USIFnet: A convolutional neural network for spatial downscaling of satellite-based SIF

Satellite-based chlorophyll fluorescence (SIF) offers substantial potential for applications in carbon cycle assessment and ecosystem monitoring. However, the utility of satellite-based SIF data is constrained by its coarse spatial resolution, spatial gaps, and significant noise interference, posing challenges for its use in fine-scale ecosystem studies. Building on previous downscaling studies, this research presents a convolutional neural network (CNN)-based model aimed at addressing issues in SIF data and overcoming the constraints of current models. We utilized this model to generate a 0.005° and 16-day resolution downscaled SIF product based on the 0.05° TROPOMI data across China, achieving validation metrics of R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.9216, RMSE = 0.0671 mW/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>/nm/sr, and a no intercept regression slope of 0.987. Furthermore, we compared the model’s performance with previous CNN model using original TROPOMI SIF and tower-based SIF measurements, confirming its effectiveness and improvements in accuracy, spatial detail, and reduction of artifacts. This work provides a promising tool for reconstructing high-resolution SIF data and demonstrates the potential of CNN-based methods in SIF downscaling applications.