Breadcrumb

Success Stories

Copernicus Sentinel-2 data for canopy height estimation

11 November 2021

Canopy height of forests is a fundamental structural and biophysical parameter, useful to a wide variety of environmental studies and applications, such as biodiversity studies; conservation planning; biomass/carbon sources estimation and monitoring forest degradation at a large scale.

Estimating canopy height over large areas is essential for sustainable ecosystem management.

There are various solutions for estimating canopy height. For instance, terrestrial manual inspection, airborne imagery, LiDAR, etc. These solutions vary in terms of cost effectiveness, intrinsic methodological limitations and output resolution upper bounds.

Therefore, selecting the most appropriate one is not trivial and depends on the use-case and the available resources. The following proposed methodology aims at providing a cost-effective, scalable solution towards pixel-based canopy height estimation, drawing inspiration from similar research efforts and using deep convolutional encoder-decoder architectures.

Partially funded and supported by the European Union's Horizon 2020 Innovation Action programme, within the framework of the e-shape project, the following study focuses on the Bohemian Forest.

It is on a mountainous terrain spreading over an area of approximately 942 km2, with an altitude ranging between 600 and 1453 m.

Sentinel-2 canopy images
Canopy Height Model of the Bohemian Forest

The Canopy Height Model (CHM) of this area (shown in Figure 1), was calculated with ground sample distance (GSD) of 10 m, using LiDAR measurements and was employed as the ground truth for training and evaluating the pixel-based canopy height estimation model (eCHM).

The dataset used for training the eCHM was derived from three Copernicus Sentinel-2 Level-2A products, representing bottom-of-atmosphere reflectance. These products were chosen to include the whole area of interest and satisfy the need for slight cloud coverage (<4%). Two variants of a deep learning architecture were tested (Alagialoglou et al., 2021).

The best object-based accuracy was achieved by the model trained with the Convolutional Encoder-Decoder with Skip-Connections architecture, which in almost all cases outperformed the state-of-the-art, both in terms of pixel and object-based evaluation. Pixel-based quantitative evaluation yielded a Mean Absolute Error (MAE) of 2.29m and a Root Mean Square Error (RMSE) of 3.15m, whereas object-based accuracy reached 91.4% and area-covering accuracy of 94.1%. In Figure 2, one can see a visualisation of the canopy height estimation results of four 48 x 48 pixels’ tiles, randomly selected from the test set.

Sentinel-2 canopy images
Tile examples of estimated canopy height

Estimation error analysis showed that such error is larger in higher altitude and steeper slopes, but does not seem to correlate with the aspect of the pixel. Lastly, it is worth noting that inference time of the model is relatively short. Efficient inference is crucial; especially in implementing the proposed solution as a web service.

Ioannis Manakos, Principal Researcher at the Centre for Research and Technology Hellas (CERTH), comments, “The methodology described is promising, since it seems that information contained in multi-spectral spaceborne imagery can indeed be accurately mapped in the height dimension. However, there is still room for improvement”.

His colleague, Leonidas Alagialoglou, PhD candidate at the Aristotle University of Thessaloniki (AUTH), emphasises that, “Future work may include investigating the generalisation ability of the model (how well it can be used on satellite imagery of other areas/time periods), using fusion techniques of heterogenous data sources to enhance the accuracy of the model, and estimating the uncertainty of the output (determining the confidence value of each pixel of the estimated height map)”.

About the Copernicus Sentinels

The Copernicus Sentinels are a fleet of dedicated EU-owned satellites, designed to deliver the wealth of data and imagery that are central to the European Union's Copernicus environmental programme.

The European Commission leads and coordinates this programme, to improve the management of the environment, safeguarding lives every day. ESA is in charge of the space component, responsible for developing the family of Copernicus Sentinel satellites on behalf of the European Union and ensuring the flow of data for the Copernicus services, while the operations of the Copernicus Sentinels have been entrusted to ESA and EUMETSAT.


Did you know that?

Earth observation data from the Copernicus Sentinel satellites are fed into the Copernicus Services. First launched in 2012 with the Land Monitoring and Emergency Management services, these services provide free and open support, in six different thematic areas.

The Copernicus Land Monitoring Service (CLMS) provides geographical information on land cover and its changes, land use, vegetation state, water cycle and Earth's surface energy variables to a broad range of users in Europe and across the World, in the field of environmental terrestrial applications.

It supports applications in a variety of domains such as spatial and urban planning, forest management, water management, agriculture and food security, nature conservation and restoration, rural development, ecosystem accounting and mitigation/adaptation to climate change.

Acknowledgements

This study has been partially funded and supported by the European Union's Horizon 2020 Innovation Action program under Grant Agreement No. 820852, e-shape (https://e-shape.eu/). LIDAR data was granted by the Cross-border cooperation programme Czech Republic-Bavaria Free State ETC goal 2014. The support of the “Data Pool Initiative for the Bohemian Forest Ecosystem” data-sharing initiative of the Bavarian Forest National Park is here also acknowledged. This article is based on Alagialoglou et al. (2021) and compiled by Rizos-Theodoros Chadoulis and Ioannis Manakos (Centre for Research and Technology Hellas - CERTH, GR).

References

&L. Alagialoglou, I. Manakos, M. Heurich, J. Červenka and A. Delopoulos, Canopy height estimation from spaceborne imagery using convolutional encoder-decoder, 27th International Conference on Multimedia Modeling, June 22-24 2021, Prague, Czech Republic, doi: 10.1007/978-3-030-67835-7_26.

Initial Floating Portlet

See announcements from the community

Sentiboard is an open digital platform facilitating sharing of information related to the users and usage of Copernicus Sentinel data
Go to sentiboard