Periodic Reporting for period 4 - SENTIFLEX (Fluorescence-based photosynthesis estimates for vegetation productivity monitoring from space)
Berichtszeitraum: 2022-07-01 bis 2022-12-31
Agricultural production is under increasing pressure by global anthropogenic changes, including rising population, diversion of cereals to biofuels, increased protein demands and climatic extremes. Through a fleet of Earth observation (EO) satellites, Europe is dedicated to keep fingers on the pulse of its agricultural lands. Among the most innovative vegetation monitoring concepts involves FLEX (FLuorescence EXplorer), which wasselected as ESA's 8th Earth Explorer and is by design timely and ground-breaking: it is the first mission concept specifically dedicated to monitoring the `breathing' of terrestrial vegetation. FLEX will fly in tandem with Sentinel-3 (S3) and will globally measure Sun-Induced chlorophyll Fluorescence (SIF) spectral emission from terrestrial vegetation. Together with S3, these two new-generation European EO missions offer huge possibilities to increase our knowledge on the basic functioning of the Earth's vegetation, i.e. the photosynthetic process of plants resulting in carbon uptake. With the selection of FLEX, Europe reinforces a forefront role in fostering SIF research. However, ESA's commitment stops at L2c, i.e. delivering geometrically corrected SIF data. Hence, it is to the responsibility of the scientists to interpret and process these data into relevant products that gain insights into terrestrial vegetation dynamics. With this project named SENTIFLEX, we aim to unravel satellite-based SIF-photosynthesis relationships given heterogeneous spatiotemporal conditions on a per-pixel basis. Consolidated relationships will serve the development of a European vegetation productivity monitoring facility that eventually will be based on assimilation of S3 surface reflectance with FLEX SIF data.
Motivation:
The scientific FLEX mission will be the first hyperspectral mission specifically designed to capture the complete broadband SIF signal as well as complementary reflectance spectra in the green to NIR region (500-780 nm) at a spatial resolution of 300 m, which is suitable for discriminating individual plots and stands. Moreover, FLEX is planned to fly in tandem with S3, which enables parallel retrieval of relevant atmospheric and key vegetation properties. Given the availability of S3 data and already-acquired or simulated FLEX-like data from FLEX science studies, the aim of SENTIFLEX is to go beyond the current state of the art in satellite-based SIF-photosynthesis research and overcome current limitations. This will be accomplished by: (1) development of a theoretical SIF-photosynthesis relationship no longer exclusively relying on indirect spectral proxies (e.g. FAPAR, APAR) but instead incorporating quantifiable biophysical variables and associated uncertainties that allow uncoupling vegetation and atmospheric dynamics from the SIF signal on a per-pixel basis; and, (2) implementation of the consolidated theoretical basis into a prototype FLEX-S3 data processing chain for vegetation photosynthesis and productivity monitoring.
Motivation:
The scientific FLEX mission will be the first hyperspectral mission specifically designed to capture the complete broadband SIF signal as well as complementary reflectance spectra in the green to NIR region (500-780 nm) at a spatial resolution of 300 m, which is suitable for discriminating individual plots and stands. Moreover, FLEX is planned to fly in tandem with S3, which enables parallel retrieval of relevant atmospheric and key vegetation properties. Given the availability of S3 data and already-acquired or simulated FLEX-like data from FLEX science studies, the aim of SENTIFLEX is to go beyond the current state of the art in satellite-based SIF-photosynthesis research and overcome current limitations. This will be accomplished by: (1) development of a theoretical SIF-photosynthesis relationship no longer exclusively relying on indirect spectral proxies (e.g. FAPAR, APAR) but instead incorporating quantifiable biophysical variables and associated uncertainties that allow uncoupling vegetation and atmospheric dynamics from the SIF signal on a per-pixel basis; and, (2) implementation of the consolidated theoretical basis into a prototype FLEX-S3 data processing chain for vegetation photosynthesis and productivity monitoring.
Task 1: Data generation. Both simulated FLEX/S3 data with associated variables, as well real data from campaigns have been collected. Regarding simulated data, physical models (radiative transfer models: RTMs) have been brought together within the ARTMO software framework (https://artmotoolbox.com/(öffnet in neuem Fenster)). For simulation of SIF data, the SCOPE model has been extended. All RTMs can be used by the developed post-processing toolboxes: 1) global sensitivity analysis (GSA); 2) scene generation module (SGM); and retrieval toolboxes. Latest version of ARTMO has been made freely available to the public. As such, the ARTMO toolbox was used for: (1) generation of training data for development retrieval models, (2) generation of test synthetic scenes, and (3) emulation of SIF for fast production of SIF images.
Task 2: SIF & vegetation properties retrieval. Retrieval algorithms have been developed for both FLEX-FLORIS and S3 OLCI data, and their synergy. The following algorithms have been successfully validated: leaf area index (LAI), leaf chlorophyll content (LCC), fraction of absorbed photosynthetically active radiation (FAPAR) and fractional vegetation cover (FVC). Also a SIF retrieval toolbox is under development. It led to several publications. Currently, retrieval models have been implemented into Google Earth Engine (GEE) for S3 data processing at the European scale. First European maps have been generated.
