Multi-way partial least squares modeling of water quality data

Area of Science:

• Environmental Chemistry
• Chemometrics
• Water Quality Monitoring

Background:

• River water quality assessment is crucial for environmental health.
• Polluted rivers pose significant ecological and health risks.
• Accurate prediction of water quality parameters like BOD is essential.

Purpose of the Study:

• To analyze a decade of river water quality data.
• To compare the predictive performance of different partial least squares (PLS) regression models.
• To assess the capabilities of N-PLS (tri-PLS and quadri-PLS) models for predicting biochemical oxygen demand (BOD).

Main Methods:

• Utilized a 10-year surface water quality dataset.
• Applied partial least squares (PLS) regression, including unfold-PLS and N-PLS (tri-PLS, quadri-PLS) models.
• Employed leave-one-out cross-validation for model calibration and assessment of predictive errors (RMSECV, RMSEP) and captured variance.

Main Results:

• N-PLS models (tri-PLS and quadri-PLS) demonstrated lower validation errors and higher variance capture compared to unfold-PLS.
• Both tri-PLS and quadri-PLS models achieved acceptable precision and accuracy in predicting BOD.
• Root mean squares errors for tri-PLS were 1.65 (calibration) and 2.17 (prediction); for quadri-PLS, they were 2.58 (calibration) and 1.09 (prediction).

Conclusions:

• N-PLS models show promise for predicting river water BOD.
• Further data arrangement may enhance prediction accuracy.
• Analysis of N-PLS model scores and loadings can reveal temporal trends in river water quality.