Variational Autoencoder Framework for Hyperspectral Retrievals (Hyper-VAE) of Phytoplankton Absorption and Chlorophyll a in Coastal Waters for NASA’s EMIT and PACE Missions
Jiadong Lou, Bingqing Liu, Yuanheng Xiong, Xiaodong Zhang, Xu Yuan · 2025
Read Variational Autoencoder Framework for Hyperspectral Retrievals (Hyper-VAE) of Phytoplankton Absorption and Chlorophyll a in Coastal Waters for NASA’s EMIT and PACE Missions by Jiadong Lou, Bingqing Liu, Yuanheng Xiong, Xiaodong Zhang, Xu Yuan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-13476.txt.
Introduction
Jiadong Lou, Bingqing Liu, Yuanheng Xiong, Xiaodong Zhang, and Xu Yuan use a generative model to read plankton levels in coastal water from satellite color, in a 2025 paper. Tiny floating plants tint the sea in subtle ways that sensitive space sensors can catch. Two NASA missions now gather such rainbow-rich data, but coastal water is optically messy, and one spectrum can match several plankton mixes.
Their framework is built on a variational autoencoder, a model suited to one-to-many puzzles. Tailored designs let it recover both the plankton's light absorption and its green pigment amount from reflected spectra. Against real observations, the models scored high precision with low bias, beating a leading rival approach above all on the richer data. The authors see a path to tracking coastal ecosystems from orbit.
Catalog record
| CREATOR | Jiadong Lou, Bingqing Liu, Yuanheng Xiong, Xiaodong Zhang, Xu Yuan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-13476
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Variational Autoencoder Framework for Hyperspectral Retrievals (Hyper-VAE) of Phytoplankton Absorption and Chlorophyll a in Coastal Waters for NASA’s EMIT and PACE Missions Jiadong Lou, Bingqing Liu, Yuanheng Xiong, Xiaodong Zhang, Xu Yuan Abstract Phytoplankton absorb and scatter light in unique ways, subtly altering the color of water-changes that are often minor for human eyes to detect but can be captured by sensitive ocean color instruments onboard satellites from space. Hyperspectral sensors, paired with advanced algorithms, are expected to significantly enhance the characterization of phytoplankton community composition (PCC), especially in coastal waters where ocean color remote sensing applications have historically encountered significant challenges. This study presents novel machine learning-based solutions for NASA’s hyperspectral missions including EMIT (7nm), and PACE (2.5 nm), tackling high-fidelity retrievals of phytoplankton absorption coefficient ( a p h y a_{phy} ) and chlorophyll a (Chl-a) from their hyperspectral remote sensing reflectance ( R r s R_{rs} ). Given that a single R r s R_{rs} spectrum may correspond to varied combinations of inherent optical properties (IOPs) and associated concentrations (e.g., Chl-a), the Variational Autoencoder (VAE) is used as a backbone in this study to handle such multi-distribution prediction problems. We first time tailor the VAE model with innovative designs to achieve hyperspectral retrievals of a p h y a_{phy} and of Chl-a from hyperspectral R r s R_{rs} in optically complex estuarine-coastal waters. Validation with extensive experimental observation demonstrates superior performance of the VAE models with high precision and low bias. The in-depth analysis of VAE’s advanced model structures and learning designs highlights the improvement and advantages of VAE-based solutions over the mixture density network (MDN) approach, particularly on high-dimensional data, such as PACE. Our study provides strong evidence that current EMIT and PACE hyperspectral data as well as the upcoming Surface Biology Geology (SBG) mission will open new pathways toward a better understanding of phytoplankton community dynamics in aquatic ecosystems when integrated with AI technologies. I Introduction Phytoplankton is an extremely diverse set of microorganisms, varying in cell morphologies, biogeochemical functions, and physiological responses to environmental disturbances [ 21 ] . As the primary producer of ocean’s food web, phytoplankton produces approximately 50 percent of Earth’s oxygen, regulate the global carbon cycle and climate, and support various ecosystem services, such as fisheries, water quality, and biodiversity [ 7 ] . Therefore, knowledge of phytoplankton biomass and their community composition is critical to understanding the food web structure, higher trophic level production (e.g., fisheries), and biological shifts among other complex Earth Science questions, especially in the context of degraded water quality (e.g., eutrophication) and climate change (e.g., warming temperatures), demanding attention at local, regional, and global scales [ 9 ] . As such, there is an increasing interdisciplinary interest in studying phytoplankton