Assessing the Benefit of the Sentinel-3/FLEX Tandem Mission Concept for Retrieval of Biophysical Parameters
Verrelst, Jochem; Rivera, Juan Pablo; Alonso, Luis; Moreno, Jose
University of Valencia, SPAIN

The Fluorescence Explorer (FLEX) mission concept was proposed to the European Space Agency (ESA) for consideration as a future Earth Explorer mission. FLEX' main objective is the measurement of vegetation chlorophyll fluorescence (Fs) from space and the exploitation of this signal to better understand the carbon cycle. FLuORescence Imaging Spectrometer (FLORIS) is the main instrument of the FLEX mission concept. The Earth Science Advisory Committee recommended the investigation of the FLEX concept to be flown as a tandem mission with Sentinel-3 (S3). S3's OLCI and SLSTR should ensure the continuity of the MERIS and AATSR measurements. These instruments are expected to provide an accurate characterization of key atmospheric and surface parameters to facilitate Fs retrieval for FLORIS.
Biophysical parameters such as leaf chlorophyll content (LCC) and leaf area index (LAI) play two important roles within the framework of Fs retrieval. On the one hand, they complement the information provided by Fs in order to determine the actual photosynthetic activity of vegetation, as Fs estimation on its own would not suffice. On the other hand, these parameters might be used during the Fs retrieval process to constrain the outcomes to realistic values according to them. Moreover, the synergy of both FLORIS and OLCI datasets may lead to improved estimations as compared to these sensors alone. This concept has been studied in two ESA projects; FLUSS and PARCS. FLUSS aimed to study atmospheric corrections for fluorescence signal using synthetic data, while the main objective of the ongoing PARCS project is to analyze the adequacy of the FLEX-Sentinel 3 tandem concept.

In FLUSS, top-of-atmosphere (TOA) radiance data were simulated across a wide range of vegetation, atmospheric and geometry parameters by coupling radiative transfer models at leaf (FluorMODleaf), canopy (FluorSAIL) and atmosphere (MOMO) level. A similar approach as the MERIS TOA_VEG vegetation retrieval algorithm was pursued to analyze the retrieval potential from TOA. For each retrievable parameter (i.e., LAI, LCC, soil type and the fluorescence parameter F_total) a regression model was trained using the simulated datasets and then its performance was evaluated. Two regression types were chosen: a conventional linear regressor and a more advanced nonlinear regressor. Kernel ridge regression (KRR) was selected as nonlinear regressor because of its fast processing and robust performance. In order to assess the robustness of the retrieval approaches, the same approach was repeated by adding instrumental and additional retrieval noise to the spectra prior to feeding them into the models. A generic sampling strategy was pursued; a uniform random subset of the complete dataset was used for training of a model and then another uniform random subset was used for validation. FLORIS was evaluated as well equipped for accurate retrieval of biophysical parameters. Its retrieval capacity potentially surpasses that of S3-OLCI, especially when noise is involved. More remarkably is that for all scenarios the synergy of FLORIS with OLCI yielded considerable better performances. Even when the dataset become very noisy (e.g. retrieval noise: 25%ó) the synergy concept still delivered excellent accuracies, with a relative RMSE of 13% for LAI, 18% for soil type and 19% for LCC. Also F_total was considerably better retrieved when making use of the combined datasets (relative RMSE around 40%) however alternative retrieval strategies will be needed to further improve accuracies.

While FLUSS yielded first encouraging results, it did not escape our attention that both simulated OLCI and FLORIS datasets were radiometrically and geometrically perfectly integrated, i.e. without pixel-onto-pixel distortions. In reality, however, due to different sensor specifications, delayed recording time and imprecise geolocation, perfect geometric and radiometric co-registration is unlikely to occur. Hence, the impact of imperfect integration on the performance of biophysical parameter retrieval was assessed in PARCS. Real data was used here, which came from an airborne CASI image that was acquired over a flat agricultural site at Barrax, Spain. LCC and LAI maps were first calculated and used as base maps. The CASI image was subsequently resampled to OLCI and FLORIS spectral settings, which were then merged pixel-onto-pixel and sorted along the spectral range to simulate their synergy. From this, halve of the spectra along with corresponding LCC and LAI values were used for training a KRR model, each for FLORIS and OLCI independently and then their synergy. The remaining halve was kept aside for analyzing the effects of imperfect geometric co-registration on parameter retrieval. Various degrees of subpixel misregistrations were analyzed, for both deformed FLORIS (with OLCI as geolocated reference) and deformed OLCI (with FLORIS as geolocated reference). Results confirmed the FLUSS study that under the conditions of a perfect geometric co-registration, the synergy of FLORIS with OLCI led to superior retrievals as compared to their independent counterparts. For both biophysical parameters FLORIS outperformed S3-OLCI. However, accuracies started to degrade when subpixel spectral deformation took place due to imperfect geometric co-registration. As a consequence of the applied regression model, deformation of FLORIS spectra degraded the retrievals more than deformation of OLCI due to its many bands that deviated from what has been presented during training.
The following step consisted in calculating the relative subpixel misregistration threshold where the synergy is no longer beneficial as opposed to FLORIS’ independent performance. Although for LCC the synergy performed superior until reaching a relative misregistration of 29% (87 m) as opposed to FLORIS’ performance, when considering only the most conservative thresholds, then about 8.5% or 25.5 m misregistration is allowed. Hence, once subpixel misregistration is above 25.5 m then it is recommended to use FLORIS data alone for biophysical parameter retrieval.

In conclusion, the S3/FLEX synergy concept has been proven to be beneficial, and therefore launching both missions in tandem is strongly encouraged. However, imperfect geometric or radiometric co-registrations are likely to occur and cause retrieval performance degradation. Solutions will have to be developed to account for these misregistrations, either through an accurate radiometric/geometric correction scheme or through a flexible retrieval scheme.