Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Estimating Canopy Height at Scale

Jan Pauls, Max Zimmer, Una M. Kelly, Martin Schwartz, Sassan Saatchi, Philippe Ciais, Sebastian Pokutta, Martin Brandt, Fabian Gieseke

TL;DR

This work tackles the challenge of generating a high-resolution global canopy height map by leveraging multi-source satellite data (Sentinel-1/2) and GEDI ground-truth measurements. It introduces a shift-resilient loss to cope with global geolocation inaccuracies in GEDI labels and applies SRTM-based filtering to mitigate mountainous-label errors, enabling robust training. The approach uses a UNet with a ResNet50 backbone to produce per-pixel canopy heights at 10 m resolution and demonstrates superior quantitative and qualitative performance compared with existing global height maps, including improved edge detail and reduced height error across height ranges. The resulting 10 m global canopy height map and publicly available code enable enhanced forest biomass monitoring and carbon accounting at scale, supporting climate mitigation and policy-relevant ecological analyses.

Abstract

We propose a framework for global-scale canopy height estimation based on satellite data. Our model leverages advanced data preprocessing techniques, resorts to a novel loss function designed to counter geolocation inaccuracies inherent in the ground-truth height measurements, and employs data from the Shuttle Radar Topography Mission to effectively filter out erroneous labels in mountainous regions, enhancing the reliability of our predictions in those areas. A comparison between predictions and ground-truth labels yields an MAE / RMSE of 2.43 / 4.73 (meters) overall and 4.45 / 6.72 (meters) for trees taller than five meters, which depicts a substantial improvement compared to existing global-scale maps. The resulting height map as well as the underlying framework will facilitate and enhance ecological analyses at a global scale, including, but not limited to, large-scale forest and biomass monitoring.

Estimating Canopy Height at Scale

TL;DR

This work tackles the challenge of generating a high-resolution global canopy height map by leveraging multi-source satellite data (Sentinel-1/2) and GEDI ground-truth measurements. It introduces a shift-resilient loss to cope with global geolocation inaccuracies in GEDI labels and applies SRTM-based filtering to mitigate mountainous-label errors, enabling robust training. The approach uses a UNet with a ResNet50 backbone to produce per-pixel canopy heights at 10 m resolution and demonstrates superior quantitative and qualitative performance compared with existing global height maps, including improved edge detail and reduced height error across height ranges. The resulting 10 m global canopy height map and publicly available code enable enhanced forest biomass monitoring and carbon accounting at scale, supporting climate mitigation and policy-relevant ecological analyses.

Abstract

We propose a framework for global-scale canopy height estimation based on satellite data. Our model leverages advanced data preprocessing techniques, resorts to a novel loss function designed to counter geolocation inaccuracies inherent in the ground-truth height measurements, and employs data from the Shuttle Radar Topography Mission to effectively filter out erroneous labels in mountainous regions, enhancing the reliability of our predictions in those areas. A comparison between predictions and ground-truth labels yields an MAE / RMSE of 2.43 / 4.73 (meters) overall and 4.45 / 6.72 (meters) for trees taller than five meters, which depicts a substantial improvement compared to existing global-scale maps. The resulting height map as well as the underlying framework will facilitate and enhance ecological analyses at a global scale, including, but not limited to, large-scale forest and biomass monitoring.
Paper Structure (36 sections, 4 equations, 14 figures, 7 tables)

This paper contains 36 sections, 4 equations, 14 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Satellite Imagery
  4. Sparse Height Measurements
  5. Generating Height Maps
  6. Approach
  7. Data Collection and Preprocessing (A)
  8. Model Training (B)
  9. Model Architecture & Parameters
  10. Shift-Resilient Loss
  11. Global-Scale Inference (C)
  12. Source Code & Canopy Height Map
  13. Results
  14. Comparison of Height Maps
  15. Metric-Based Comparison
  16. ...and 21 more sections

Figures (14)

  • Figure 1: A global canopy height map at a $10$ m resolution.
  • Figure 2: An RGB image (based on multi-spectral image data collected by one of the Sentinel 2 satellites) along with GEDI height measurements (red/yellow dots) are shown for an area in France.
  • Figure 3: Visual comparison of a regional map schwartzFORMSForestMultiple2023 and two global maps langHighresolutionCanopyHeight2023potapovMappingGlobalForest2021 (heights from $0$ m to $35$ m; see \ref{['fig:global_map']} for the colormap).
  • Figure 4: Our approach to generate height maps at a global scale: (A) data collection and preprocessing, (B) model training, and (C) global-scale inference. Initially, image composites based on Sentinel-1 and Sentinel-2 imagery are constructed and corresponding GEDI measurements as well as SRTM data are collected. After training the model using hyperparameter grid search, it is applied in a distributed manner, with the canopy height estimates being subsequently reprojected and made available through streaming services.
  • Figure 5: The GEDI instrument records the signals returned from the ground within a $25$ m diameter. The height is then essentially computed based on the differences of the first and last signals. While satisfying height values are obtained for most areas, this often leads to inaccurate height measurements for slopes.
  • ...and 9 more figures