How should a country organise waste collection so that sorting facilities serve the right areas without making transport unnecessarily expensive? Research by MBA student Michael Camden-Smith developed a scalable method for KOLEKT’s Waste Simulator that uses population and geography to design collection areas and locate sorting facilities. The developed method revealed solutions within 2% of the lowest possible cost, while remaining fast enough to use at national scale.

Why can a fixed grid be a problem for waste planning?

The geography of waste collection is uneven. Dense cities, sparsely populated rural areas, borders, coastlines and physical terrain all affect how far waste has to travel and how many people a sorting facility needs to serve.

An earlier version of KOLEKT’s simulator divided countries into fixed 50-square-kilometre cells. That made large-scale calculations manageable, but the same grid structure does not reflect differences in population density or geography. A square covering a dense urban area can represent a very different collection challenge from an equally sized square covering a sparsely populated or partly uninhabitable area.

The research question was therefore geographical as well as financial: how can collection areas be designed around where people actually live while keeping the calculation manageable at national scale?

The planning problem starts with getting the geography right.

Why is this an optimisation problem?

Each collection area needs to serve a set of locations and connect them to a sorting facility. Changing the number of collection areas changes both transport distances and infrastructure requirements.

A country with too few sorting facilities may require waste to travel long distances. Adding more facilities can shorten those journeys, but each facility also carries a cost. The decision is therefore not simply to minimise travel or maximise the number of facilities, but to find a workable balance between them.

An exact optimisation model can calculate the mathematical optimum for smaller problems, but national waste systems can contain millions of spatial demand points. Camden Smith’s research therefore looked for an approach that could get very close to that optimum without requiring the computational effort of solving every national problem exactly.

The challenge was to combine accuracy with scale.

How was the model developed?

The model combines high-resolution WorldPop population information with a geospatial indexing system that divides the map into consistently referenced cells. It then uses population-weighted clustering to group those cells into collection areas and identify possible locations for Material Recovery Facilities, or MRFs — large facilities where collected waste is sorted before further processing.

The approach was developed and tested across countries with very different spatial characteristics, including densely populated, mountainous, island and geographically large settings. The purpose was not to create one map that fits every country, but a method that can adapt the collection areas to the population and geography supplied to it.

This turns collection-area design from a fixed assumption into a modelling decision.

What did the testing show?

The model was benchmarked against Mixed-Integer Linear Programming, an exact optimisation method that can identify the minimum-cost solution when a problem is small enough to solve exactly.

In those validation tests, the faster approach stayed within 2% of the absolute minimum cost while greatly reducing the computational burden. The research therefore produced a method capable of handling national-scale logistics problems without requiring an exact optimisation run for every possible configuration.

That distinction matters for a global simulator. A method can be mathematically sophisticated, but it is only useful in this setting if it can run across very large datasets and many different countries.

The result is a practical approximation that remains close to the exact benchmark.

How does the research add to the Waste Simulator?

The research was developed directly around KOLEKT’s need to replace its earlier static grid with dynamic collection areas. The findings will be directly incorporated into the deployment of the simulator's version 3.0.

For users, the significance is less about the clustering method itself and more about what it changes downstream. More realistic collection areas can improve estimates of how far waste travels, how many sorting facilities may be required and what the collection network could cost.

The research therefore strengthens the geographical foundation on which later waste-planning decisions depend.

What can the findings mean for other applications?

The immediate value of the research is practical: KOLEKT can replace fixed collection grids with areas that better reflect where people live and where sorting facilities could be located. Because the method can handle large national datasets in a reasonable amount of time, it gives the Waste Simulator a more realistic basis for estimating transport distances, infrastructure needs and collection costs.

The model can also be refined as better data become available. Future versions could use actual road-network distances, account for different facility capacities and respect administrative boundaries or other geographic constraints. These additions would bring the model closer to the conditions planners face when designing a real collection network.

For governments and municipalities, the contribution is therefore a stronger starting point for deciding how a national waste collection network could be organised before committing to costly infrastructure choices.