EnGeoMAP - A geological mapping tool applied to the EnMAP mission

Hyperspectral imaging spectroscopy offers a broad range of spatial applications that are primarily based on the foregoing identification of surface cover materials. In this context, the future hyperspectral sensor EnMAP will provide a new standard of highly qualitative imaging spectroscopy data from space that enables spatiotemporal monitoring of surface materials. The high SNR of EnMAP offers the possibility to differentiate and to identify minerals that are showing characteristic absorption features as a 30 m x 30 m spatial mixture in the visible, the near infrared and the short wave infrared range (0.4 - 2.5 micrometre). For this purpose, spectral mixture analysis (SMA) approaches are traditionally used. However, these approaches lack in transferability, repeatability and inclusion of sensor characteristics. Additionally, they rely on image-based and randomly detected endmembers as well as on in situ or laboratory spectra that are not spatially stable in case of an image-based extraction and are assumed to be spectrally pure. In this work, a new framework is proposed that addresses these limitations considering the EnMAP sensor characteristics. It is named EnMAP Geological Mapper - EnGeoMAP. It consists of several new and adapted approaches to identify spectrally homogeneous regions. In parallel, minerals are identified and semiquantified by a se-nsor-related and knowledge-based fitting approach. Supplementary outputs are abundance, classification, homogeneity and uncertainty maps. First results show that the proposed approach offers 100% percent repeatability and gains an identification error for minerals of about 2 percent on average for different studies. In this work, an approach is proposed that aims on spectroscopic mineral modelling by image synthesis that might be applied for geological mapping.

高光谱成像光谱学（Hyperspectral Imaging Spectroscopy）拥有诸多空间应用场景，其核心基于地表覆盖物质的先期识别。在此背景下，未来高光谱传感器EnMAP将提供全新标准的高质量星载高光谱成像光谱数据，可实现地表物质的时空监测。EnMAP具备高信噪比（Signal-to-Noise Ratio, SNR），能够在可见光、近红外及短波红外波段（0.4~2.5微米）范围内，对30米×30米空间分辨率下呈现特征吸收特性的矿物混合体进行区分与识别。传统上，此类任务多采用光谱混合分析（Spectral Mixture Analysis, SMA）方法，但这类方法存在可迁移性不足、重复性欠佳，且未纳入传感器特性等短板。此外，此类方法依赖基于图像提取或随机选取的端元（endmembers），以及原位（in situ）或实验室获取的光谱数据：基于图像提取的端元缺乏空间稳定性，且默认其光谱纯净度。本研究提出一种全新的处理框架，可针对EnMAP的传感器特性弥补上述缺陷，该框架被命名为EnMAP地质制图器（EnMAP Geological Mapper, EnGeoMAP）。其包含多项创新及改进算法，用于识别光谱均一区域；同时，通过结合传感器特性与知识驱动的拟合方法，实现矿物的识别与半定量分析。该框架的附加输出包括丰度图、分类图、均一性图及不确定性图。初步实验结果显示，所提方法可实现100%的重复性，且在不同测试场景下，矿物识别的平均误差仅约2%。本研究同时提出一种基于图像合成的光谱矿物建模方法，可应用于地质制图任务。