Determination of optically inactive water quality variables using Landsat 8 data: A case study in Geshlagh reservoir affected by agricultural land use

Water chemical variables such as total nitrogen (TN) and total phosphorus (TP) are soluble and are optically inactive. Remote sensing (RS) technique is able to monitor the optical activity of water but it has limited capability to monitor the TN and TP concentrations. In order to determine the TN and TP concentrations of Geshlagh Reservoir (GR), RS was used. The other water quality variables that were utilized included chlorophyll-a, Secchi disk depth (SDD) and total suspended solids (TSS). These variables are optically active and have a high correlation coefficient with TN and TP concentrations. The five images of Landsat 8 Operational Land Imager (OLI) taken at different times were used. Artificial neural network (ANN) model and linear regression (LR) model were utilized to determine the relationship of OLI data and TP and TN concentrations in GR. Results showed that the TN and TP concentrations have a high correlation with chlorophyll-a concentration. The spectral reflectance of bands ratio (band3/band2) and the single band of bands 3 and 4 were obtained as the most suitable bands to determine the chlorophyll-a concentration in GR. Compared with LR model, the accuracy of the ANN model was higher in the testing phase of TP prediction with R = 0.81 and TN prediction with R = 0.93. It can be concluded that the Landsat 8 OLI data has a high potential to provide the data of solubility variables of the water quality for the management of reservoirs. The contribution of this paper is that the developed ANN model will improve the prediction accuracy of inactive water quality variables based on the RS data. [Display omitted] •Saptiotemporal patterns of TP-TN concentrations were mapped in GR.•Changes in TP-TN concentrations affected by different factors were elucidated.•Landsat 8 OLI-based, the ANN and LR estimation models of TP-TN concentrations were developed.•The ANN model results were more accurate compared with the those of LR model.