Remote Sensing and Machine Learning for Forest Inventory

Forest inventories are fundamental to sustainable management, carbon accounting and biodiversity conservation. Traditional field‐based surveys, while accurate at plot scale, are labour‐intensive and cannot feasibly cover vast or remote regions. Advances in remote sensing—including multispectral satellite imagery, airborne and terrestrial LiDAR, unmanned aerial vehicles and synthetic aperture radar—have revolutionised forest measurement by providing wall‐to‐wall data on canopy structure, species composition and biomass. Concurrently, machine learning algorithms such as random forest, support vector machines and deep neural networks offer powerful tools to extract complex patterns from high‐dimensional remote sensing datasets. Together, these approaches enable estimation of key metrics—stem volume, basal area, mean tree height and above‐ground biomass—at unprecedented spatial resolution and temporal frequency. This integration supports real‐time monitoring of forest dynamics, assessment of carbon sequestration potential and early detection of disturbances such as pest outbreaks or illegal logging. The global reach of satellite platforms and the scalability of cloud-based processing workflows herald a new era in precision forestry, underpinning evidence‐based policy and management across diverse biomes.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

In a 2023 study of the Helan Mountains in China, researchers developed a novel red-edge vegetation index derived from Sentinel-2 data and employed a random forest model to predict forest stock volume. The combined use of spectral bands and indices yielded high accuracy (R²≈0.93 in training and R²≈0.60 in independent tests), demonstrating the value of red-edge information for carbon stock assessments. A 2020 investigation in Hunan Province integrated in situ plot measurements with Sentinel-2 imagery within the Google Earth Engine environment, comparing random forest, support vector regression and linear regression. The random forest approach outperformed alternatives (training R²≈0.91, testing R²≈0.58), highlighting the importance of red-edge bands and tailored vegetation indices for regional biomass mapping. A seminal 2019 comparison of Sentinel-2 and Landsat 8 for boreal forest structure in Finland employed multilayer perceptron and regression tree models. Sentinel-2 data consistently outperformed Landsat 8 across stem volume, mean height, diameter and basal area estimates, underscoring the value of higher spectral resolution—especially red-edge and shortwave infrared bands—for forest inventory applications.

Remote Sensing and Machine Learning for Forest Inventory publication trend

The graph below shows the total number of articles in remote sensing and machine learning for forest inventory across all publications each year (not limited to Nature Index journals).

Technical terms

Multispectral imagery: Remote sensing data acquired in several discrete wavelength bands, enabling discrimination of vegetation properties.

Vegetation index: Mathematical combination of spectral bands (e.g., red and near-infrared) designed to highlight plant health and density.

Random forest: Ensemble machine learning method using multiple decision trees to improve predictive accuracy and control overfitting.

LiDAR (Light Detection and Ranging): Active remote sensing technique that measures distance by timing laser pulses, yielding detailed 3D structure of forests.

Synthetic aperture radar (SAR): Microwave radar system capable of penetrating cloud and canopy to capture structural information irrespective of illumination conditions.

References

1. A Novel Vegetation Index Approach Using Sentinel-2 Data and Random Forest Algorithm for Estimating Forest Stock Volume in the Helan Mountains, Ningxia, China. Remote Sensing (2023).
2. Comparison of Sentinel-2 and Landsat 8 imagery for forest variable prediction in boreal region. Remote Sensing of Environment (2019).
3. Estimating Forest Stock Volume in Hunan Province, China, by Integrating In Situ Plot Data, Sentinel-2 Images, and Linear and Machine Learning Regression Models. Remote Sensing (2020).