关键词: 汉江中下游, 典型河段, 水质评价, 神经网络, 多光谱数据, 遥感反演模型

Abstract: Typical segments in the middle and lower reaches of Hanjiang River were taken as study areas for water quality. According to sampling results and synchronized multi-spectral CCD data of HJ-1A satellite in spring, summer and autumn of 2012, we establish a retrieval model of BP neural network for TN (total nitrogen) concentration ,and assess water quality of the study areas based on the retrieval results. The results show that, on the basis of resilient BP training algorithm (heuristic-based training algorithm), the retrieval model of BP neural network established is of high accuracy and wide application fields, which can truly reflect the changes in TN concentration in different reaches and different seasons , and is easy to utilize domestic satellite data to carry out assessment work of water quality ; furthermore, assessment results indicate that water quality of the research areas varies a lot with seasons and reaches the value of TN indicator in spring significantly exceeds standard value , in other words, value of this indicator in summer or autumn is lower than that in spring. Finally, concentration of TN of downstream area is lower than that of upstream area.

Key words: middle and lower reaches of Hanjiang River, typical segments of river, assessment of water quality, neural network, multi-spectral data, remote sensing inversion model, typical segments of river