Annual canopy height mapping in Italy from Landsat and multi-source LiDAR for disturbance and recovery monitoring
Abstract
Accurate and continuous monitoring of forest canopy height is essential for assessing forest structure, biomass, disturbance, and recovery. However, airborne laser scanning (ALS) has incomplete national coverage, is updated infrequently, and is costly to acquire. Here, we present an annual 30 m canopy height dataset for Italy spanning 2004-2024, developed by integrating Landsat imagery with multi-source LiDAR observations in a time-series deep learning framework. Two U-Net models were trained independently using canopy height reference data from ALS and NASA's Global Ecosystem Dynamics Investigation (GEDI) spaceborne LiDAR mission. Annual canopy height predictions from the two models were then fused using Bayesian Model Averaging (BMA) to improve spatial consistency and temporal continuity. Validation against field measurements from the Italian National Forest Inventory (NFI) showed high predictive accuracy, with a mean absolute error of 3.98 m. Despite its coarser spatial resolution, our 30 m product outperformed canopy height products currently available in Italy, including a Europe-scale PlanetScope-based dataset at 3 m resolution and a global Sentinel-based canopy height product at 10 m resolution, while also providing a harmonized annual record over two decades rather than only one or a few years. In contrast, an existing Europe-scale Landsat-based product yielded a low R 2 in this mountainous region, highlighting the limited transferability of broad-scale models in complex terrain. To demonstrate the utility of the dataset for forest dynamics monitoring, we derived a canopy height change-based disturbance product and evaluated it against independent observed events, obtaining a mean precision of 0.64 for 2005-2016. We also quantified post-disturbance recovery by tracking canopy height regrowth in areas disturbed during 2004-2005 across major Italian biomes through 2024. These results show that combining Landsat time series, multi-source LiDAR, deep learning, and Bayesian fusion provides an effective framework for annual canopy height mapping and for monitoring forest disturbance and recovery. The final dataset is publicly
Related
- Physical disturbance expands the climatic envelope of annual plant dominance
- Different strategies, shared trajectories: biomass-allocation scaling during establishment and recovery from disturbance in perennial herbs
- Soil viral response to drought disturbance and ecosystem recovery in an artificial rainforest
- A composite disturbance index for monitoring Sudanian dryland forests: structural and compositional indicators diverge under anthropogenic pressure
- Mapping and dynamic monitoring of desertification on the Qinghai-Tibetan Plateau using surface Albedo and Solar-Induced Chlorophyll Fluorescence
BibTeX
@article{su2026annual,
title = {{Annual canopy height mapping in Italy from Landsat and multi-source LiDAR for disturbance and recovery monitoring}},
author = {Yang Su and Nikola Besic and Saverio Francini and Xianglin Zhang and Yidi Xu and Giovanni d'Amico and Gherardo Chirici and Martin Schwartz and Ibrahim Fayad and Sarah Brood and Agnes Pellissier-Tanon and Ke Yu and Haotian Chen and Haoruo Li and Songchao Chen and Alexandre d'Aspremont and Philippe Ciais},
journal = {Ecological Informatics},
year = {2026},
month = oct,
doi = {10.1016/j.ecoinf.2026.104036},
url = {https://hal.inrae.fr/hal-05799677v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Annual canopy height mapping in Italy from Landsat and multi-source LiDAR for disturbance and recovery monitoring
AU - Yang Su
AU - Nikola Besic
AU - Saverio Francini
AU - Xianglin Zhang
AU - Yidi Xu
AU - Giovanni d'Amico
AU - Gherardo Chirici
AU - Martin Schwartz
AU - Ibrahim Fayad
AU - Sarah Brood
AU - Agnes Pellissier-Tanon
AU - Ke Yu
AU - Haotian Chen
AU - Haoruo Li
AU - Songchao Chen
AU - Alexandre d'Aspremont
AU - Philippe Ciais
PY - 2026
DA - 2026/10/14
JO - Ecological Informatics
DO - 10.1016/j.ecoinf.2026.104036
UR - https://hal.inrae.fr/hal-05799677v1
AB - Accurate and continuous monitoring of forest canopy height is essential for assessing forest structure, biomass, disturbance, and recovery. However, airborne laser scanning (ALS) has incomplete national coverage, is updated infrequently, and is costly to acquire. Here, we present an annual 30 m canopy height dataset for Italy spanning 2004-2024, developed by integrating Landsat imagery with multi-source LiDAR observations in a time-series deep learning framework. Two U-Net models were trained independently using canopy height reference data from ALS and NASA's Global Ecosystem Dynamics Investigation (GEDI) spaceborne LiDAR mission. Annual canopy height predictions from the two models were then fused using Bayesian Model Averaging (BMA) to improve spatial consistency and temporal continuity. Validation against field measurements from the Italian National Forest Inventory (NFI) showed high predictive accuracy, with a mean absolute error of 3.98 m. Despite its coarser spatial resolution, our 30 m product outperformed canopy height products currently available in Italy, including a Europe-scale PlanetScope-based dataset at 3 m resolution and a global Sentinel-based canopy height product at 10 m resolution, while also providing a harmonized annual record over two decades rather than only one or a few years. In contrast, an existing Europe-scale Landsat-based product yielded a low R 2 in this mountainous region, highlighting the limited transferability of broad-scale models in complex terrain. To demonstrate the utility of the dataset for forest dynamics monitoring, we derived a canopy height change-based disturbance product and evaluated it against independent observed events, obtaining a mean precision of 0.64 for 2005-2016. We also quantified post-disturbance recovery by tracking canopy height regrowth in areas disturbed during 2004-2005 across major Italian biomes through 2024. These results show that combining Landsat time series, multi-source LiDAR, deep learning, and Bayesian fusion provides an effective framework for annual canopy height mapping and for monitoring forest disturbance and recovery. The final dataset is publicly
SN - 1574-9541
ER -