The forests of Dohoku (northern Kamikawa, Soya, and inland Rumoi) — stands of Sakhalin fir, Sakhalin spruce, and Erman's birch — account for a significant share of Hokkaido's total forest carbon stock. As research using satellite remote sensing and deep learning to estimate carbon stocks advances both in Japan and abroad, we lay out a perspective that reads Dohoku's forests not merely as "candidate sites for carbon credits" but as an intersection of forestry, tourism, and climate adaptation.
In the context of carbon neutrality by 2050, the capacity of forests to absorb and store carbon is drawing international attention. Hokkaido’s forests make up roughly one quarter of Japan’s total forest area, and Dohoku (northern Kamikawa, Soya, and inland Rumoi) is a distinctive region within it, where relatively young stands and boreal tree species are intermixed. In this article, we revisit the current state of research on estimating forest carbon stocks through satellite remote sensing and deep learning, repositioning it within the context of Dohoku.
1. The Difficulty of Measuring “Forest Carbon”
A forest’s carbon stock is distributed across the trunk, branches and leaves, roots, and soil. The trunk alone can be estimated from diameter at breast height and tree height, but measuring trees one by one across a vast, sparsely populated region like Dohoku is not realistic. This is why methods that combine satellite imagery with ground-based plot surveys to estimate carbon stocks over wide areas have developed rapidly over the past decade.
2. Estimating Carbon Stocks with High-Resolution Satellite Imagery and Deep Learning
This paper proposes a method for learning a model that directly estimates forest carbon stocks from high-resolution satellite imagery, using style transfer to suppress domain shift across multiple regions. The fact that “a model trained in one region can also be used in another” carries important implications for a region like Dohoku, which has its own distinctive forest structure.
However, satellite imagery alone reveals only the upper canopy. To estimate the total carbon stock — including the forest floor, fallen trees, and soil organic matter — it must be combined with LiDAR and ground-based plot surveys.
3. Wide-Area Estimation with LiDAR and Statistical Models
This paper addresses a method for statistically conducting wide-area forest inventory and small area estimation using the sparse sampling of G-LiHT (NASA’s airborne LiDAR system) in Alaska. Dohoku lies in the same high-latitude boreal zone as Alaska, and it shares the geographic and ecological similarities that would allow such methods to be transplanted.
Within Japan, the Forestry Agency has acquired airborne LiDAR mainly for national forests, but much of the privately owned and municipally owned forest remains uncovered. There is room here for the private sector, local governments, and universities to collaborate in filling in the data.
4. The Forest as an Intersection of Forestry, Tourism, and Climate Adaptation in Dohoku
If forest carbon stock is grasped only as “an object for calculating carbon credits,” the discussion becomes one-dimensional. Dohoku’s forests are places where the following multiple functions overlap.
| Function | Stakeholder | Evaluation Axis |
|---|---|---|
| Timber production | Forestry cooperatives, forestry companies | Timber revenue, employment |
| Carbon absorption and storage | National and local governments, credit-purchasing companies | Absorption volume, credit unit price |
| Tourism and experiences | nbyn, activity operators | Number of visitors, experience quality |
| Biodiversity conservation | Ministry of the Environment, research institutions, NGOs | Number of species, population counts |
| Climate adaptation | Watershed municipalities, flood-control parties | Water-source recharge, sediment-disaster mitigation |
These are interlinked with one another. For example, excessive timber harvesting reduces short-term carbon stock while simultaneously eroding tourism value and biodiversity. On the other hand, appropriate thinning promotes the growth of a stand, raising its medium- to long-term carbon absorption capacity while also maintaining the biodiversity of the forest floor.
As satellite remote sensing makes it possible to estimate wide-area carbon stocks, the foundation for evaluating these multiple functions in an integrated way falls into place. Numbers will become increasingly important going forward as a supporting line for decision-making.
5. Implementation Steps for Dohoku (Tentative)
| Phase | Period | Content |
|---|---|---|
| 0 | 1–3 months | Integrate existing satellite data (Sentinel-2 / Landsat) with Forestry Agency data; baseline estimation |
| 1 | 3–12 months | Plot surveys in representative stands of Dohoku, plus accuracy validation of satellite estimates |
| 2 | 12–24 months | Prototype a watershed-level carbon-stock-trend dashboard for local governments and forestry companies |
| 3 | 24 months onward | Develop into a decision-making platform integrating credit conversion, tourism experience design, and forestry planning |
The technical hurdles have been coming down. The remaining challenge is how to translate the numbers, and to which decisions to connect them, amid stakeholders whose interests differ.
Conclusion
Dohoku’s forests occupy a distinctive position within Hokkaido’s overall forest carbon stock. Satellite remote sensing and deep learning have pushed the means of estimating wide-area carbon stocks at low cost into a practical stage. However, when numbers are handled in isolation, the discussion tends to drift toward “credit maximization” alone. As an intermediate layer that integrates the multiple functions of forestry, tourism, biodiversity, and climate adaptation, nbyn aims to play the role of translating the numbers within the region’s context.