|Paper title||Forest Flux - Cloud service for mapping forest structural variables and carbon assimilation|
|Form of presentation||Poster|
Today, information to support carbon emission control and carbon assimilation by forests is of very variable quality. The information sources are versatile: different types of field data, aerial photography, laser scanning data, and satellite imagery. This information is used as input for calculation models using which it is decided whether forest is a carbon sink or an emission source. The results can be further used to value the carbon on the growing market of the voluntary carbon trade.
In future, it will be ever more important that forest owners, governments, academia, investors, and organizers of the voluntary carbon market and can base their decisions on as accurate, reliable and comparable information as possible and that the information is easily accessible.
In the VTT-led Horizon 2020 Innovation Action project Forest Flux a service was developed to offer reliable and comparable information on forest resources and forest carbon. Forest Flux cloud service on the Forestry Thematic Exploitation Platform F-TEP includes a seamless service chain from field observations and satellite imagery. It produces estimates of the present forest resources and carbon assimilation on a certain area and their future forecasts. The forecasts can be computed by applying different climate scenarios. It is to our knowledge the first of its kind globally.
The main satellite data source was Sentinel-2 of the Copernicus program. Additional data sources included very high-resolution optical imagery and airborne laser scanning (ALS) data. Ground reference data were provided by the users or were acquired from open sources.
The services were offered for nine users in Finland, Germany, Portugal, Romania, Paraguay, and Madagascar located in the boreal, temperate, and tropical vegetation zones. The user types included private and governmental large forest owners and managers, forest industries, associations of forest owners, and a development aid organization.
The users could select their desired map products from the portfolio of 51 alternatives. They included natural and color infrared image maps, forest cover map, nine traditional forest structural variables, site fertility type, three change map types, five forest fragmentation variables plus five variables for their changes, four biomass variables, nine carbon flux variables plus nine variables indicating their change, and eight variables to forecast the biomass and carbon assimilation. In addition, statistical information on the carbon balance of an organization was computed. Inputs for the organizational carbon balance were, in addition to the satellite image based carbon assimilation products, emissions from the silvicultural measures, harvest, and transportation, provided by the user.
The main method for satellite image analysis was the in-house probability software whose benefit is its adaptivity to different quality and amount of reference data because the models can be checked and modified manually (Häme et al., 2001, 2013). For the mapping of change, another in-house tool Autochange was used (Häme et al., 2020).
The process model PREBAS was used to compute and forecast the primary production variables. It used as inputs the outputs of the structural variable estimation and daily data on temperature and precipitation (Minunno et al., 2019; Tian et al., 2020). The model that was originally developed for boreal forest was parametrized for several other species that grew in the study sites. Comparison of the model predictions with flux tower measurements indicated voi very good match.
Software components for the Forest Flux services we developed for the F-TEP platform where they are applicable for operational services. The processing chain is largely automated. The main challenges were the variable quality, amount, and formats of the reference data as well as residual clouds in the pre-processed imagery, which led to manual work in the development of the models for the estimation of the structural variables. The uncertainty of the results was computed using a random sample from the reference data. However, in some cases, the reference data were not adequate for an independent set for uncertainty assessment and the results had to be assessed using the training data.
The relative root mean square error (RMSE) for the growing stock volume estimation varied between 29% and 67%. The error was always smaller for the other estimated structural variables stem basal area, mean height, and stem diameter than for volume. The bias was usually few percent with an exception at two sites in the same country where the overestimation was over 20% computed with a limited reference data. In Finland, the pure Sentinel-2 based estimation provided a relative error of 45%. By including the ALS data in the model, the RMSE dropped to 31%.
In total, about 1300 raster maps at ten-meter pixel size or vector outputs were computed in two phases. The user feedback was collected after both phases. In the short term, the most desired services concern forest change and the traditional structural variables and biomass. Carbon market is still poorly developed but this is expected to grow fastest within coming few years due to international regulations, and pressure from company shareholders and public.
The three-year Innovation Action project Forest Flux started in 2019 and it was completed in November 2021. The operational services can be started immediately after the completion of the project.
Project partners, in addition to VTT Technical Research Centre of Finland Ltd. were Unique Land Use GmbH (DE), Simosol Oy (FI), University of Helsinki (FI), Instituto Superior De Agronomia (PT), and The National Institute for Research and Development in Forestry (RO). The project was supported by the Horizon2020 Program of the EU, Grant Agreement #821860.
Häme, T. et al. (2001) ‘AVHRR-based forest proportion map of the Pan-European area’, Remote Sensing of Environment, 77(1), pp. 76–91. doi: 10.1016/S0034-4257(01)00195-X.
Häme, T. et al. (2013) ‘Improved mapping of tropical forests with optical and sar imagery, part i: Forest cover and accuracy assessment using multi-resolution data’, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6(1), pp. 74–91. doi: 10.1109/JSTARS.2013.2241019.
Häme, T. et al. (2020) ‘A Hierarchical Clustering Method for Land Cover Change Detection and Identification’, Remote Sensing. MDPI AG, 12(11), p. 1751. doi: 10.3390/rs12111751.
Minunno, F. et al. (2019) ‘Bayesian calibration of a carbon balance model PREBAS using data from permanent growth experiments and national forest inventory’, Forest Ecology and Management. Elsevier B.V., 440, pp. 208–257. doi: 10.1016/j.foreco.2019.02.041.
Tian, X. et al. (2020) ‘Extending the range of applicability of the semi‐empirical ecosystem flux model PRELES for varying forest types and climate’, Global Change Biology. Blackwell Publishing Ltd, 26(5), pp. 2923–2943. doi: 10.1111/gcb.14992.