Assessment of Chlorophyll-a Algorithms Considering Different Trophic Statuses and Optimal Bands

Abstract

Numerous algorithms have been proposed to retrieve chlorophyll-a concentrations in Case 2 waters; however, the retrieval accuracy is far from satisfactory. In this research, seven algorithms are assessed with different band combinations of multispectral and hyperspectral bands using linear (LN), quadratic polynomial (QP) and power (PW) regression approaches, resulting in altogether 43 algorithmic combinations. These algorithms are evaluated by using simulated and measured datasets to understand the strengths and limitations of these algorithms. Two simulated datasets comprising 500,000 reflectance spectra each, both based on wide ranges of inherent optical properties (IOPs), are generated for the calibration and validation stages. Results reveal that the regression approach (i.e., LN, QP, and PW) has more influence on the simulated dataset than on the measured one. The algorithms that incorporated linear regression provide the highest retrieval accuracy for the simulated dataset. Results from simulated datasets reveal that the 3-band (3b) algorithm that incorporate 665-nm and 680-nm bands and band tuning selection approach outperformed other algorithms with root mean square error (RMSE) of 15.87 mg·m −3 , 16.25 mg·m −3 , and 19.05 mg·m −3 , respectively. The spatial distribution of the best performing algorithms, for various combinations of chlorophyll-a (Chla) and non-algal particles (NAP) concentrations, show that the 3b_tuning_QP and 3b_680_QP outperform other algorithms in terms of minimum RMSE frequency of 33.19% and 60.52%, respectively. However, the two algorithms failed to accurately retrieve Chla for many combinations of Chla and NAP, particularly for low Chla and NAP concentrations. In addition, the spatial distribution emphasizes that no single algorithm can provide outstanding accuracy for Chla retrieval and that multi-algorithms should be included to reduce the error. Comparing the results of the measured and simulated datasets reveal that the algorithms that incorporate the 665-nm band outperform other algorithms for measured dataset (RMSE = 36.84 mg·m −3 ), while algorithms that incorporate the band tuning approach provide the highest retrieval accuracy for the simulated dataset (RMSE = 25.05 mg·m −3 ).

Spectral Feature Selection Optimization for Water Quality Estimation

“…Table 4 shows the model performance comparisons for validation. This result agrees with the conclusions of previous studies [38,39] and indicates that the performance of three-band model is slightly better than that of two-band models and NDCI. In addition, the comparisons show that the RMSE of the best model is 6.37 mg·m −3 (Figure 4).…”

Excessive Fertilizer Usage Drives Agriculture Growth but Depletes Water Quality

“…While contamination and its variation in Indian rivers are well studied (Kumar et al, 2016), contamination of lakes are not explored much. Lakes form a major portion of freshwater sources around the world (see Figure 1) and are heavily used for various anthropogenic purposes (Salem et al, 2017). So, studying current levels of contamination along with its temporal variation in lakes is essential to better understand the changes occurring in lake water quality and the potential causes for this change.…”

Evaluation of MERIS Chlorophyll-a Retrieval Processors in a Complex Turbid Lake Kasumigaura over a 10-Year Mission

“…Salem et al [56] found that the differences in retrieval accuracies among linear, quadratic polynomial, and power regression approaches were relatively small for the measured dataset. Linear regression was therefore adopted during this research.…”