Enhancing chlorophyll-a predictions using optimal machine learning models and field spectral reflectance
Mona Allam (National Water Research Center, Egypt, Aerospace Information Research Institute)
Lifu Zhang (Aerospace Information Research Institute)
Xuejian Sun (Aerospace Information Research Institute)
Ozgur Kisi (Ilia State University (ISU), Luebeck University of Applied Sciences, Korea University)
Nand Lal Kushwaha (ICAR-Indian Agricultural Research Institute, Punjab Agricultural University)
Liu Yongxin (Aerospace Information Research Institute)
Massimo Menenti (TU Delft - Optical and Laser Remote Sensing, Aerospace Information Research Institute)
Salim Heddam (University 20 Août 1955 Skikda)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This study evaluates the effectiveness of hyperspectral data to retrieve chlorophyll a (Chl-a) concentrations using various Machine Learning (ML) methods, specifically to determine whether spectral reflectance can provide accurate estimations of Chl-a. The study aims to address the gap in understanding how hyperspectral measurements correlate with Chl-a concentrations and to explore the potential for improving water quality assessment by accurately estimating Chl-a concentrations, which is essential for environmental monitoring, especially in aquatic ecosystems. The method proposed is evaluated using different Chl-a concentrations defined by the experiment design using Rhodamine B. The main reason for preparing pre-defined solutions of Chl-a is to verify the sensitivity of spectral measurements to Chl-a concentrations. In this paper, we aim to measure the pure signature of the Chl-a in which spectral reflectance of each Chl-a concentration is measured with 10 replicates by the spectrometer HS-1000WFL3. Six ML methods were investigated; (i) the multilayer perceptron artificial neural network (MLPNN), (ii) the support vector regression (SVR), (iii) the random forest regression (RFR), (iv) the Gaussian process regression (GPR), (v) Relevance Vector Machine (RVM) and (vi) Extreme Gradient Boosting (XGboost). 70 % of the data is used in training the models and 30 % of the data was used for their validation. We applied two bands 446 nm and 595 nm that are highly correlated with Chl-a. The models are evaluated using coefficient of determination (R2), Nash-Sutcliffe efficiency (NSE), root-mean-square error (RMSE), and mean absolute error (MAE). The results for the input variable, band 595 nm achieved the best predictive accuracy using the MLPNN method with R2, NSE, RMSE and MAE of approximately ≈0.859, ≈0.853, ≈26.722 and ≈19.05, respectively. The research also aims to lay the groundwork for future studies in water quality monitoring and management, using hyperspectral data and ML to improve our understanding of aquatic environments.