Training a GA-PLS Model for Chl-a Concentration Estimation over Inland Lake in Northeast China

摘要

Apparent Spectral properties (AOPs) of phytoplankton pigment concentration were analyzed in Shitoukoumen Reservoir, the Changchun city drinking water resource, in order to investigate the feasibility of the using remote sensing to monitor chlorophyll a (Chl-a) concentration in Northeast China. 225 samples were collected for laboratory Chl-a analysis during 12 field campaigns from 2006 to 2008. Correlation analysis between Chl-a and spectra and derivative spectral reflectance revealed that derivative at spectral band between 420 to 700nm is more sensitive to Chl-a concentration, however correlation coefficient on reflectance is relatively low as Chl-a is not the dominating optical active component in the reservoir. A combination of Genetic Algorithms and Partial Least Square (GA-PLS) model was established for Chl-a estimation in this study. A preprocessing procedure was conducted to selected the comparatively high correlated spectral variable by application of correlation analysis between Chl-a with each spectral band, reflectance derivative ranging from 350 to 1000 nm and all possible band ratios (101, 250 combinations all together). 100 sensitive spectral variables were selected for GA-PLS modeling with above mentioned.preprocessing procedure. A number of 2/3 samples were selected to train the GA-PLS model, and the rest was utilized to validate the performance of the model. It is found that the relationship between the most sensitive reflectance band, reflectance derivative and band ratio and Chl-a concentration agreed well with linear function with R-Square range from 0.45 to 0.78. However, the GA-PLS model for Chl-a estimation performs much better, with model validation R-Square of 0.81. As our results were derived from large number of ground truth samples, representing a spatio-temporal variation of pigment conditions, so the GA-PLS model has great potential for Chl-a estimation for inland water bodies with similar background.