石油储层评价方案：集成VNIR-LWIR光谱与机器学习的工作流程

A Revolutionary Workflow for Petroleum Reservoir Evaluation: Integrating VNIR-LWIR Spectroscopy and Machine Learning

在地质勘探领域，如何快速、准确地从数百米长的岩心中找到最关键的那几块样本，一直是个巨大挑战。一项针对沙特阿拉伯油气储层的研究，展示了一种具有前景的解决方案。

In the field of geological exploration, rapidly and accurately identifying the most critical samples from hundreds of meters of core has long been a significant challenge. A study focusing on hydrocarbon reservoirs in Saudi Arabia demonstrates a highly promising solution.

注：

地质勘探是指为了查明地下地质情况，探寻矿产资源（如油气、矿产、地下水等）而开展的一系列综合性调查与研究工作。对于油气领域而言，地质勘探的核心任务是确定油气聚集的有利区域，并最终定位可供开采的油气藏。

油气储层是地下具备储集空间（孔隙、裂缝）和流体渗流能力（渗透性）的岩层，油气就储存于这些空间内。它是油气藏形成的核心要素之一。最常见的油气储层岩石类型是砂岩和碳酸盐岩（如石灰岩、白云岩）。

通过地质勘探，来发现、圈定和评价具有商业价值的油气储层。

Note:

Geological Exploration refers to a series of comprehensive investigations and research activities conducted to understand subsurface geological conditions and explore for mineral resources (such as oil, gas, minerals, groundwater, etc.). In the context of the petroleum industry, the core objective of geological exploration is to identify areas favorable for hydrocarbon accumulation and ultimately locate exploitable oil and gas reservoirs.

A Hydrocarbon Reservoir is a subsurface rock unit possessing storage space (pores, fractures) and the ability to allow fluid flow (permeability); hydrocarbons are stored within these spaces. It is one of the core elements for the formation of a hydrocarbon accumulation. The most common rock types for hydrocarbon reservoirs are sandstones and carbonates (e.g., limestone, dolomite).

The purpose of geological exploration is to discover, delineate, and evaluate commercially viable hydrocarbon reservoirs.

研究地区（Wadi Daqlah）出露岩相的综合岩石地层柱状图 / the generalized lithostratigraphy of exposed facies in  the study area (Wadi Daqlah)

「光谱范围」

在这项研究的核心技术基础是全波段高光谱成像，光谱范围覆盖了从可见光到长波红外的广阔区间：

可见光-近红外 (VNIR, 400-1000 nm)：主要对岩石中的铁离子（Fe²⁺）等过渡金属元素敏感。通过计算铁指数，可以快速评估岩石的化学组成变化。

短波红外 (SWIR, 1000-2500 nm)：方解石和白云石在此区域的特征吸收谷位置，存在约20纳米的微小但稳定的偏移（方解石~2345 nm，白云石~2325 nm）。这个差异是快速、准确绘制矿物分布图的关键。

中波红外 (MWIR, 3-7μm) 与长波红外 (LWIR, 8-14μm)：包含了碳酸盐矿物更复杂的分子振动信息。本研究通过融合这三个红外区间的数据，能够捕捉到单靠SWIR无法识别的、与岩石晶体结构和粒度相关的细微差异，从而成功区分了不同结构的白云岩。

「Spectral Range」

The core technical foundation of this study is full-range hyperspectral imaging, covering a broad spectrum from the visible to the long-wave infrared:

Visible-Near Infrared (VNIR, 400-1000 nm): This range is primarily sensitive to transition metal elements such as ferrous iron (Fe²⁺) in rocks. Calculating an iron index allows for rapid assessment of changes in the rock's chemical composition.

Short-Wave Infrared (SWIR, 1000-2500 nm): A key finding of the study is the consistent, approximately 20-nanometer shift in the characteristic absorption trough positions between calcite and dolomite in this region (calcite ~2345 nm, dolomite ~2325 nm). This difference is crucial for rapidly and accurately mapping mineral distribution.

Mid-Wave Infrared (MWIR, 3-7μm) and Long-Wave Infrared (LWIR, 8-14μm): These ranges contain more complex molecular vibration information from carbonate minerals. By integrating data from these three infrared intervals, the study was able to detect subtle differences related to rock crystal structure and grain size that are indistinguishable using SWIR alone, thereby successfully differentiating dolomites with varying textures.

白云岩（绿）与方解石（蓝）的光谱特征：白云岩的特征吸收峰位置约为2325nm，方解石的特征吸收峰位置约为2345nm。

Spectral signature of dolomite (green) and calcite (blue) with characteristic absorption band position around 2325 nm (dashed green line) for dolomite and 2345 nm (dashed blue line) for calcites.

