This research explores the application of geospatial intelligence and machine learning to map and analyse routes used in illicit supply chains, specifically targeting the networks that transport illegal goods from South America to Europe. With drug trafficking routes becoming more adaptive and complex, particularly in the corridors connecting South America to Europe, traditional methods of monitoring and interception have proven insufficient. Enforcement strategies often rely on seizure data, which is delayed and offers an incomplete view of these evolving networks. To address this, the study proposes a novel methodology that combines satellite data with machine learning to identify and map elements of the supply chain, such as cultivation fields, clandestine laboratories, and transportation routes. This approach aims to create a scalable solution for tracking these illicit networks and supporting law enforcement.

Through the use of supervised machine learning models, potential cultivation fields and clandestine laboratories are detected in satellite imagery. The methodology includes sourcing data, configuring machine learning models based on literature, and evaluating model performance. The challenge in training models is the limited labelled data for illegal activities.

Once these nodes are identified, a connected graph is constructed that integrates both licit and illicit transport routes. This graph serves as the foundation for estimating smuggling routes. The process involves combining graphs of waterways and roads with airstrips, ports, potential cultivation fields and laboratories. To estimate paths between illicit sites and the main infrastructure network, an algorithm is applied that estimates the least steep path based on a slope raster.

Route estimation is the next step, where Dijkstra's algorithm is employed to calculate the most probable paths. Each edge is weighted based on travel time, cost, and risk, with factors adjusted to account for various modes of transport, including trucks, boats, planes, and donkeys, depending on terrain. There is also a factor added to account for transloading. These weighted paths offer a realistic picture of choices traffickers might make to minimize travel time, cost or risk.

To test the model, a case study is performed focusing on the Colombia-Venezuela border region, a region with high trafficking activity. The case study applies the graph-based methodology to identify plausible trafficking routes, connecting cultivation sites to ports. By calculating optimal paths, the study reveals which routes are likely to be used most frequently, providing insights that could potentially aid in targeted enforcement.

There are some limitations and areas for future improvement in this proof of concept, such as challenges with data quality and model accuracy, as well as potential refinements to the methodology to enhance real-world applicability. However, there is potential in combining machine learning with geospatial intelligence to create a tool for tracking and disrupting illicit supply chains. This research offers a proof of concept for enabling scalable monitoring of illegal activities, supporting efforts to disrupt global trafficking networks.