The overlooked contribution of trees outside forests to tree cover and woody biomass across Europe

Siyu Liu, Martin Brandt, Thomas Nord‐Larsen, Jérôme Chave, Florian Reiner, Nico Lang, Xiaoye Tong, Philippe Ciais, Christian Igel, Sizhuo Li, Maurice Mugabowindekwe, Sassan Saatchi, Yuemin Yue, Zhengchao Chen, Rasmus Fensholt

Abstract

Trees are an integral part of almost all European landscapes, but only forest resources are systematically assessed by national inventories, and the extent to which trees in urban and agricultural areas contribute to biomass and carbon stocks at the national level remains largely unknown. Here we make use of nanosatellite imagery to generate canopy cover, height, and above-ground biomass maps for the entire European continent from 3-m resolution imagery. Our country-scale biomass estimates have a systematic bias of only 7.6% (overestimation; R = 0.98) when compared to national inventories of 30 countries, and our new dataset is sufficiently highly resolved spatially to support the inclusion of tree biomass outside forests, which we quantify to 0.8 Pg. Although this represents only 2% of the total European tree biomass on average, large variations between countries are found (for example 10% for UK) and trees in urban areas contribute substantially to national carbon stocks (for example 8% for the Netherlands). The good agreement with national inventory data, the scalability, and the high level of detail across landscapes, including trees outside forests, make our approach attractive for operational implementation to support national carbon stock inventory schemes.

Introduction

Monitoring terrestrial carbon stocks is critical for climate mitigation strategies. However, current national and international forest assessments, often guided by the Food and Agriculture Organization (FAO) definition of a "forest" (minimum 0.5 hectares and 5m height), overlook millions of trees scattered across agricultural and urban landscapes. These "trees outside forests" (TOF)—including hedgerows, garden trees, and isolated patches—are vital for biodiversity and local climate regulation but have remained difficult to quantify at scale. Medium-resolution satellites like Landsat (30m) or Sentinel-2 (10m) often fail to resolve individual tree crowns, leading to significant gaps in our understanding of non-forest biomass.

Methods and Data

To address this "blind spot," researchers developed a deep-learning framework leveraging high-resolution nanosatellite imagery.

High-Resolution Nanosatellite Data

The study utilized 2019 PlanetScope mosaics, providing 3-meter multispectral imagery daily across the European continent. At this resolution, individual tree crowns and their shadows become visible, providing critical features for height estimation that are lost in coarser data.

Deep Learning Architecture

A U-Net architecture with an EfficientNet-B4 backbone was trained for two primary tasks:

  1. Segmentation: Binary classification of pixels as "tree" or "not tree."
  2. Regression: Estimating canopy height for every identified tree pixel.

The model was trained using 0.7 million km² of high-accuracy airborne LiDAR data from countries including Denmark, the Netherlands, and Spain.

Biomass Estimation

Canopy height and cover were converted into Above-Ground Biomass (AGB) using allometric relationships derived from over 11,000 plots in the Danish National Forest Inventory (NFI).

Figure 1: Continental-scale canopy height mapping at 3m resolution. (a) European map of canopy height. (b-d) Zoom-ins showing the transition from multispectral imagery to high-resolution canopy height products.

Key Results

The resulting maps provide an unprecedented view of Europe's woody resources.

Validation Against National Inventories

The model’s biomass estimates demonstrated exceptional fidelity, achieving a correlation of R = 0.98 with official national inventories across 30 European countries. The systematic bias was remarkably low at 7.6% (overestimation).

Quantifying Trees Outside Forests (TOF)

The study identified 0.8 petagrams (Pg) of biomass in TOF across Europe. While this accounts for roughly 2% of total European tree biomass on average, its importance varies significantly by region. In the United Kingdom, TOF represent nearly 10% of national tree biomass. In the Netherlands, urban trees alone contribute 8% of national carbon stocks. Of the identified TOF biomass, 44.6% is located in grasslands, 34% in urban areas, and 16.2% in croplands.

Sensor Comparison

The 3m PlanetScope-based approach significantly outperformed existing products. Comparisons with Sentinel-2 and Landsat-8 data showed that coarser resolutions lead to massive biases (exceeding 200% in some cases) when attempting to map sparse or isolated trees.

Figure 2: Comparison of canopy height and cover estimates across different sensors (LiDAR, PlanetScope, Sentinel, and Landsat). PlanetScope shows the closest alignment with LiDAR reference data.

Discussion and Limitations

The ability to use shadows as a proxy for height at 3m resolution is a key technical breakthrough for satellite-based forest monitoring. This approach is highly scalable and could support annual updates to national carbon inventories.

However, several challenges remain:

  • Geographic Sensitivity: The model performed less optimally in Mediterranean zones due to sparse vegetation and a lack of recent LiDAR training data.
  • Allometric Calibration: Current biomass conversions rely heavily on Danish equations; future work requires more region-specific field data to refine these estimates across diverse European climates.
  • Data Requirements: The reliance on massive airborne LiDAR datasets for training currently limits the method's applicability in regions where such data is unavailable, such as the Global South.

Despite these limitations, the study proves that high-resolution nanosatellites can effectively map the "hidden" carbon sink of trees outside forests, providing a more complete picture of terrestrial biomass.

Summary updated

Abstract

Abstract Trees are an integral part of almost all European landscapes, but only forest resources are systematically assessed by national inventories, and the extent to which trees in urban and agricultural areas contribute to biomass and carbon stocks at the national level remains largely unknown. Here we make use of nanosatellite imagery to generate canopy cover, height, and above-ground biomass maps for the entire European continent from 3-m resolution imagery. Our country-scale biomass estimates have a systematic bias of only 7.6% (overestimation; R = 0.98) when compared to national inventories of 30 countries, and our new dataset is sufficiently highly resolved spatially to support the inclusion of tree biomass outside forests, which we quantify to 0.8 Pg. Although this represents only 2% of the total European tree biomass on average, large variations between countries are found (for example 10% for UK) and trees in urban areas contribute substantially to national carbon stocks (for example 8% for the Netherlands). The good agreement with national inventory data, the scalability, and the high level of detail across landscapes, including trees outside forests, make our approach attractive for operational implementation to support national carbon stock inventory schemes.
BibTeX
@inproceedings{liu2023overlooked,
  author = {Siyu Liu and Martin Brandt and Thomas Nord‐Larsen and Jérôme Chave and Florian Reiner and Nico Lang and Xiaoye Tong and Philippe Ciais and Christian Igel and Sizhuo Li and Maurice Mugabowindekwe and Sassan Saatchi and Yuemin Yue and Zhengchao Chen and Rasmus Fensholt},
  title = {{The overlooked contribution of trees outside forests to tree cover and woody biomass across Europe}},
  booktitle = {Research},
  year = {2023},
  doi = {10.21203/rs.3.rs-2573442/v1},
}
Figure