Remote sensing of water quality based on HJ-1A HSI imagery with modified discrete binary particle swarm optimization-partial least squares (MDBPSO-PLS) in inland waters: A case in Weishan Lake

Remote sensing has been recognized as an effective tool to monitor water quality in inland waters. An adaptive model coupled with a modified discrete binary particle swarm optimization algorithm utilizing the catastrophe strategy and partial least squares (MDBPSO-PLS) was developed to retrieve the water quality indexes (chlorophyll a (chl-a), total suspended matter (TSM), and turbidity) in Weishan Lake. Based on the selective bands for water quality retrieval, the proposed MDBPSO algorithm was compared with original DBPSO and Genetic Algorithm (GA) in order to validate the feasibility and efficiency of the proposed algorithm. The comparison results indicated that MDBPSO with the catastrophe strategy could avoid the premature convergence phenomenon of DBPSO algorithm. The hyperspectral data from HJ-1A Hyperspectral Imager (HSI) selected by MDBPSO were utilized to establish the PLS model by correlating the spectral data with measured water quality parameters. Then the established PLS model was compared with the PLS model established with full spectral data and measured water quality parameters. MDBPSO-PLS showed the better performance in retrieving the above-mentioned three parameters in Weishan Lake. The comprehensive errors between measured and retrieved parameters in MDBPSO-PLS (chl-a (R2 = 0.97, CE = 3.16%), TSM (R2 = 0.97, CE = 5.84%), and turbidity (R2 = 0.97, CE = 7.40%)) were lower than those in PLS model (chl-a (R2 = 0.55, CE = 18.71%), TSM (R2 = 0.49, CE = 49.50%), and turbidity (R2 = 0.50, CE = 50.46%). The three water quality parameters were mapped by using MDBPSO-PLS. Our observations indicated that the MDBPSO-PLS model showed the significantly higher retrieval accuracy than the PLS model and could provide realistic information on the distribution of chl-a, TSM, and turbidity in Weishan Lake based on HJ-1A HSI.