i-$\phi$-male: A novel hybrid machine learning phasor-based approach to retrieve a full-set of solar-induced fluorescence metrics and biophysical parameters.
Scodellaro R., Cesana I., D'Alfonso L., Bouzin M., Collini M., Miglietta F., Celesti M., Schuettemeyer D., Colombo R., Chirico G., Cogliati S., Sironi L.
Solar-induced Fluorescence (F) is a pivotal parameter to monitor vegetation health as it provides crucial information on photosynthetic processes. Actually, only the atmospheric O$_2$-bands of high-resolution apparent reflectance spectra ($R_$app$$) are exploited to retrieve F. We propose a new method, i-$\phi$-male, based on the phasor approach coupled to supervised machine learning (S-ML) techniques. The 650-800nm $R_app$ spectra are Discrete Fourier transformed on consecutive spectral windows, where the S-ML algorithm, trained with a radiative transfer (RT) model, estimates: F, Leaf Area Index, Chlorophyll Content, Absorbed Photosynthetic Active Radiation, F Quantum Yield and F at photosystem level. We validated i-$\phi$-male on RT simulations (error $<$ 3% $)$ and provided analysis of field measurements acquired for increasing spectrometer-canopy distances, up to 100m (where $O_2$ bands are affected by atmospheric oxygen absorption). The simultaneous retrieval of biophysical variables and F from spectra acquired at multiple distances from the canopy paves the way to new important perspectives related to the real time monitoring of vegetation stress level on high scales.