Multi-Algorithm Indices and Look-Up Table for Chlorophyll-a Retrieval in Highly Turbid Water Bodies Using Multispectral Data

Salem Salem,Hiroto Higa,Hyungjun Kim,Kazuo Oki,Komatsu Kazuhiro,Hiroshi Kobayashi,Taikan Oki

doi:10.3390/rs9060556

Abstract

Many approaches have been proposed for monitoring the eutrophication of Case 2 waters using remote sensing data. Semi-analytical algorithms and spectrum matching are two major approaches for chlorophyll-a (Chla) retrieval. Semi-analytical algorithms provide indices correlated with phytoplankton characteristics, (e.g., maximum and minimum absorption peaks). Algorithms’ indices are correlated with measured Chla through the regression process. The main drawback of the semi-analytical algorithms is that the derived relation is location and data limited. Spectrum matching and the look-up table approach rely on matching the measured reflectance with a large library of simulated references corresponding to wide ranges of water properties. The spectral matching approach taking hyperspectral measured reflectance as an input, leading to difficulties in incorporating data from multispectral satellites. Consequently, multi-algorithm indices and the look-up table (MAIN-LUT) technique is proposed to combine the merits of semi-analytical algorithms and look-up table, which can be applied to multispectral data. Eight combinations of four algorithms (i.e., 2-band, 3-band, maximum chlorophyll index, and normalized difference chlorophyll index) are investigated for the MAIN-LUT technique. In situ measurements and Medium Resolution Imaging Spectrometer (MERIS) sensor data are used to validate MAIN-LUT. In general, the MAIN-LUT provide a comparable retrieval accuracy with locally tuned algorithms. The most accurate of the locally tuned algorithms varied among datasets, revealing the limitation of these algorithms to be applied universally. In contrast, the MAIN-LUT provided relatively high retrieval accuracy for Tokyo Bay (R2 = 0.692, root mean square error (RMSE) = 21.4 mg m−3), Lake Kasumigaura (R2 = 0.866, RMSE = 11.3 mg m−3), and MERIS data over Lake Kasumigaura (R2 = 0.57, RMSE = 36.5 mg m−3). The simulated reflectance library of MAIN-LUT was generated based on inherent optical properties of Tokyo Bay; however, the MAIN-LUT also provided high retrieval accuracy for Lake Kasumigaura. MAIN-LUT could capture the spatial and temporal distribution of Chla concentration for Lake Kasumigaura.

Highlights

Ocean color satellites have provided synoptic observation for oceans, coastal areas and inland water bodies [1,2]
The current study proposed multi-algorithm indices and the look-up table technique
The MAIN-LUT technique combines the advantage of a look-up table that covers wide ranges of trophic status with the ability of the existing

Summary

Introduction

Ocean color satellites have provided synoptic observation for oceans, coastal areas and inland water bodies [1,2]. Several approaches were developed for Case 1 waters such as quasi-analytical algorithm (QAA) [13], the Garver–Siegel–Maritorena (GSM) algorithm [14], and the band ratio in the blue-green wavelengths as ocean color V4 (OC4E) algorithm [15]. Both QAA and GMS derive inherent optical properties of open oceans, while the difference between is that GMS requires the absorption spectral shape of water constituents, while QAA does not require these data [16]

Objectives

Results

Conclusion