Turning Geospatial Data into Planning Decisions: Evaluating a Participatory Accessibility Model for Urban Drinking Water

UBDC's Oluwatimilehin Adenike Shonowo has been conducting research into these multidimensional access challenges.

Introduction

Urban planning increasingly relies on geospatial data to identify underserved populations and prioritise infrastructure investment. Spatial models can map where services are located, estimate travel time, and generate deprivation surfaces that appear precise and objective. But producing a map is not the same as understanding access.

In rapidly urbanising cities across Sub-Saharan Africa, proximity to a waterpoint does not automatically guarantee adequate service. Infrastructure functionality, drinkability, waiting time, affordability, and seasonal reliability all shape lived access conditions. Accessibility is multidimensional.

This research asks a critical question:

When we model drinking water accessibility using advanced spatial methods, how well do those outputs reflect lived realities?

To answer this, a participatory accessibility modelling framework was implemented and quantitatively evaluated in two large Nigerian cities: Kano and Lagos.

Modelling Accessibility

Spatial accessibility models estimate the potential interaction between population demand and service supply. One widely used method is the Enhanced Two-Step Floating Catchment Area (E2SFCA) approach. It integrates population demand, service supply, and network-based traveltime with distance decay.

The model generates continuous accessibility scores under defined behavioural and infrastructural assumptions. It is widely used in health service planning and increasingly applied to infrastructure contexts. However, accessibility models remain structural abstractions. They capture physical reachability under specific parameters but do not automatically capture lived experience.

To strengthen interpretation, this study explicitly evaluates model outputs against participatory validation data collected at grid level.

Conceptual Framework

Drinking water accessibility was conceptualised as a spatial interaction between 100m population grid cells and qualified infrastructure supply.

Unlike simple proximity models, service quality was incorporated into the supply term. Waterpoints were classified into three service categories: Optimal water (functional, improved,drinkable), Moderate water (functional but not meeting all optimal criteria), and Limited water(non-drinkable sources).

These categories were operationalised using weighted supply coefficients (1.0, 0.5, 0.2respectively). Accessibility therefore reflects both spatial reachability and relative service adequacy.

However, affordability magnitude, queueing duration, volumetric capacity, and seasonal variability were not explicitly modelled due to data constraints. The model estimates physical accessibility under network-constrained walking conditions, not full multidimensional service adequacy.

Travel-Time and Threshold Assumptions

Accessibility was estimated using pedestrian network travel times derived from OpenRouteService and OpenStreetMap data. A walking speed of 5 km/hour was assumed.

A 60-minute catchment threshold was adopted. While international guidelines often recommend 30 minutes as a benchmark for acceptable access, empirical evidence from the case study areas indicates that households frequently travel up to an hour to collect water in underserved neighbourhoods. The 60-minute threshold therefore reflects observed coping realities rather than ideal standards.

Accessibility scores were standardised within each city and classified into three deprivation levels: Low, Medium, and High.

Evaluating Model Performance

Rather than stopping at map production, grid-level model classifications were compared directly with participatory validation data aggregated at the same spatial resolution.

Performance was assessed using overall accuracy, class-specific recall, and macro-averaged F1 scores. This allows examination not only of whether the model aligns with community-reported conditions, but how it behaves across deprivation levels.

Results: Kano

In Kano, 1,474 grid-level validation observations were analysed.

Overall accuracy: 26%

Macro F1 score: 0.20

Class-level performance revealed strong asymmetry:

1. Low deprivation recall: 0.00

2. Medium deprivation recall: 0.16

3. High deprivation recall: 0.99

The model demonstrated near-complete sensitivity to severe deprivation but systematically over-predicted the High category and failed to identify validated Low-deprivation areas. This implies potential over-prioritisation of interventions if outputs are used without contextual interpretation.

Results: Lagos

In Lagos, 543 grid cells were classified and validated against 2,094 participatory assessments aggregated to 423 grid cells.

Overall accuracy: 29.31%

Macro F1 score: 0.23

Class-level performance revealed strong asymmetry:

1. Low deprivation recall: 0.50

2. Medium deprivation recall: 0.456

3. High deprivation recall: 0.012

The model moderately captured intermediate deprivation but almost completely failed to detect severe deprivation zones. This creates the risk of under-prioritising the most constrained areas if relying solely on model outputs.

Cross-City Insights

Although Lagos showed slightly stronger overall performance than Kano, the two cities displayed opposing classification biases. Kano exhibited systematic over-detection of High deprivation, while Lagos demonstrated systematic under-detection of High deprivation.

This demonstrates that accessibility model behaviour is highly context dependent and class sensitive. Misclassification is not random and cannot be assumed to transfer uniformly across urban contexts.

Why This Matters for Planning

Geospatial accessibility models produce visually persuasive deprivation surfaces that appear precise and technically rigorous.

However, network-constrained physical accessibility does not fully capture lived access conditions. Classification blind spots are unevenly distributed across deprivation levels, and unvalidated model outputs may introduce systematic bias into infrastructure targeting decisions.

Participatory validation strengthens spatial modelling by revealing where assumptions diverge from community-reported realities, enabling more cautious and defensible planning interpretation.

Conclusion

Spatial accessibility modelling is a powerful tool for infrastructure planning. It enables systematic comparison between population demand and service supply across large urban areas.

However, turning geospatial data into planning decisions requires more than computation. It requires evaluation.

By integrating participatory validation into accessibility modelling, this research demonstrates that quantitative spatial methods can be strengthened through community-grounded evidence, producing more cautious, context-sensitive, and defensible planning insights.

Future work will focus on refining threshold specification and recalibrating classification boundaries using validation informed adjustment procedures. Such refinement aims to improve alignment between structural accessibility modelling and lived water access realities while maintaining methodological transparency and reproducibility.