A Chlorophyll-a Algorithm for Landsat-8 Based on Mixture Density Networks

Abstract

Retrieval of aquatic biogeochemical variables, such as the near-surface concentration of chlorophyll-a (Chla) in inland and coastal waters via remote observations, has long been regarded as a challenging task. This manuscript applies Mixture Density Networks (MDN) that use the visible spectral bands available by the Operational Land Imager (OLI) aboard Landsat-8 to estimate Chla. We utilize a database of co-located in situ radiometric and Chla measurements (N = 4,354), referred to as Type A data, to train and test an MDN model (MDNA). This algorithm’s performance, having been proven for other satellite missions, is further evaluated against other widely used machine learning models (e.g., support vector machines), as well as other domain-specific solutions (OC3), and shown to offer significant advancements in the field. Our performance assessment using a held-out test data set suggests that a 49% (median) accuracy with near-zero bias can be achieved via the MDNA model, offering improvements of 20 to 100% in retrievals with respect to other models. The sensitivity of the MDNA model and benchmarking methods to uncertainties from atmospheric correction (AC) methods, is further quantified through a semi-global matchup dataset (N = 3,337), referred to as Type B data. To tackle the increased uncertainties, alternative MDN models (MDNB) are developed through various features of the Type B data (e.g., Rayleigh-corrected reflectance spectra ρs). Using held-out data, along with spatial and temporal analyses, we demonstrate that these alternative models show promise in enhancing the retrieval accuracy adversely influenced by the AC process. Results lend support for the adoption of MDNB models for regional and potentially global processing of OLI imagery, until a more robust AC method is developed. Index Terms—Chlorophyll-a, coastal water, inland water, Landsat-8, machine learning, ocean color, aquatic remote sensing.

Impact of Spectral Resolution on Quantifying Cyanobacteria in Lakes and Reservoirs: A Machine-Learning Assessment

“…However, these factors are not considered in the development of band-ratio algorithms [2], which might contribute to the poor performance of these algorithms compared with the ML models in the current study. Previous studies in [45], [104], [121], and [122] show the value of ML models and clearly demonstrate their advantages over empirical algorithms with hard-coded coefficients. The ML models tend to be flexible and learn the nonlinear association between R rs and IOPs.…”

“…That said, the MdSA value of these models is still larger than (or around) the uncertainties in field-based measurements. These uncertainties arise from random and systematic errors and are propagated to the model predictions [122]. Even though the test dataset (N = 181) was independent of the data used in the training of each ML model, the data were still originating from the same study sites with similar optical characteristics as the training dataset which brings uncertainties in the generalizability of the conclusions of this study to other sites.…”

Water Mixing Conditions Influence Sentinel-2 Monitoring of Chlorophyll Content in Monomictic Lakes

“…That is to say, different combinations of IOPs of each water component can lead to an identical sum of IOPs and, thus, to similar reflectance values [67,68]. On the contrary, the MDN algorithm attempts to overcome this one-to-many issue by modeling a conditional probability distribution of the target variables given input R rs data rather than directly modeling the R rs -Chl-a relationship like standard NNs [27,48,69]. However, the results obtained in the present study show a comparable performance, indicating that this non-uniqueness problem is yet to be fully solved for the investigated lakes.…”

CHLNET: A novel hybrid 1D CNN-SVR algorithm for estimating ocean surface chlorophyll-a

“…MDN can implement one-to-many inversion, that is, the same input can output multiple different values. In recent years, the model has been frequently used for regression applications with multiple normal distribution patterns (Pahlevan et al, 2020;Smith et al, 2021). MDN is a combined model based on MLP and Gaussian Mixture Model (GMM).…”