Spectral indices for estimating total dissolved solids in freshwater wetlands using semi-empirical models. A case study of Guartinaja and Momil wetlands

Water is one of the most important and abundant resources on earth; therefore, determining easier and more accurate methods of measuring water quality has become challenging in recent years. Images of the Sentinel-2 multispectral (MSI) sensor and field measurements for three seasons and 39 spectral indices were evaluated to identify those that can be used as patterns for the generation of semi-empirical bio-optical models to predict the total dissolved solids (TDS) in surface water of freshwater wetlands. The Guartinaja and Momil wetlands, located in the Wetland Complex of Bajo Sinu, Northern Colombia, were used as references in the field. The individual bands of the Sentinel 2 MSI sensor and the spectral indices derived from these bands were tested (and will be referred to in this paper as spectral indices). They were classified as follows: 10 individual spectral bands of the MSI sensor of the Sentinel 2 satellite; 11 indices reported by the literature to determine water, vegetation, or soils; 11 indices reported to estimate salinity; and 7 to estimate total dissolved solids and proposed in this research. The indices were evaluated based on three exclusive conditions: 1) the correlation between the TDS values measured in the field and indices spectral values >0.7 for at least two sampling seasons; 2) for the three seasons, the regression models derived from the indices have a determination coefficient >0.7; and 3) the spatial correlation matrix between the images derived from the models is >0.8 for at least two out of the three analysis seasons. We achieved important results: 1) seven indices (B3, Brightness-3, SI-3, SI-5, Ferdous-2020, TDS-1, and TDS5) met the three conditions proposed above and therefore, they were preliminarily defined as patterns for estimating TDS in the selected wetlands; 2) Additionally, this research provided two new efficient indices for calculating TDS through bio-optical models: TDS-1 and TDS-5. The validation of the regression models derived from these indices indicated a high accuracy of the prediction surfaces, expressed in values <10% of the normalized root-mean-square deviation. The results of this study contribute to the determination of semi-empirical bio-optical models for predicting optically inactive water quality parameters, which are generally predicted with empirical models whose use is spatiotemporally limited. Additionally, the research contributions that are carried out to improve the methods of evaluating water quality from remote sensing products become investigations that positively impact the evaluation of water resources, especially in geographical locations in which there is no with sufficient financial resources for on-site sampling.