Retrieval of sun-induced plant fluorescence in the O$_2$-A absorption band from DESIS imagery

Abstract

We provide the first method allowing to retrieve spaceborne SIF maps at 30 m ground resolution with a strong correlation ($r^2=0.6$) to high-quality airborne estimates of sun-induced fluorescence (SIF). SIF estimates can provide explanatory information for many tasks related to agricultural management and physiological studies. While SIF products from airborne platforms are accurate and spatially well resolved, the data acquisition of such products remains science-oriented and limited to temporally constrained campaigns. Spaceborne SIF products on the other hand are available globally with often sufficient revisit times. However, the spatial resolution of spaceborne SIF products is too small for agricultural applications. In view of ESA's upcoming FLEX mission we develop a method for SIF retrieval in the O$_2$-A band of hyperspectral DESIS imagery to provide first insights for spaceborne SIF retrieval at high spatial resolution. To this end, we train a simulation-based self-supervised network with a novel perturbation based regularizer and test performance improvements under additional supervised regularization of atmospheric variable prediction. In a validation study with corresponding HyPlant derived SIF estimates at 740 nm we find that our model reaches a mean absolute difference of 0.78 mW / nm / sr / m$^2$.

Figures (8)

• Figure 1: DESIS and HyPlant data. (a) RGB composite of a DESIS acquisition (13/06/2023 14:37 CEST) and, in red, extent of spatially and temporally overlapping HyPlant acquisitions (13/06/2023 14:11 - 14:38 CEST). (b) Top: Sample DESIS at-sensor radiance spectra, Bottom: sample HyPlant at-sensor radiance spectra. $\mathcal{W}_\mathrm{out}$ denotes the spectral emulator domain.
• Figure 2: Proposed network architecture. Data: gray blocks denote different data sources: L1B smile-corrected DESIS L1B at-sensor radiance, L2A reflectance and atmospheric variables provided in the DESIS L2A product, GEO geometrical variables from L1C metadata and L2A geolayer: RAA (relative azimuth angle), TA (tilt angle), SZA (sun zenith angle), $h_\mathrm{gnd}$ (digital elevation model). other: $u$ denotes trainable sensor state identifier and $x_1$ the across-track pixel position. Network: variables ($\rho_{740}$, $s$, $e$, $f_{740}$) predicted by $d_\mathrm{px}$ and ($\mathrm{AOT}_{550}$, H$_2$O) predicted by $d_\mathrm{patch}$ as well as ($\Delta\lambda$, $\Delta\sigma$) predicted by $q$ are passed to the simulation layer implemented as the emulator $E$patoFastMachineLearning2023patoPhysicsbasedMachineLearning2024.
• Figure 3: (a) Relative and absolute reconstruction error of best performing model configuration over all DESIS acquisitions. (b) Red/Pink: Spectrally explicit error distribution in the DESIS acquisition matching the OCO-3 validation data (\ref{['fig:data_overview']}), light colors denote the 25 - 75 percentiles. Blue/Green: Sample reconstruction (blue) of a single spectral DESIS observation (green) matching HyPlant (2023) data.
• Figure 4: Overview of an image excerpt of a DESIS acquisition matching HyPlant (2023) validation data (\ref{['fig:err_per_wvl']}). Left to right: RGB composite, NDVI derived from L2A, relative reconstruction error, fluorescence estimate $f_{740}$, initial fluorescence estimate $f^\mathrm{init}_{740}$.
• Figure 5: Conditional 2d-histogram of DESIS SIF estimates of the best performing run ($\gamma_\mathrm{c} = 5 \times 10^3$, $\gamma_\mathrm{AOT} = 100$) compared to HyPlant (2023), HyPlant (2020) and OCO-3 validation data sets \ref{['tab:data']}, red dashed lines denote the 10, 25, 75 and 90 percentiles.