「基于高光谱成像的数据分析工作流程」

传统地质采样可被视为“经验驱动"模式，严重依赖专家的肉眼观察和主观判断。本研究提出一套更客观的、更高效、可重复的“数据驱动"解决方案，其工作流程清晰体现了从宏观到微观的分析逻辑：

第一步：全域扫描，绘制矿物地图。研究团队首先利用SWIR波段特征，对50米长的岩心进行快速扫描，生成一张高精度的矿物分布图，清晰界定出白云石化的目标区域。

第二步：识别结构差异。在锁定白云石区域后，通过综合SWIR、MWIR和LWIR的光谱信息，并采用主成分分析（PCA）算法，系统能够放大那些与晶体结构、粒度相关的细微光谱差异。这些差异是肉眼无法分辨的。

第三步：定位最佳采样点。基于光谱差异，K-means聚类算法将白云石像素自动划分为4个类别。随后，系统会计算出每个类别的光谱中心，并推荐最靠近这些中心的岩心位置作为具代表性的采样点。

「Data-Driven Workflow Based on Hyperspectral Imaging」

Traditional geological sampling can be considered an "experience-driven" model, heavily reliant on experts' visual observation and subjective judgment. This study proposes a more objective, efficient, and reproducible "data-driven" solution. The workflow clearly demonstrates an analytical logic progressing from macro to micro:

1. Full-Scene Scanning and Mineral Mapping: The research team first utilized SWIR spectral features to rapidly scan the 50-meter core, generating a high-precision mineral distribution map that clearly delineated the target dolomitized zones.

2. Identifying Textural Differences: After identifying the dolomite regions, the system amplified subtle spectral differences related to crystal structure and grain size by integrating spectral information from SWIR, MWIR, and LWIR and applying Principal Component Analysis (PCA). These differences are entirely undetectable to the naked eye.

3. Locating Optimal Sampling Points: Based on the spectral differences, the K-means clustering algorithm automatically classified the dolomite pixels into four categories. The system then calculated the spectral centroid for each category and recommended the core locations closest to these centroids as the most representative sampling points.

A. 标准化混淆矩阵，展示基于岩相学分析与高光谱成像分析的白云岩结构分类结果（对比）。B. 针对相同的岩芯取样点样品（该样品同时用于ICP-OES实验室检测），高光谱成像计算得出的铁指数与铁浓度之间的相关性分析图。

A. Normalized confusion matrix showing the dolomite texture classification from petrographic analysis and HSI based classification. B. Correlation between HSI derived iron index and Fe concentration for the same plug samples used for ICP-OES lab measurements.

A.不同类型白云岩沿岩芯的分类及分布情况。各类别白云岩在B.可见近红外-短波红外（VNIR-SWIR）波段、C.中波红外（MWIR）波段、D.长波红外（LWIR）波段的代表性光谱。

A. Classification and distribution of the different dolomite types along the drill core. Representative spectra for each class in the VNIR- SWIR (B.), MWIR (B) and LWIR(C).

「结语」

综上所述，这篇论文的亮点在于：它展示了一个从全域扫描到采样的完整数据驱动工作流程，为地质研究提供了新思路；同时，它突破了常规，通过集成VNIR-SWIR-MWIR-LWIR多波段数据，实现了对碳酸盐岩从矿物识别到结构分类的精细刻画。

这不仅对油气行业有直接价值，也为未来在矿产勘查等领域的精细表征指明了方向。

如果您想了解可见光-近红外、短波红外波段，甚至是中波红外或长波红外波段的高光谱成像系统，欢迎联系我们！

「Conclusion」

In summary, the highlights of this paper are twofold: it demonstrates a complete data-driven workflow from full-scene scanning to sampling, providing a new paradigm for geological research; furthermore, it breaks from convention by integrating VNIR-SWIR-MWIR-LWIR multi-band data to achieve fine characterization of carbonate rocks, progressing from mineral identification to texture classification.

This holds direct value for the petroleum industry and also points the way for detailed characterization in future applications such as mineral exploration.

If you are interested in hyperspectral imaging systems covering the VNIR, SWIR, or even MWIR and LWIR ranges, please do not hesitate to contact us!

论文 / Article：

Gairola, G. S., Thiele, S. T., Khanna, P., Ramdani, A. I., Gloaguen, R., & Vahrenkamp, V. (2024). A data-driven hyperspectral method for sampling of diagenetic carbonate fabrics – A case study using an outcrop analogue of Jurassic Arab-D reservoirs, Saudi Arabia. Marine and Petroleum Geology, 161, 106691.