KOLEKT, the University of Amsterdam and Analytics for a Better World collaborated to extend the optimisation behind the Waste Simulator. The resulting models allow the tool to explore ten-year waste-system scenarios across 199 countries, more than 1,000 cities and 14 material types, while working with imperfect data through more than 500 adjustable parameters. This work was developed through three Master’s research projects, each addressing a different decision across the waste chain: how collection areas should be designed, which collection systems are most cost-effective, and how processing infrastructure could be developed over time.
Why is planning a waste system difficult without a shared cost picture?
Governments and municipalities make interconnected decisions about waste collection, infrastructure and financing. To understand what a system will cost, they need information about how much waste is generated, where people live, how waste will be collected and sorted, which processing facilities are available, and what new infrastructure may be required.
Yet the public data available for making these decisions are often fragmented and incomplete. KOLEKT developed its Waste Simulator in response to this gap, creating a configurable starting point from which governments, municipalities and financiers can estimate waste-system costs and explore how much of those costs could be financed through Extended Producer Responsibility, or EPR — policies that make producers financially responsible for managing the products and packaging they put on the market.
The simulator now provides waste data for countries worldwide and 3,800 cities, estimates operating and infrastructure costs, calculates EPR revenue potential and compares the recovery costs of 14 materials. It draws on sources including the World Bank’s What a Waste 3.0, WorldPop, the European Space Agency and Eurostat.
The starting point is therefore a common view of the system and its costs.
Where does optimisation add to this picture?
A simulator can show what a waste system looks like under different assumptions. Optimisation goes a step further by systematically comparing possible choices and identifying which configurations perform best against a defined objective.
For waste planning, those choices occur at several stages. Planners need to determine how a country should be divided into collection areas and where sorting facilities should sit. They need to compare collection approaches, such as collecting waste directly by truck or asking households to bring materials to drop-off points. Once waste has been sorted, they also need to decide which processing technologies to invest in and where different materials should go.
KOLEKT therefore collaborated with the University of Amsterdam and Analytics for a Better World on three research projects supervised by the ABW Science-to-Impact Director, Prof. Dick den Hertog. Michael Camden-Smith and João Monteiro conducted their research through the MBA in AI, Data & Analytics, while Stefano Alting von Geusau carried out his project within the MSc in Data Science and Business Analytics.
Together, the projects examine the waste system as a sequence of connected decisions.
How did the three research projects approach different parts of the system?
Michael Camden-Smith started with geography. Earlier versions of the simulator divided countries into fixed 50-square-kilometre grids. His research developed an approach that instead uses high-resolution population data to create collection areas dynamically and determine suitable locations for large sorting facilities, known as Material Recovery Facilities or MRFs. When benchmarked against an exact optimisation method, the approach stayed within 2% of the minimum cost while remaining fast enough for national-scale problems.
Stefano Alting von Geusau examined how waste should be collected within those areas. His model compares collecting waste directly from households by truck with using shared drop-off points, where people bring waste to a central location for collection. It also tests a hybrid system that chooses the lower-cost option for each collection area. Across six countries, more than 355,000 spatial cells and 859 collection areas, the results showed that one collection method does not necessarily make sense everywhere: the lower-cost choice depends on local conditions such as population density, waste volumes and infrastructure costs.
João Monteiro focused on the next stage: what happens after waste reaches a material recovery facility and has been sorted. His research developed an optimisation prototype for deciding how different materials could be allocated between processing and landfill options, and which processing capacity could be added over a multi-year planning horizon. Instead of returning one universal answer, the model allows decision-makers to compare plans according to financial return, emissions, recovery and landfill use.
The three projects therefore move from where to collect, to how to collect, to what to invest in after sorting.
What did the research show?
One recurring finding is that national averages can hide important local differences.
Camden-Smith’s work addressed the first question: how should a country be divided into collection areas, and where should sorting facilities be located?
His method uses population and geography to create collection areas rather than applying the same fixed grid everywhere. When tested against a more computationally demanding exact optimisation method, it produced results within 2% of the minimum cost. This means national-scale planning can use a faster approach while still producing cost estimates that remain close to the mathematical benchmark, giving decision-makers a more realistic basis for planning collection networks and infrastructure.
Once those collection areas are defined, Alting von Geusau’s research asks how waste should be collected within them. His model compares direct truck collection with shared drop-off points and a hybrid approach that chooses the lower-cost option for each area.
In the six-country analysis, truck collection was cheaper when applied as a single system across all areas, while drop-off points became the lower-cost option in 70 of the 859 collection areas. The finding suggests that governments do not necessarily need to choose one collection model for an entire country. They can identify where a different approach may reduce costs and focus further planning and investment on those locations.
Monteiro’s work takes the next step, after waste has been collected and sorted. His model examines where different material streams should go and which processing capacity could be added over time.
When applied to Saudi Arabia, Brazil and the Netherlands, the same optimisation framework produced different plans because each country starts with different infrastructure, waste flows and constraints. It also showed that financial and environmental objectives do not always point to the same investment choices. For decision-makers, this makes the trade-offs visible before investment decisions are made, allowing them to compare how different priorities could shape the waste system over time.
Across all three projects, the same lesson emerges: effective waste planning depends on matching decisions to local conditions. Where collection areas are drawn, how waste is collected and which processing investments make sense can all vary by place; therefore, better decisions come from comparing options in context rather than applying one national solution everywhere.
What can this mean for waste planning next?
The Waste Simulator gives practitioners a common starting point for understanding waste flows, costs, infrastructure needs and potential EPR revenues over a ten-year planning horizon. KOLEKT launched the tool after more than a year of development, with the University of Amsterdam and ABW among its research partners.
The optimisation research adds different decision layers around that foundation. Camden-Smith’s work provides a scalable method for defining collection geography. Alting von Geusau’s work provides a framework for comparing collection systems and testing MRF assumptions. Monteiro’s prototype extends the decision process towards processing investments and material allocation.
The research also shows why these tools still depend on good local information. Optimisation can compare the consequences of different assumptions, but the usefulness of those comparisons depends on the quality of the data and the way local constraints are represented.
The broader opportunity is to connect waste data with increasingly specific planning decisions, from collection through to processing investment.
