Another major milestone has been reached in the project, with all the eight main demonstrations successfully concluded. (Work on the extra demonstrations in Austria still continues.) Most of the demonstration products were finalized by the end of last year as reported in the blog post. The rest of the products were finalized and delivered to the users during Q1 2023. All the products were validated using a combination of statistical approaches and user evaluation. The User Workshop held in early March allowed fruitful discussions and feedback from the user partners.
Altogether, 79 prototype map products were delivered to the users in the demonstrations, including forest structure variables, biomass and growth variables and change detection products. These maps covered the user partners interest areas, ranging from a few hundred hectares to full European coverage. The products were produced for two to three years in each of the demonstration sites, increasing the total number of maps delivered to the users to 164.
The prototype products were received with great interest by the users. The products were considered to have the potential to support large area forest carbon monitoring, but the current accuracy was not considered sufficient for stand level management decisions in most cases. Overestimation in low volume forest and underestimation in high volume forests were the main problems in high detail, affecting particularly voluminous forests in Central Europe. However, the demonstration provided several further development ideas for the project team, allowing improvement of the products in the future.
Relative RMSE for the forest structure variables produced with empirical methods using field reference data was typically 20-50% of the mean at plot level. Inter-year consistency of the estimates varied between demonstration areas depending on the stability of the remotely sensed imagery between years (Figure 1). In forest change maps, total clearances were well detected, while uncertainty of partial clearance strongly variable between and within demonstration sites.
Figure 1. Examples of reached accuracies: [a] scatter plot of estimates and validation field plots for height (in dm) in Romania (plot level RMSE 22,2% of mean; Bias -0,7%), [b] year-to-year consistency for growing stock volume in Galicia at stand level. The stands in the lower right section [b] have been cleared either totally or partially between 2019 and 2020.
Throughout the demonstration sites and output variables, biases were typically small (< 4% of the mean). This indicates potential usability of the approaches demonstrated for large area monitoring. For example, in the Finland demonstration, the bias for above ground biomass estimation was less than 2% for the total Finland for both of the demonstration years (Figure 2). The small biases and the stabile large area estimates (e.g. for Catalonia in Figure 2) indicate consistency of the results of large areas, with increasing variation in stand and plot level (Figure 1).
Figure 2. [a] Accuracy assessment statistics of the 2017 and 2019 above ground biomass (AGB) maps for Finland and [b] Average estimated forest structure values in 951 validation plots in Catalonia for 2020 and 2021.
A new ‘Demonstrations’-page has now been created. In this page you can find information on the locations and basic characteristics of the different types of demonstrations conducted. Prototype output products produced with openly available data will be made publicly available in the near future. Links to the product portal will open on the ‘Demonstrations’-page as soon as available. Examples of the prototype products can also be seen in the 'Platform'-page and on an earlier blog post about the prototype products.
Stay tuned for the blog post announcing the opening of the product portal!