Research / Lab journal
Canopy Monitor: Exploring Romania’s Forests Through Time
Explore Romania’s forests through time with Canopy Monitor, an agent-led research project and interactive map.
Canopy Monitor: Exploring Romania’s Forests Through Time
What can satellite maps tell us about Romania’s forests—and how can we make the evidence easier to explore?
That question led to Canopy Monitor, an Innolabs experiment in agent-led research. Under human direction, AI agents investigated forest datasets, developed and tested analyses, and helped turn the results into a public research interface.
The result is Forest Time Machine: an interactive way to explore annual forest maps in the Făgăraș and Retezat mountains.
From a research question to an explorer
The project began with Google’s Forest Typology, or ForTy, datasets. These maps offer a way to study different forest types, going beyond a simple distinction between tree cover and open land.
The underlying ForTy model and datasets were developed by Maxim Neumann and colleagues at Google DeepMind, Google Research, the World Resources Institute, and the International Institute for Applied Systems Analysis. Innolabs used those published datasets; we did not train the forest model.
Our agents investigated how the data could support research in Romania. They worked through documentation, prepared data, wrote analysis code, ran experiments, and checked whether results could be reproduced. People directed the work, set its scope, and requested the later public explorer and forest-balance study.
A closer look at change
Forest Time Machine lets visitors move through nine annual maps, examine areas identified for further investigation, and compare findings with additional disturbance evidence. It also provides methods and interpretation notes alongside the maps.
A separate forest-balance study looks at forest and canopy change across the two mountain massifs. It presents evidence from different sources separately, helping readers understand what each dataset measures and why their results may differ.
Evidence with its limits
Satellite-derived maps can reveal patterns worth investigating, but a model’s classification is not a field observation. The project did not perform field verification or establish the causes or legal status of mapped changes.
That care shaped the research process. The team recorded data sources and checksums, tested calculations, and compared repeated runs. One conservation screen returned no qualifying patches under its chosen criteria—a bounded result that does not mean unrecorded primary forest is absent from Romania.
Canopy Monitor shows how a research question can become an inspectable chain of methods, results, and public tools.
Explore Forest Time Machine or read the technical case study for the experiments, verification process, and research record.