Spectral fingerprinting: sediment source discrimination and contribution modelling of artificial mixtures based on VNIR-SWIR spectral properties
Spectral fingerprinting: sediment source discrimination and contribution modelling of artificial mixtures based on VNIR-SWIR spectral properties
复制标题
DOI：
10.1007/s11368-014-0925-1
复制
发表时间：
2014-12-01
影响因子：
3.6
通讯作者：
Kaufmann, Hermann
中科院分区：
文献类型：
--
作者：
Brosinsky, Arlena;Foerster, Saskia;Kaufmann, Hermann
悬浮泥沙的来源知识是重要的，以提高我们的泥沙动力学的理解，从而支持可持续的流域管理。一种直接追踪沉积物来源的方法是指纹技术。它所依据的假设是，可以区分潜在的沉积物来源，并可以根据不同的特征（指纹）确定这些来源对沉积物的贡献。近年来的研究表明，土壤的可见-近红外（VNIR）和短波-红外（SWIR）反射特性可能是传统指纹特性的快速、廉价的替代品（例如地球化学或矿物磁性）。为了进一步探索VNIR-SWIR光谱数据用于沉积物追踪目的的适用性，在Isabena流域收集了源样品，位于西班牙比利牛斯山脉中部的一个445公里（2）的旱地集水区。从主要的潜在沉积物源类型中采集上层土壤样品，沿原位反射光谱进行沿着采集。将样品干燥并过筛，并产生已知比例的人工混合物用于算法验证。然后，在实验室中采集潜在源和人工混合物样品的光谱读数。颜色系数和基于物理的参数计算从原位和实验室测量的光谱。所有参数通过一系列的先决条件测试，随后被应用于判别函数分析的源判别和混合模型分析的源贡献评估，三个源类型（即荒地，森林/草地和其他源的集合，包括农田，灌木丛，未铺砌的道路和开放的斜坡）可以可靠地识别基于光谱参数。实验室测量的光谱指纹允许量化的源贡献的人工混合物，和引入源的异质性到混合模型降低某些源类型的准确性。无法区分的源类型的聚合并没有改善混合模型的结果。尽管提供了类似的歧视精度实验室源参数，在现场获得的源信息被认为是不够的贡献modelling.The实验室混合物实验提供了宝贵的见解的能力和局限性的光谱指纹属性。从这项研究中，我们得出结论，光谱特性的组合可用于混合模型分析的有限数量的源组，而更直接的原位测量源参数似乎不合适。然而，基于实验室参数的建模结果也需要谨慎解释，不应仅依赖于平均值的估计，还应考虑不确定区间。
Knowledge of the origin of suspended sediment is important for improving our understanding of sediment dynamics and thereupon support of sustainable watershed management. An direct approach to trace the origin of sediments is the fingerprinting technique. It is based on the assumption that potential sediment sources can be discriminated and that the contribution of these sources to the sediment can be determined on the basis of distinctive characteristics (fingerprints). Recent studies indicate that visible-near-infrared (VNIR) and shortwave-infrared (SWIR) reflectance characteristics of soil may be a rapid, inexpensive alternative to traditional fingerprint properties (e.g. geochemistry or mineral magnetism).To further explore the applicability of VNIR-SWIR spectral data for sediment tracing purposes, source samples were collected in the Isabena watershed, a 445 km(2) dryland catchment in the central Spanish Pyrenees. Grab samples of the upper soil layer were collected from the main potential sediment source types along with in situ reflectance spectra. Samples were dried and sieved, and artificial mixtures of known proportions were produced for algorithm validation. Then, spectral readings of potential source and artificial mixture samples were taken in the laboratory. Colour coefficients and physically based parameters were calculated from in situ and laboratory-measured spectra. All parameters passing a number of prerequisite tests were subsequently applied in discriminant function analysis for source discrimination and mixing model analyses for source contribution assessment.The three source types (i.e. badlands, forest/grassland and an aggregation of other sources, including agricultural land, shrubland, unpaved roads and open slopes) could be reliably identified based on spectral parameters. Laboratory-measured spectral fingerprints permitted the quantification of source contribution to artificial mixtures, and introduction of source heterogeneity into the mixing model decreased accuracies for some source types. Aggregation of source types that could not be discriminated did not improve mixing model results. Despite providing similar discrimination accuracies as laboratory source parameters, in situ derived source information was found to be insufficient for contribution modelling.The laboratory mixture experiment provides valuable insights into the capabilities and limitations of spectral fingerprint properties. From this study, we conclude that combinations of spectral properties can be used for mixing model analyses of a restricted number of source groups, whereas more straightforward in situ measured source parameters do not seem suitable. However, modelling results based on laboratory parameters also need to be interpreted with care and should not rely on the estimates of mean values only but should consider uncertainty intervals as well.