Task 3: assimilation to GPP. A post-doc has been hired for this task and he was preparing a manuscript about the topic. Unfortunately, he left the team within one year. However, he continues to work on the manuscript in his free time. Alternatively, as outlined in the proposal, recently we have teamed up with the groups working with the dynamic vegetation models BETHY and CCDAS for achieving assimilation towards GPP.
Task 4: FLEX-S3 prototype vegetation productivity monitoring facility. This task is slowly getting shape. Although not explicitly stated in the proposal, based on positive experience, we decided to develop this prototype vegetation productivity monitoring facility into GEE. A PhD student is dedicated to this task. The running of vegetation properties retrieval is already in place. We currently work on gap filling of S2 and S3 time series, which will enable to calculate phenology indicators (see also next task). A final step is to develop a regression model to enable productivity estimation.
Task 5: Integrating spatiotemporal methods. A time series toolbox called DATimeS has been developed with conventional and latest machine learning algorithms. It enables gap-filling and phenology indicators calculation. Also a data fusion tool option has been added, e.g. for fusing FLEX and S3 data. This research line led to several publications. We are now working on integrating the time series processing models into GEE to enable operational and larger-scale processing. Eventually we aim to be prepared for vegetation productivity processing at the European scale.
Task 6: developments of applications. Within the ARTMO framework existing toolboxes have been improved and multiple new toolboxes have been developed: TOC2TOA, BRDFplot, LabelMe, DATimeS, FLUOR, ALG, MLCA. See also https://artmotoolbox.com/(öffnet in neuem Fenster).
Each of these tasks led to several papers and several more manuscripts are in preparation.
Task 2: SIF & vegetation properties retrieval. Retrieval algorithms have been developed for both FLEX-FLORIS and S3 OLCI data, and their synergy. The following algorithms have been successfully validated: leaf area index (LAI), leaf chlorophyll content (LCC), fraction of absorbed photosynthetically active radiation (FAPAR) and fractional vegetation cover (FVC). Also a SIF retrieval toolbox is under development. It led to several publications. Currently, retrieval models have been implemented into Google Earth Engine (GEE) for S3 data processing at the European scale. First European maps have been generated.
Task 3: assimilation to GPP. A post-doc has been hired for this task and he was preparing a manuscript about the topic. Unfortunately, he left the team within one year. However, he continues to work on the manuscript in his free time. Alternatively, as outlined in the proposal, recently we have teamed up with the groups working with the dynamic vegetation models BETHY and CCDAS for achieving assimilation towards GPP.
Task 4: FLEX-S3 prototype vegetation productivity monitoring facility. This task is slowly getting shape. Although not explicitly stated in the proposal, based on positive experience, we decided to develop this prototype vegetation productivity monitoring facility into GEE. A PhD student is dedicated to this task. The running of vegetation properties retrieval is already in place. We currently work on gap filling of S2 and S3 time series, which will enable to calculate phenology indicators (see also next task). A final step is to develop a regression model to enable productivity estimation.
Task 5: Integrating spatiotemporal methods. A time series toolbox called DATimeS has been developed with conventional and latest machine learning algorithms. It enables gap-filling and phenology indicators calculation. Also a data fusion tool option has been added, e.g. for fusing FLEX and S3 data. This research line led to several publications. We are now working on integrating the time series processing models into GEE to enable operational and larger-scale processing. Eventually we aim to be prepared for vegetation productivity processing at the European scale.
Task 6: developments of applications. Within the ARTMO framework existing toolboxes have been improved and multiple new toolboxes have been developed: TOC2TOA, BRDFplot, LabelMe, DATimeS, FLUOR, ALG, MLCA. See also https://artmotoolbox.com/(öffnet in neuem Fenster).
Each of these tasks led to several papers and several more manuscripts are in preparation.
1) Development of retrieval algorithms of vegetation properties for FLEX and Sentinel-3 OLCI. The added value of the sensor synergy has been demonstrated.
2) Development of a time series toolbox that includes latest machine learning methods. The toolbox further calculates phenological indicators. The toolbox has been made freely available to the interested user, and is now used by various groups all over the world.
3) Development of a remote sensing synergy toolbox where multiple time series data sources can be fused for an improved capability of filling up gaps.
4) Development of toolboxes within the ARTMO software framework that has been provided to the community. New or improved toolboxes include: retrieval of vegetation properties, scene generation, global sensitivity analysis, fluorescence retrieval, emulation.
5) Implementation of retrieval and gap-filling models into GEE. It is the first time ever that Gaussian processes regression has been implemented for enabling cloud-free time series processing of satellite data. With GEE we are now able to process S2 or S3 data at the European scale.
2) Development of a time series toolbox that includes latest machine learning methods. The toolbox further calculates phenological indicators. The toolbox has been made freely available to the interested user, and is now used by various groups all over the world.
3) Development of a remote sensing synergy toolbox where multiple time series data sources can be fused for an improved capability of filling up gaps.
4) Development of toolboxes within the ARTMO software framework that has been provided to the community. New or improved toolboxes include: retrieval of vegetation properties, scene generation, global sensitivity analysis, fluorescence retrieval, emulation.
5) Implementation of retrieval and gap-filling models into GEE. It is the first time ever that Gaussian processes regression has been implemented for enabling cloud-free time series processing of satellite data. With GEE we are now able to process S2 or S3 data at the European scale.