't build a roof (healthcare programs) if you don't know where the walls (villages) are.

By creating this detailed map, Madagascar can now:

• Find the invisible: Identify exactly which families are cut off from healthcare.
• Plan better: Build new clinics in the exact spots where they are needed most, not just where they are easiest to reach.
• Save lives: Ensure that ambulances or health workers know the real paths to take, not just the ones on a generic map.

In short: This paper proves that with enough effort, money, and human dedication, we can turn a "blank page" into a life-saving roadmap. It's a heavy lift, but for a country like Madagascar, it's the only way to ensure no one is left behind.

Technical Summary

1. Problem Statement

In low-income countries, particularly in sub-Saharan Africa, geographic accessibility to healthcare is a major barrier due to sparse infrastructure and incomplete spatial data. While OpenStreetMap (OSM) is a critical open-source database for geospatial analysis, its completeness in rural Africa is significantly lower than in the Global North (e.g., <50% road completeness vs. near-perfect in Europe).

• The Gap: Existing models often rely on simplified Euclidean distances or generalized travel speeds, which fail to capture the reality of rural travel (footpaths, unpaved roads, terrain variations).
• The Challenge: Previous pilot studies demonstrated that exhaustive, household-level mapping combined with locally calibrated travel models yields actionable data, but the feasibility and resource requirements of scaling this approach to a national level remained unknown.

2. Methodology

The study employed a three-phase scaling approach across eight districts in southeastern Madagascar (covering ~30,200 km² and a population of >2.5 million) to test scalability and estimate national resource needs.

A. Data Collection and Mapping

• Phases:
  • Phase 1 (Pilot): Ifanadiana district (2019–2020).
  • Phase 2: Five districts in Fitovinany and Atsimo Antsinanana (2022–2023).
  • Phase 3: Two districts in Vatovavy (2024–2025).
• Technique: The study area was divided into 1km² tiles managed via a customized Tasking Manager.
• Imagery: High-resolution satellite imagery (Bing and Maxar Premium) was used for photointerpretation.
• Elements Mapped: All buildings, residential areas, road networks (including unpaved roads and footpaths), rice fields, and hydrographic networks.
• Quality Control: A dual-step process was used: one cartographer mapped the tile, and a second validated/corrected the data.
• Participatory Mapping: Community Health Workers (CHWs) were engaged to name villages and hamlets, adding 4,870 village names to OSM.

B. Accessibility Modeling

• Routing: The OSRM (Open Source Routing Machine) was used to calculate the shortest path from every mapped building to the nearest Primary Health Center (PHC) or Community Health Site (CHS).
• Travel Time Estimation: Routes were segmented into 100m intervals. Travel speeds were assigned based on a locally calibrated statistical model incorporating:
  • Land cover (forest, savanna, rice fields).
  • Slope (derived from SRTM elevation data).
  • Seasonal conditions (dry vs. rainy).
• Output: Raster maps of travel time and distance for both dry and rainy seasons.

C. National Scale-Up Estimation

To estimate the resources required to map all of Madagascar, two complementary methods were used:

1. Completeness-Based Extrapolation: Used Kontur's Disaster Ninja framework to classify districts by OSM completeness (Low/Medium/High) and population density. Mapping times per km² were derived from Phase 2 data and extrapolated based on these classes.
2. Statistical Predictive Model: A multiple linear regression model predicted mapping time per tile based on variables: building count, road length, population density, rice field area, and waterway length. This model was applied to national-scale layers.

• AI Reference: AI-generated datasets (Microsoft Building Footprints, MapWithAI/Facebook) were used as reference layers to assess baseline completeness, though they required significant scaling factors due to under-detection in rural areas.

3. Key Results

A. Mapping Output (8 Districts)

• Volume: Added 1.15 million buildings, 82,600 residential areas, 176,000 km of footpaths, and 168,000 ha of rice fields to OSM.
• Baseline: Pre-mapping completeness was extremely low (e.g., <1% footpaths in several districts; <30% buildings in most).
• Participatory Success: 75% of village names in the mapped districts were added via CHW collaboration.

B. Health Accessibility Findings

• Primary Health Centers (PHC): Access is poor. In the dry season, **<42%** of the population lives within 1 hour of a PHC. In districts like Ifanadiana and Vondrozo, >60% live >1 hour away.
• Community Health Sites (CHS): Access is significantly better. >90% of the population lives within 1 hour of a CHS. However, ~10% in specific districts still face >1 hour travel times.
• Seasonality: Rainy season conditions further degrade accessibility, though the study focused primarily on dry season baselines for the main metrics.

C. National Resource Estimation

• Completeness: >90% of Madagascar's districts have "Low" OSM completeness (Class A).
• Time/Cost Estimates:
  • Method 1 (Completeness Class): Estimated 350 person-years to map the entire country.
  • Method 2 (Statistical Model): Estimated 220 person-years.
• Cost: Based on a local salary of $300/month, the direct labor cost is estimated at **$1 million USD** (excluding coordination and overhead).
• Timeframe: 100 full-time mappers could complete the national mapping in 2 to 3.5 years.

4. Key Contributions

1. Scalability Proof: Demonstrated that exhaustive, household-level mapping is feasible at a regional scale (30,000 km²) using professional cartographers and participatory validation.
2. Actionable Data: Generated the first high-resolution, household-level travel time models for Madagascar, moving beyond "as-the-crow-flies" approximations to actual routing on footpaths and unpaved roads.
3. Resource Roadmap: Provided the first quantitative estimates (220–350 person-years) for national-scale OSM completion in a large, data-scarce African country.
4. AI Limitations Analysis: Highlighted that while AI datasets (Microsoft/Meta) are useful for reference, they significantly undercount rural footpaths and informal structures in Madagascar, necessitating human validation and scaling factors.
5. Participatory Integration: Validated the critical role of local Community Health Workers in adding semantic data (village names) that automated systems cannot detect.

5. Significance

• Health Planning: The data enables ministries and NGOs to optimize the placement of new health facilities, schedule last-mile delivery campaigns, and model disease spread with high precision.
• Global South Infrastructure: The study provides a reproducible framework for other nations to assess the cost and effort required to build national geospatial infrastructure.
• Policy Impact: By quantifying the "data gap," the paper argues that investing ~$1M in mapping yields high returns for universal health coverage (UHC) and sustainable development, transforming OSM from a volunteer hobby into a critical public health utility.
• Future AI Training: The creation of high-quality, human-validated ground truth data in these regions will eventually improve the accuracy of AI models for future automated mapping efforts.