Authors: Francesco Minunno and Jukka Miettinen
EO data products are sensitive to variations caused by atmospheric or seasonal effects, sometimes significantly affecting forest variable predictions. When used repeatedly in the same area to monitor the development of forest resources these effects can cause inconsistencies in the time series. The Data Assimilation (DA) approach to be demonstrated in the Norway use case aims to mitigate these inconsistencies of forest variable values in the monitoring time series.
The DA approach takes advantage of the combination of the PREBAS process-based ecosystem model and EO based estimates to produce consistent times series of forest variable maps with improved prediction accuracy. The PREBAS model is initialized with the first satellite-based prediction (2017) and run to the next available EO dataset (2019). DA is then performed combining model predictions with EO based prediction producing new forest structural variable predictions that account for both sources of information. The DA based predictions for 2019 are used to initialize the model and the DA will be iteratively repeated for the 2021 and 2023 databases.
As the DA implementation of the Norway demonstration is about to start soon, the timing of the recently published 'Data Assimilation of forest status using Sentinel-2 data and a process-based model' article (Minunno et al. 2025) could not have been more perfect. The paper describes the DA framework to be used in the Norway demonstration combining model and satellite-based predictions to improve the reliability and robustness of forest monitoring. The journal article utilizes datasets from an earlier ESA Assesscarbon project, but the framework development was continued during the ongoing FCM project. The article presents two frameworks: one for forest structural variables and one for quantification of the site fertility class. As the findings of the paper indicate that site fertility class prediction does not significantly benefit from DA in the boreal zone in short simulations due to low forest growth rates, only the forest structural variable framework will be used in the Norway demonstration. This framework reduced the uncertainty of forest structural variables in the Assesscarbon study and limited the impact of biased predictions. As an example, the RMSE were reduced from 5.8 cm (s2019) to 4.5 cm (DA2019) for diameter and from 5.1 m (s2019) to 3.3 m (DA2019) for height.
In the Norway use case we can further improve the DA framework and find optimal implementation approaches for large scale operation application. The demonstration area provides excellent conditions for evaluating the DA approach because yearly NFI field plots are available for uncertainty assessment. We are very much looking forward to reporting the findings of the demonstration later in this blog. But in the meanwhile, you can read the journal article to better understand the scientific background of the demonstration.
Full citation of the article: Minunno, F., Miettinen, J., Tian, X., Häme, T., Holder, J., Koivu, K. and Mäkelä, A. (2025) Data assimilation of forest status using Sentinel-2 data and a process-based model. Agricultural and Forest Meteorology 363:110436. DOI: 10.1016/j.agrformet.2025.110436