Accurate deep-learning estimation of chlorophyll-a concentration from the spectral particulate beam-attenuation coefficient

Graban, Sebastian; Dall’Olmo, Giorgio; Goult, Stephen; Sauzède, Raphaëlle

doi:10.1364/oe.397863

Cited by 10 publications

(2 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most traditional and based on laboratory techniques are the spectrophotometric methods and highperformance liquid chromatography (HPLC; Marino 2017;Batista and Fonseca 2018;Garrido et al 2019). Although these methods provide very reliable quantifications, they are in vitro methods which demand benchtop protocols with several steps and substantial consumption of chemical reagents that can deteriorate algae at the time of extraction (Lorenzen 1967;Van Heukelem and Thomas 2001;Marino 2017;Graban et al 2020). In addition, analyses are timeconsuming, require a large sample volume and involve high logistic and analytical costs (Ferreira et al 2012;Kuha et al 2020), which can delay the availability of results and response actions for prevention and control.…”

Section: Introductionmentioning

confidence: 99%

Fluorescence in the estimation of chlorophyll-a in public water reservoirs in the Brazilian cerrado

Souza de Barros,

de Almeida,

Moraes Soares

et al. 2024

Mar. Freshw. Res.

View full text Add to dashboard Cite

Context The usual strategy for monitoring of eutrophication process is the use of traditional limnological methods, based on laboratory analysis. These procedures involve costly and time-consuming analyses, usually with in vitro methodologies, which can still have limitations in terms of sensitivity and reliability, if poorly managed. Phytoplankton pigments, such as chlorophyll-a (Chl-a), are highly fluorescent and can provide the environmental status of water bodies. Aims This study aims to analyse, compare and evaluate an estimation of Chl-a through fluorescence in public water sources in the Brazilian cerrado. Exploratory statistical analyses were conducted by using absolute fluorescence units (AFU) and relative fluorescence units (RFU) compared with traditional laboratory data (standard procedure for the determination of Chl-a by spectroscopic methods) to evaluate the significance of differences in estimating Chl-a concentration. Subsequently, empirical models, based on spectral band combinations, were generated to convert fluorescence measurement in Chl-a concentration, by linear regression. Key results The generated model found a strong correlation and coefficient of determination (r = 0.88; R2 = 0.78). The efficiency of the model was also confirmed by statistical indicators (RMSE = 1.27, MAPE = 26.72 and BIAS = −6.32). Conclusions We concluded that the estimate of Chl-a through RFU was better than through AFU. Implications Therefore, based on the results of this study, it is recommended that RFU be used to obtain more precise and accurate estimates of Chl-a concentration through empirical models based on linear regression.

show abstract

Section: Introductionmentioning

confidence: 99%

Fluorescence in the estimation of chlorophyll-a in public water reservoirs in the Brazilian cerrado

Souza de Barros,

de Almeida,

Moraes Soares

et al. 2024

Mar. Freshw. Res.

View full text Add to dashboard Cite

show abstract

“…In addition to traditional machine learning methods, deep learning has attracted significant attention for water quality monitoring and remote sensing. Few researchers have used deep learning methods to estimate Chla from Rrs, such as feedforward neural networks, (18,19) which are multilayered neural networks used as regression models and trained using the backpropagation algorithm. Convolutional neural networks (20) operate on an image level instead of a pixel level, similar to other deep learning techniques, and they consist of multiple filters that are also trained using the backpropagation algorithm.…”

Section: Introductionmentioning

confidence: 99%

1D Convolutional Neural Network-based Chlorophyll-a Retrieval Algorithm for Sentinel-2 MultiSpectral Instrument in Various Trophic States

Salah,

Higa,

Ishizaka

et al. 2023

Sensors and Materials

View full text Add to dashboard Cite

Despite extensive research on chlorophyll-a (Chla) concentration retrieval methods from remote sensing reflectance (Rrs, sr −1 ) data, there remains a need for more reliable Chla retrieval techniques. In this study, we introduce a deep learning approach based on a 1D convolutional neural network (1D CNN) architecture. In addition, we provide a new method of representing the Rrs as a sequential vector. The model architecture targets the Sentinel-2 MultiSpectral Instrument (MSI) sensor. The proposed model was trained and tested on simulated and in situ data collected from broad trophic states in Japan and Vietnam waters with Chla concentrations ranging from 0.02 to 148.26 mg/m 3 . The proposed model was evaluated against well-accepted state-of-the-art methods: ocean color three-band (OC3), ocean color index (OCI), two-band ratio, Blend, and a neural network model with a mixture density network. The evaluation shows that the proposed method outperforms other methods with a 7.48-38.02% reduction in root mean squared error (RMSE) and an 11.50-39.17% lower mean absolute error (MAE) than the other methods. The promising performance of the proposed model suggests that more attention should be paid to the domain of sequence modeling for Rrs and the architecture of 1D CNN.

show abstract

Analyzing the Role of Salinity in the Chlorophyll Prediction in the Ashtamudi Estuary, India

Raj,

Krishna Priya,

Hisana

et al. 2024

Springer Proceedings in Earth and Environmental Sciences

View full text Add to dashboard Cite

Accurate deep-learning estimation of chlorophyll-a concentration from the spectral particulate beam-attenuation coefficient

Cited by 10 publications

References 28 publications

Fluorescence in the estimation of chlorophyll-a in public water reservoirs in the Brazilian cerrado

Fluorescence in the estimation of chlorophyll-a in public water reservoirs in the Brazilian cerrado

1D Convolutional Neural Network-based Chlorophyll-a Retrieval Algorithm for Sentinel-2 MultiSpectral Instrument in Various Trophic States

Analyzing the Role of Salinity in the Chlorophyll Prediction in the Ashtamudi Estuary, India

Contact Info

Product

Resources

About