工矿业城市区域易受工业活动、 矿产开采影响, 使其水环境遭受不同程度的破坏, 以至于水体污染问题突出。 当前常规水质监测主要采用“以点代面”的工作方式进行野外采样及其室内化验分析, 然而环境复杂多变, 空间差异大, 导致调查点代表性受限, 整体精度不高, 效率低下, 更难以实现区域性动态监测。 以因矿兴市的矿业重镇湖北黄石大冶市为研究区, 同步开展无人机高光谱航飞、 地面光谱测量和水体样品采集测试, 分别获得具有49个波段的高光谱影像数据和光谱分辨率为1 nm的水体光谱曲线, 其中影像数据波谱范围为505~890 nm, 光谱分辨率为7.78 nm, 空间分辨率为30 cm。 对获取的高光谱影像和光谱数据剔除异常值、 光谱定标、 辐射校正等预处理后, 对比分析研究区内水体的不同光谱吸收、 反射及光谱曲线形态特征信息, 从而提取出高光谱影像和测量光谱的反射光谱曲线形态特征、 去包络线后光谱曲线形态特征、 三阶求导后光谱曲线形态特征和光谱四值编码共四大类25个光谱特征。 采用皮尔森相关系数分析样品水质参数与光谱特征之间的相关性, 以此筛选出存在显著相关的水质参数与光谱特征。 在此基础上, 采用逐步回归分析方法筛选出最大反射率及其波长位置、 对称度、 光谱编码Ⅲ、 三阶导最大及最小值等光谱特征作为模型变量, 构建出水质参数的多元线性反演模型, 并对模型进行 F检验和 t检验。 将检验后的反演模型推广运用于研究区内高光谱影像, 获得尾矿库、 河流、 湖泊等典型区域的水质参数反演结果, 从而实现“由点到面”水质参数信息的快速获取。 结果显示水质参数pH、 硬度(Ca²⁺+Mg²⁺)、 钾与氯离子比值(K⁺/Cl^-)、 镁与碱度比值[Mg²⁺/(HCO^3-+CO32-)]的反演精度较高, 其pH的判定系数 R²最小为0.669, 镁与碱度比值的判定系数 R²最大为0.895, 相对均方根误差均小于28%; 而总溶解固体(TDS)反演精度较低, 其判定系数 R²仅为0.463, 相对均方根误差达36.762%。 提出了一种基于光谱曲线形态特征的高光谱遥感水质参数定量反演模型方法, 实现了pH值、 硬度、 镁离子与碱度之比等水质参数的高光谱定量反演, 为区域水环境动态监测提供了新方法。

Industrial and mining cities are affected by activities of these industries that damage the water environment to various degrees and make water pollution a particular problem. At present, field sampling with a grid pattern and indoor laboratory analysis are mainly used for routine water quality monitoring. However, the environment is complex and variable, and spatial differences are considerable. Therefore, the survey sites render limited representativeness, low overall accuracy, and poor efficiency, making it difficult to realize dynamic monitoring regionally. In this study, we took Daye City under Huangshi City, Hubei Province, as the research area. Daye is an important mining city that thrives on mining. Unmanned aerial vehicle (UAV) hyperspectral imaging, ground spectral measurements, and water body sampling were carried out simultaneously. As a result, 49-band hyperspectral imaging data and water body spectra with a spectral resolution of 1 nm were obtained. The imaging data have a wavelength range of 505~890 nm, a spectral resolution of 7.78 nm, and a spatial resolution of 30 cm. After outlier removal, spectral calibration and radiometric correction were performed on the hyperspectral imaging data and spectral measurements and a comparative analysis was carried out between the spectral data of various water bodies located in the study area in terms of their absorption/reflectance spectra and the morphological features of their spectral curves. We subsequently extracted 25 spectral features from these hyperspectral images and measurement spectra, and these were classified under the following categories: morphological features of reflectance spectra, morphological features of continuum-removed spectra, morphological features of third-derivative spectra, and 4-value spectral encoding. The Pearson’s correlation coefficient was used to analyze the correlation between the water quality parameters and the spectral features of the water specimens and to select the water quality parameters and spectral features that were significantly correlated with each other. On this basis, a multivariate linear inversion model was constructed for the water quality parameters using the following model variables, which were selected via stepwise regression analysis: the maximum reflectance and its corresponding wavelength, symmetry, and spectral code Ⅲ, and the maximum and minimum third derivative values. F-tests and t-tests were then performed on this model. After the tests, our inversion model was used to obtain the water quality parameters of typical water bodies, such as tailings ponds, rivers, and lakes, from the hyperspectral imaging data from the study area. We have thus succeeded in achieving the rapid acquisition of water quality information in a “point-to-surface” manner. The results of this study indicate that our model has high inversion accuracies for water quality parameters such as pH, hardness (Ca²⁺+Mg²⁺), potassium-to-chloride ratio (K⁺/Cl^-), and magnesium-to-alkalinity ratio [Mg²⁺/(HCO^3-+CO32-)]. Between these parameters, pH has the lowest coefficient of determination ( R²) of 0.669, whereas the magnesium-to-alkalinity ratio has the highest R² of 0.895. The relative root mean square errors (RRMSE) were generally lower than 28%. In contrast, the inversion accuracy of our model for total dissolved solids (TDS) was relatively low, and its R² and RRMSE were 0.463 and 36.762%, respectively. This study proposes a hyperspectral remote-sensing quantitative inversion method of water quality parameters based on spectral curve patterns. The method achieves hyperspectral quantitative inversion of water quality parameters such as pH, hardness and magnesium-to-alkalinity ratio, and it provides a new technique for dynamic monitoring of the regional water environment.