Evaluating and Unmixing Spaceborne DESIS Solar-Induced Fluorescence; A Comparative Analysis with Airborne HyPlant Observations

By Mohammed Hajaldaw

Abstract
Solar-Induced Fluorescence (SIF) serves as a promising proxy for photosynthesis, facilitating the monitoring of plant health and agricultural productivity. In recent decades, remote sensing has been used in research to monitor SIF at various scales, with satellite-level research primarily conducted using atmospheric chemistry satellites that have accidental SIF retrieval capabilities. However, the limited spatial resolution of these satellites poses challenges in evaluating and unmixing their SIF products. This study aims to improve large-scale agricultural monitoring by evaluating the relative accuracy of SIF retrievals generated from satellites with finer spatial resolution. Additionally, it introduces a novel framework that utilizes land cover fraction maps to unmix SIF signals, enabling the generation of vegetation-based SIF maps. The methodology involved performing a comparative analysis between SIF retrievals from spaceborne DESIS and airborne HyPlant sensors, using HyPlant data as a reference. Furthermore, a novel SIF unmixing framework was introduced by training a random forest model to predict fractional SIF values across different vegetation classes (Crops, Grass and Weeds, and Trees). The model performance was then evaluated by aggregating data to spatial resolutions of 30 m, 60 m, and 300 m. The results showed that the Spectral Fitting Method Neural Network (SFMNN) SIF retrieval from DESIS achieved moderate relative accuracy when compared with HyPlant Spectral Fitting Method (SFM) SIF (R2 = 0.63). It outperformed other DESIS retrievals across all classes, demonstrating its potential for large-scale agricultural monitoring. In addition, the proposed unmixing framework demonstrated strong performance in predicting per-class average SIF (R2 = 0.95, NRMSE = 4.22%), highlighting the effectiveness of using landcover fraction maps for SIF unmixing. This study advances large-scale vegetation monitoring by assessing the relative accuracy of DESIS SIF retrievals and introducing a novel SIF unmixing framework. Future studies could focus on evaluating and unmixing SIF products produced by the upcoming FLuorescence EXplorer (FLEX) satellite.