Firstly, we re-tune an algorithm based on empirical orthogonal functions (EOF) for globally retrieving the chlorophyll a concentration (Chl-a) of phytoplankton functional types (PFTs) from multi-sensor merged ocean color (OC) products. The re-tuned algorithm, namely EOF-SST hybrid algorithm, is improved by: (i) using 30% more matchups between the updated global in situ pigment database and satellite remote sensing reflectance (Rrs) products, and (ii) including sea surface temperature (SST) as an additional input parameter. In addition to the Chl-a of the six PFTs (diatoms, haptophytes, dinoflagellates, green algae, prokaryotes and Prochlorococcus), the fractions of prokaryotes and Prochlorococcus Chl-a to total Chl-a (TChl-a), are also retrieved by the EOF-SST hybrid algorithm. Matchup data are further separated for low and high temperature regimes based on different PFT dependences on SST, to establish the SST-separated hybrid algorithms which further shows improved performance as compared to the EOF-SST hybrid algorithm. The per-pixel uncertainty of the retrieved TChl-a and PFT products is estimated by taking into account the uncertainties from both input data and model parameters through Monte Carlo simulations and analytical error propagation. The uncertainty assessment provided within this study sets the ground to extend the long-term continuous satellite observations of global PFT products by transferring the algorithm and its method to determine uncertainties to similar OC products until today. Satellite PFT uncertainty is also essential to evaluate and improve coupled ecosystem-ocean models which simulate PFTs, and furthermore can be used to directly improve these models via data assimilation.

The terrestrial energy balance represents a measure of the excess energy stored in the climate system. A possible measurement of this variable can be made at the top of the atmosphere by quantifying the imbalance between the incoming solar flux and the outgoing reflected and infrared flux. This is the objective of the UltraViolet and infrared Sensors at high Quantum efficiency on-board a small SATellite (UVSQ-SAT) mission, validating miniaturized technologies on-board a CubeSat with 1U standards (about 11 cm x 11 cm x 11 cm). This satellite was put into orbit in January 2021 by SpaceX’s Falcon 9 launcher and is totally functional. In order to measure the various fluxes with accuracy it is necessary to know precisely the orientation of the satellite at each time. Indeed, the knowledge of this orientation makes it possible to dissociate the various fluxes and to correct them from the angle to the considered source (Earth, Sun). To do so, two methods were implemented to retrieve the satellite’s attitude based on Sun and Nadir pointing along with inertial measurement unit (IMU) data. To ensure more accurate knowledge of the attitude determination in every configuration (daylight and eclipse), neural networks were implemented based on the available sensors. A multilayer perceptron was thus trained in order to find the orientation of the satellite. Based on the attitude retrieved the different flux were computed at each time from the sensors signals. We present here the development and the outcomes of the neural network applied to in-orbit data recovered from the UVSQ-SAT mission.