Inversion Models Of Water Quality Parameters In The Bosten Lake Based On Multi-source Remote Sensing Data

Bosten Lake is an important water source of the Bayinguoleng Mongol Autonomous Prefecture and also the entire Southern Xinjiang, Water quality is related to the development, prosperity and stability of Southern Xinjiang. The thesis sets up inversion models by the measured water quality data, the hyperspectral data and the quasi synchronous satellite remote sensing data, expecting to complement and replace the traditional way of water quality monitoring with the remote sensing, and consequently to achieve a wide range of water quality monitoring in the Boston Lake. The conclusions of the thesis are as follows.1) Water quality prediction model building based on the satellite remote sensing data.Based on the satellite remote sensing data, the thesis realizes an establishment of mineralization, suspended solids, total dissolved solids and pH models. For the first two kinds of water quality parameters, the thesis sets up models by three kinds of data, which are the single band data, normalized data and band combination data. Through analysis and comparison, band combination data, i.e B2 + B3, is optimal for mineralization inversion; band B7 of normalized data is the best for suspended solids. Two data forms, normalized data and band combination data, are used to build the model for the total dissolved solids. Results show that band combination(B3×B5) and its built exponentiation model is suited for the total dissolved solid inversion. As for pH, single band, indirect data and multiple linear regression models are employed to set up model. It shows that multiple linear regression model is the best way of pH inversion.2) Water quality prediction model building based on the measured hyperspectral data.For the mineralization, the thesis sets up an model by three kinds of data, the normalized data,band combination data and the first-order differential data. Results show that the first-order differential data, namely 869 nm, 571 nm, 708 nm and 736 nm these four bands constructed multiple linear regression model more suitable to inversion. For suspended solids, three kinds of data, i.e. the single band data, the normalized data and the first-order differential data, are used and the first-order differential data is confirmed to build the band combination, namely571 462R’ ?R’and constructed the quadratic model more suitable to inversion. For the total dissolved solids, the normalized data and the first-differential data, are used and the multiple linear regression model constructed by the first-order differential data, namely 363 nm, 426 nm, 428 nm and 567 nm four bands are more suitable to inversion. For pH, the single band data, normalized data and the first-order differential data are used and the multiple linear regression model constructed by the first-order differential data, namely 1064 nm, 1063 nm,1070nm and 1032 nm these four bands are more suitable to inversion.3) Model comparisons based on the two data sourceFor the mineralization and suspended solids, the established model based on remote sensing image is better than that based on the measured hyperspectral data, so choose the remote sensing image to invert these two water quality parameters. For the total dissolved solids, the established model based on the measured hyperspectral data is a little bit better than that based on remote sensing image data, so choose the hyperspectral data to invert this water quality parameter. For pH, the models based on these two data source are similar, but the remote sensing image data source is chosen to invert pH on account of its accessibility than the hyperspectral data.