community dynamics in estuarine-coastal waters, where massive riverine inputs of nutrient-rich freshwaters often lead to eutrophication, harmful algal blooms (HABs), and the annual recurrence of bottom-water hypoxia events, which cause widespread and severe impacts on the aquatic ecosystem [ 3 , 43 , 57 , 23 , 54 ] . In the field of ocean color remote sensing, the concentration of chlorophyll a (Chl-a) and phytoplankton absorption properties ( a_{phy} ) are two of the most commonly used metrics for assessing phytoplankton abundance and diversity in aquatic environments [ 41 , 56 ] . Those phytoplankton-related satellite algorithms are rooted in the physical principle that remote sensing reflectance ( R_{rs} , sr^{-1)} ), the ratio of water-leaving radiance to the total downwelling irradiance just above water, is determined by the inherent optical properties (IOPs), most importantly the total backscattering coefficient ( b_{total} , m^{-1} ) and the total absorption coefficient ( a_{total} , m^{-1} ) [ 38 ] . These components are interrelated through the radiative transfer theory, presented here in its simplified form [ 38 ] : R_{rs}({\lambda})=c\times(\frac{bb_{total}}{a_{total}+bb_{total}})\ , (1) where R_{rs}({\lambda}) represents remote sensing reflectance at wavelength \lambda ; c is the proportionality constant that accounts for factors such as the geometry of the observation and specific conditions of the water body. Further, a_{total} can be separated into absorption by four major constituents (Roesler and Perry, 1995) including water ( a_{w} ), colored dissolved organic matter ( a_{CDOM} ), non-algal particle ( a_{NAP} ) and phytoplankton ( a_{phy} ), i.e., a_{total}=a_{w}+a_{CDOM}+a_{NAP}+a_{phy}\ . (2) Recognizing this physical principle, Chl-a, a key indicator of phytoplankton biomass, has been widely estimated from multi-spectral satellite data using semi-analytical approaches that firstly infer a_{phy} peaks within the red portions of the spectrum from R_{rs} and then estimate Chl-a [ 18 , 29 , 34 , 50 ] . Alternatively, empirical methods that utilize band-ratio metrics, have a long history of being employed to estimate Chl-a directly from R_{rs} [ 35 , 24 , 49 ] . Band ratios, particularly those in the red and near-infrared (NIR) regions [ 17 ] , as well as fluorescence line-height (FLH) approaches that utilize three bands in either red or red-NIR bands to compute characteristic spectral reflectance peaks, are commonly used for Chl-a estimation in estuarine and coastal waters [ 20 , 31 ] . Additionally, recent phytoplankton-related algorithms have increasingly favored machine learning techniques, such as Mixture Density Networks (MDNs) [ 42 , 51 ] , to enhance the accuracy and efficiency based on large amounts of historical data, such as GLORIA [ 30 ] . Pahlevan et al. (2021) [ 41 ] not only applied MDNs—originally developed for multispectral sensors (e.g., Sentinel-3 OLCI) to enhance the estimation of Chl-a and, more critically, a_{phy} , but also adapted this approach for the Hyperspectral Imager for the Coastal Ocean (HICO), the first spaceborne hyperspectral imaging spectrometer designed specifically for sampling coastal oceans. This adaptation paves new possibilities for ocean color algorithms, especially in the context of the increasing availability of hyperspectral missions, such as PACE (Fig. 1). Hyperspectral sensors, with their narrow spectral bands from ultraviolet to NIR, capture far more spectral details than traditional multispectral sensors and, when paired with advanced retrieval algorithms, can extend beyond phytoplankton abundance (e.g., Chl-a) to quantify certain aspects of phytoplankton community composition (PCC). Hyperspectral a_{phy} is a crucial optical proxy for characterizing PCC as the variability in phytoplankton cellular pigment compositions significantly affects the magnitude and spectral shape of a_{phy} , which is further reflected in R_{rs} [ 5 , 11 , 25 , 33 , 60 , 59 ] . Semi-analytical models [ 29 , 46 ] , and machine learning approaches [ 41 , 2 , 56 ] have been popular methods for solving the inverse problem of estimating IOPs (e.g., a_{phy} ) from R_{rs} , most of which generally perform well in the open ocean and are well-suited to multispectral satellites. Studies focusing on the retrieval of a_{phy} in bio-geo-chemically and optically complex estuarine and coastal waters are comparatively sparse. This is because coastal waters, typically rich in colored dissolved organic matter (CDOM) and/or non-algal particles (NAP), can overwhelm phytoplankton optical signals [ 32 , 15 , 16 ] , which further leads to uncertainty regarding the degree to which PCC information (e.g., taxa, function, or size) can be accurately gleaned from satellite data in coastal waters and the levels of confidence. Last but not