Methodology
The methodological backbone of SENTIFLEX lies in the fields of plant physiology, physically-based modeling, and satellite data processing and assimilation schemes - and is output driven. Activities are organized in six major tasks: two preparatory and experimental tasks, three data processing tasks and one application-oriented task, as detailed below:
Task 1: Processing FLEX-like & Sentinel-3 data
While awaiting actual FLEX data, an important step is preparing FLEX-like data for use in modeling and experimental studies. Various datasets are readily available, especially Sentinel-3 and synthetic datasets generated by the FLEX-E2E simulator. SIF retrieval algorithms have been recently developed within the FLEX scientific activities, e.g., spectral fitting methods. The next step involves developing retrieval strategies for key biophysical and atmospheric variables. With the ARTMO toolbox I have extensive experience in developing and running diverse retrieval methods which are readily applicable to Sentinel-3 data. New hybrid retrieval methods, i.e. coupling of RTMs with latest machine learning methods, will be evaluated based on precision, processing speed, portability and capacity to quantify uncertainties.
Task 2: Disentangling SIF estimates to obtain unbiased estimates of photosynthesis
Here, the challenging science of satellite-based photosynthesis mapping is fostered. This task initiates experimental and modeling studies to ascertain how SIF is related to canopy photosynthesis under various spatiotemporally heterogeneous canopy conditions. Available RTMs (e.g., SCOPE) and datasets form a solid basis for experimental, retrieval and validation studies. The retrieval of multiple photosynthesis products will be considered, i.e. instantaneous net/gross photosynthesis of the canopy. This task should lead to a refined theoretical basis on how to best derive unbiased estimates of photosynthesis from SIF and SENTINEL-3, especially in complex canopies.
Task 3: Data assimilation processing scheme.
This core task aims to move beyond ESA FLEX activities by consolidating the insights obtained into a new data-assimilation processing scheme. This will first be addressed based on ARTMO foundations, but eventually aims at stand-alone capability enabled as a plug-in for a processing chain. The promising concept of emulation, i.e., approximating RTMs with machine learning, will be made applicable to the retrieval toolboxes. It then becomes possible to replace tedious processing tasks (e.g., running or inversion of SCOPE) by emulators with a vast gain in processing speed while maintaining precision. The development of the toolboxes will not only make possible to analyze and compare various processing schemes, but also will be released to the broader community to serve other processing applications.
Task 4: Development of a FLEX-Sentinel-3 prototype photosynthesis monitoring facility
The above activities will be streamlined into a prototype processing scheme that will enable automated processing of Sentinel-3 and FLEX-like satellite input data into accurate photosynthesis products. The processing scheme will be verified and validated with Sentinel-3 and FLEX-like data coming from the FLEX-E2E (end-to-end) simulator and field campaigns. This task will demonstrate that the facility is able to run autonomously as soon as Level-2 FLEX data becomes available. The final objective here is to achieve a high level of automation in producing vegetation photosynthesis and productivity products.
Task 5: Integrating spatiotemporal prediction methods
This task involves integrating advanced data processing techniques to fill in missing data (e.g. due to clouds) to enable spatiotemporal prediction and temporal upscaling. This task will be conducted in close collaboration with the signal processing group. We will use RTM couplings with ensemble Kalman filter and with support of machine learning algorithms in a Bayesian framework such as Gaussian processes. Emulation techniques will be applied to accelerate processing. Eventually this activity should lead to a scaling toolbox that can be implemented into the processing chain. Also, efforts towards temporally upscaling to achieve daily GPP are planned.
Task 6: Development of scientific & practical monitoring applications
Here the focus will be on the R&D and dissemination of scientific and practical applications. Specifically, the prototype monitoring facility can become attractive in support of two higher-level processing systems: i.e. CCDAS, for carbon flux estimations; and MARS, the European crop yield forecasting system in support of food security assessment. The monitoring facility can also pave the way for new public \& commercial initiatives in the smart-farming business and food security initiatives.
