- Remote sensing technologies are methods for acquiring information about the Earth’s surface and subsurface without direct physical contact.
- A sub-bottom profiler generates low-frequency sonar waves that penetrate the seafloor, reflecting off various subsurface sediment layers.
- Continuous refinement of these models can enhance accuracy over time, ensuring more efficient and effective analysis in the future.
- Continuous refinement of these models can enhance accuracy over time, ensuring more efficient and effective analysis in the future.
- Existing methods play a very important role in solving this issue, but a future of machine learning and artificial intelligence can provide an efficient and intelligent solution.
Overview
This article will discuss the different methods currently used for sediment classification, ranging from traditional field methods to remote sensing and GIS. It will also explain why sediment classification is essential and the need for a more straightforward approach to this problem. It will detail how machine learning can provide approximate yet valuable answers to this problem and make it a much simpler problem.
What is sedimentation?
It might seem unusual to start with a question, but questions are often the best way to explore and learn about a topic, especially as important as this one. So, let us dive right into it but in steps:
- Particles: Sediments are broken-down particles that settle on the underwater floor. These sediments come from various sources, such as rocks, cliffs, glaciers, etc. They erode over time, and their particles are transported to basins by natural forces.
- Accumulation: The processes described above occur over extended periods, accumulating layers of different particles on the seabed. This accumulation is driven by gravity and influenced by water flow and turbulence, forming sediment layers.
- Sediment Layers: The layers described in the above step are unique and important to understanding the underwater ecosystem. They are characterized by soil composition, grain size, etc.
What do these layers hold for us? What is the need for their classification? Sediment layers, in general, can be labelled as a key to the past. They can tell us about past geological events, the geography of the area, and much more. It also helps us understand climate change.
What exactly is sediment classification, and how does it happen presently?
Classifying different sediments based on their physical and chemical characteristics is a very detailed process. It consists of many segments, which are listed below:
- Grain Size Classification
- Mineral Composition
- Chemical Composition
- Chemical Composition
- Sediment Texture & Structure
- Analysis of Organic Content
Here, I will discuss some modern techniques used for this analysis. Methods for sediment classification vary in all verticals from basic to intricate and approximate to very accurate. The most basic method is analysing the different layers by hand. Field Sieving and Core Sampling, with the help of grain charts, are other methods used to perform this analysis. In field sieving, you use sieves of different sizes to filter out the soil, find the grain size, and then classify it. We also have more advanced methods, which are listed below:
- Laser Diffraction: This method uses the principle of light scattering. We shine a laser on the sample, and particles of different sizes deflect light at different angles. The basic relation is that the scattering angle is inversely proportional to the particle size. After tabulating all the angles, we will generate a particle distribution graph, which will be analysed later to get the results.
- Sediment Image Analysis: It is what it sounds like. A detailed analysis of their visual properties is done by looking at images. This method uses advanced software and technologies to analyse the sediment particles’ size, shape, and texture.
- Remote Sensing Technologies: Remote sensing technologies are methods used to acquire information about the Earth’s surface and subsurface without direct physical contact. These technologies utilize electromagnetic radiation (EMR) to detect and monitor various environmental and geological features. Data collection involves deploying sensors on satellites, aircraft, or drones to detect EMR across various spectral bands. Data processing includes preprocessing to correct errors and image analysis to classify and map sediment properties. Validation is done through ground truthing with field measurements to ensure data accuracy and reliability.
- Spectral Analysis: Spectral analysis is a technique used to analyse the properties of materials based on their interaction with electromagnetic radiation (EMR). Scientists can infer various characteristics, such as composition and structure, by examining the specific wavelengths of light that a material absorbs, reflects, or emits. In sediment classification, spectral analysis helps identify different types of sediments based on their unique spectral signatures.
All these methods have advantages, yet they all have a disadvantage: extensive on-field assessment and post-processing of data. This is costly, time-consuming, and labour-intensive. Machine learning can provide another path to this issue. But before we come to our ML algorithm, we first need to understand the data this algorithm will work on.
A sub-bottom profiler generates low-frequency sonar waves that penetrate the seafloor, reflecting off various subsurface sediment layers. The instrument records the time and amplitude for these waves to return, creating a vertical subsurface profile. This data is processed using mathematical models based on energy-time analysis and reflection coefficients to classify sediment types. Advanced signal processing techniques, such as filtering and deconvolution, are applied to enhance data quality. The reflection coefficients help identify interfaces between different sediment layers, allowing for the creation of detailed maps that show sediment layer thickness, continuity, and composition.
Although this is a novel approach, it still does not solve the extensive post-processing issue. Mathematical models are very complex, and it takes a lot of time to compute results. To solve this issue, I plan to make an ML model that can use this data to generate sediment classification. Now, the main question is, how?
Per my understanding, each value in my data tables represents a signal amplitude value. In my case, these data tables are of size 732×1867. The dataset consists of 732 rows, each representing variations in depth. Additionally, 1867 points represent different horizontal positions along a survey line at a particular depth. We need to use this data to generate values of some essential features, such as
- Time of reflection
- Amplitude of reflected signals
- The energy of reflected signals
- The energy of reflected signals
- Reflection coefficients
- Frequency of the sonar waves
We’ll proceed by normalizing the features to ensure consistency across the dataset. We’ll then develop a predictive model, preferably utilizing a support vector machine or neural network algorithms based on their respective accuracies. The dataset will then be divided into training and testing subsets to evaluate the model’s performance. Ground truth data can be obtained through mathematical models or real-world analysis for precise comparisons. Once the model is trained, it can quickly analyse input data and provide approximate results. Continuous refinement of these models can enhance accuracy over time, ensuring more efficient and effective analysis in the future.
What I have proposed above is still yet to be tested by myself, and let us hope that the next digest article will describe the reasons for the success of this model and a more detailed explanation of the model itself.
Bhavin Jain, MRC Intern, IIT Bombay
About Author
Bhavin Jain is pursuing a BTech degree in Civil Engineering from IIT Bombay. He is interested in data analytics and marine ecosystems and loves to read all forms of literature.