Boreal forests (taiga) account for roughly 30 percent of the planet's total forest area. As Japan's southern limit of the boreal zone and a transition belt, Dohoku occupies a geographic position where change can be detected earliest. Taking research on mapping human disturbance from Sentinel-2 satellite imagery and the statistical treatment of ecosystem "memory" as a starting point, this article outlines a method for reading Dohoku's forests by combining satellites with on-the-ground observation.
Boreal forests (taiga) are a vast ecosystem that accounts for roughly 30 percent of the planet’s total forest area. While their main range lies across Russia, Canada, and northern Europe, Japan’s Dohoku region sits at their southern limit and transition belt, occupying a geographic position where vegetation change driven by climate change can be detected earliest. Taking recent research that reads forest change using satellite imagery and AI as a starting point, this article outlines a method for reading Dohoku’s forests by combining that research with on-the-ground observation.
1. Dohoku as the “Southern-Limit Testing Ground of the Boreal Forest”
Dohoku (northern Kamikawa, the Soya mountains, and inland Rumoi) is a boundary zone where boreal and cool-temperate tree species intermingle, including Sakhalin fir, Sakhalin spruce, Erman’s birch, and Mongolian oak. When climate change shifts this boundary northward and upward in elevation, the region becomes one of the first places where change on the scale of decades can be observed.
However, tracking wide-area forest change through ground surveys alone is not realistic. This is where forest monitoring that combines satellite imagery with machine learning, which has developed rapidly in recent years, comes in.
2. Automatically Mapping “Human Disturbance” with Satellite Imagery
Sentinel-2 (the European Space Agency’s Earth observation satellite) captures multispectral imagery at roughly 10 m resolution every five days across the entire globe. Combined with deep learning, it can automatically extract “human-made linear disturbances” such as roads, forest roads, and logging sites.
This study uses semantic segmentation (a method that classifies each pixel of an image) to verify an approach for automatically mapping linear disturbances in boreal forests from Sentinel-2 imagery. It points toward dramatically lowering the cost of surveying an entire country by hand, and is especially valuable in wide-area, sparsely populated regions like Dohoku.
That said, what satellite imagery sees is limited to what is “visible from above.” Information that satellites cannot capture, such as forest-floor vegetation, soil organisms, and bird nesting conditions, only becomes meaningful when combined with ground-based observation.
3. Treating Ecosystem “Memory” Statistically
A forest’s present form carries the influence of the past on the scale of decades and centuries. Past logging, wildfires, pests and diseases, and climatic fluctuations remain as “memory” in the current species composition and growth rates. This is called ecological memory.
This paper proposes EcoMem, an R package for statistically estimating ecosystem memory. In a place like Dohoku, where multiple layers of “the memory of the land” overlap in the current forest, such as postwar afforestation expansion, Meiji-era pioneering, and use within Ainu culture, this kind of statistical framework serves as a guideline for organizing observations.
4. A Three-Layer Model of Satellite × Ground × Data
When designing forest monitoring in Dohoku, combining the following three layers is realistic.
| Layer | Data | Strengths | Weaknesses |
|---|---|---|---|
| Satellite (Sentinel-2 / Landsat) | 10 m resolution / 5-day cycle | Wide-area / repeated observation / free | Forest floor and understory not visible |
| Airborne LiDAR | Sub-1 m / 3D | Tree height / canopy structure | High acquisition cost / hard to repeat |
| Ground survey | Plot surveys / camera traps | Species identification / individual observation | Narrow coverage / labor costs |
The natural-diversity data that nbyn is accumulating in Dohoku sits at the lowest layer of the three-layer model (ground survey). Matching it against satellite data creates a distinctive layer that translates “what change detected by satellite means on the ground.” This is difficult for other regions to imitate.
5. The Challenge: Connecting Open Data with On-the-Ground Knowledge
We have entered an era in which satellite data can be obtained for free from Copernicus and the USGS. Many machine learning models, too, are published as open source. The barrier is connecting these with local on-the-ground knowledge: which streams the sika deer gather at, which ridgelines you are more likely to encounter bears on, which stands still hold Sakhalin spruce.
In Dohoku, the knowledge of local residents, forestry cooperatives, municipalities, researchers, and activity operators has not yet been integrated into data. What nbyn wants to work on over the medium to long term is building a translation layer that stands between the satellites and on-the-ground knowledge.
Summary
Forest monitoring using satellite imagery and AI is becoming a powerful means of tracking wide-area change in Dohoku at low cost. However, what satellites see is a flat image from overhead, and the essential value of Dohoku’s forests, such as forest-floor biodiversity and the memory of the land that includes Ainu culture, becomes three-dimensional only when combined with ground-based observation and local knowledge. nbyn’s role is to serve as the intermediate layer responsible for this three-dimensionalization, moving back and forth between the satellites and the field.