Single tuned algorithm to estimate the SPM concentration in a cascade reservoir system using OLI/L8 images

Abstract

Suspended particulate matter (SPM) affecting light propagation in the water column, which, in turn, affects primary production, is considered an important water quality indicator. Monitoring SPM in cascading reservoir system is of a particular challenge due to the widely differing optical properties. The Tietê River Cascade System (TRCS) is considered a representative case that comprises different water types with varying optical properties. At 1150 km long, the Tietê River crosses the State of São Paulo from East to West across a variety of land uses and land covers, which makes SPM monitoring of this system extremely challenging via traditional methods. This research aims to investigate the relationship between heterogeneous SPM configuration in the TRCS and the remote sensing reflectance (Rrs) by identifying the most suitable empirical model to quantify a wide range of SPM concentrations. Empirical models based on single-band and band ratios were tuned which is then applied to OLI/Landsat-8 images to estimate the SPM concentrations. We tested three approaches to obtain the best fit to retrieve the SPM concentration: the first approach (i) the dataset from the first fieldwork of each reservoir was used do calibrate and the others fieldworks data were used do validate the algorithm; the second (ii) all data from a single reservoir were selected for calibration and the others for validation and the third (iii) approach we used methods for randomly divide the calibration and validation dataset. The results showed that only approach (iii) returned significant results. The best algorithm to estimate the SPM concentration was based on the band ratio B3/B2 (SPM = 10.34 × [561 nm/482 nm]-12.32; with an r = 0.65). This algorithm resulted in the lowest error on average (RMSE = 6.5 mg/L, nRMSE = 32.97% and MAPE = 47.05%). The highest errors in retrieve the SPM were observed for reservoirs with dominance of phytoplankton; in fact, considering the dataset from all reservoirs the correlation between SPM and Chl-a was 0.95, proving that there are influence from the Chl-a pigment on the signal which increases the error. Therefore, a method to reduce the influence of phytoplankton on the remote sensing reflectance should be developed in order to reduce the error in retrieve the SPM concentration.