基于光谱平滑的浑浊水体叶绿素a浓度一阶微分估算模型

A First Derivative Estimation Model of Chlorophyll-a Concentration in Turbidity Water Based on Spectral Smoothing

一阶微分处理常用于浑浊水体叶绿素a浓度的遥感估算,直接的微分处理容易扩大噪声的影响,该文使用光谱平滑降低噪声,讨论平滑处理对一阶微分估算模型的影响。对2004年夏季和2011年春季太湖水面实测光谱数据进行核回归平滑处理并建立一阶微分模型,结果表明,平滑处理后的光谱数据建立的一阶微分模型稳健性更好。将2004年7月数据建立的模型用于8月数据,2011年3月建立的模型用于4月数据,平滑处理后模型验证的均方根误差分别降低了13.1mg/m3和3.9mg/m3。光谱平滑对一阶微分模型应用精度的改进作用最显著,优于波段比值和三波段模型。平滑处理后的一阶微分和三波段模型应用精度较高,其中一阶微分模型使用的波段信息少,较易用于浑浊水体的叶绿素a浓度估算。

First derivative analysis was usually used in remote sensing estimation of chlorophyll a concentration in turbid water. Direct derivative processing may amplify the impact of noise, thus this study used spectral smoothing to reduce the noise and discussed the influence of spectral smoothing on the first derivative estimation model. The in situ spectra above water surface in summer（2004） and spring（2011） of Taihu Lake were smoothed by the kernel regression smoothing method, and the first deriva- tive models were then built. Results show that the first derivative model built by the spectra after smoothing has better robustness. When the model variable built by data in July, 2004 was used in the data of August, 2004; and the model variable built by data in March, 2011 was used in the data of April, 2011, the RMSE in validation reduced 13.1 mg/m^3 and 3.9 mg/m^3, respec- tively after spectral smoothing. After comparing the validation accuracy improvement of first derivative, band ratio and three band model after spectral smoothing, it was found that spectral smoothing can most significantly improve the validation performance of first-order derivative model. The first derivative model and three-band model after spectral smoothing have highest validation accuracy. Since the first derivative model involves least band information, it can be used to estimate chlorophyll a con- centration in turbid water.