Akua McLeod: An Automated Approach to Processing and Detection of Artifacts in Phase-Sensitive Ice Penetrating Radar Data

Authors

Akua M. McLeod1, Sean T. Peters1, Dustin M. Schroeder1,2, Nicole Bienert1, Tun Jan Young3, Poul Christoffersen3

1: Stanford University, Department of Electrical Engineering, Stanford, CA United, States

2: Stanford University, Department of Geophysics, Stanford, CA United, States

3: University of Cambridge, Scott Polar Research Institute, Cambridge, United Kingdom

Abstract

Recent work has demonstrated how large variations in surface meltwater created clipping artifacts in the processed autonomous phase-sensitive radio-echo sounder (ApRES) signal. These artifacts can obscure real signals, significantly impact the interpretation of the measurement, and thus reduce the accuracy of the results. To expand on this existing work, we present methods for automating the identification of these artifacts and several other data anomalies while analyzing and processing multi-input multi-output (MIMO) ApRES data. In this paper, we analyze MIMO ApRES data gathered in the ablation zone of Store Glacier, West Greenland, from July 2017 to July 2019. Our analysis highlights several artifacts identified in the data, including radio frequency interference (RFI) issues, receiver failure events such as elevated thermal noise, and signal leakage between channels. We present methods for automating identification of each of these anomalies and processed ApRES signal artifacts while working with large datasets. Specifically, we leverage mean squared error (MSE) analysis, clipping detection and quantification, and calculation of total power over time in the frequency domain and the time domain. We then evaluate the efficacy of these methods for processing data artifacts. Additionally, we demonstrate that by highlighting artifacts caused by clipping, instrumentation error, and ambient noise, our analysis provides an effective approach to avoid misinterpretation of one’s data. We find these methods to be particularly useful for glaciologists seeking to mitigate the impact of large variations in the environment on their ApRES signal. Ultimately, by characterizing and minimizing signal disruption, our methods increase the accuracy of ice-penetrating radar data analysis and interpretation.

Click HERE to view my slides.

6 Comments on “Akua McLeod: An Automated Approach to Processing and Detection of Artifacts in Phase-Sensitive Ice Penetrating Radar Data

  1. What a great presentation! Very clear and straight forward to me as a novice in this field of radar data.
    Can you tell me how you might think about setting the limit on the clipping? What factors would you look at to know if you were to aggressive or were leaving in too much?
    Also, I’m curious about splitting of the ice. I remember that the ice broke apart under the instrument. Did that impact the processing?
    – Jenny

    • Thank you Jenny!

      I’m still in the process of deciding how to determine what amount of clipping is acceptable, but I’m inspired by an experiment described in this clipping paper published earlier this summer: https://www.cambridge.org/core/journals/annals-of-glaciology/article/depthdependent-artifacts-resulting-from-apres-signal-clipping/74C8B78B415646A16B15CF418749D9E0. In this paper, Vaňková and others generate a synthetic deramped signal and simulate sampling that signal with an ApRES. They then introduce clipping to evaluate how it changes the synthetic signal. I think a similar approach could be used to determine what percentage of clipping makes the data unusable. By gradually increasing the amount of clipping that occurs in the simulation, we could work to identify at what levels of clipping, if any, the signal remains relatively unchanged. Applying this process to several different synthetic signals might help us determine whether our limit is too aggressive, or not stringent enough.

      About the crevasse that formed in the ice below the instrument – it didn’t change much about how we process the data. I’m still hoping to find a change in the recorded signal, however! Hopefully, after filtering out all of these data anomalies, I’ll be able to better identify when the crevasse opened.

      Thank you for your questions and feedback!
      – Akua

  2. Great Presentation Akua!

    It’s good to know all potential artifacts that could obscure our interpretation of the data and to see a compelling underlying signal after those artifacts have been removed. What advice would you give glaciologists that are planning on putting out a-pres systems to best avoid these?

    Great job again,
    – Dusty

    • Thank you so much, Dusty!

      I think in terms of advice, I would borrow from the recommendation made in the Vaňková paper, and suggest that glaciologists deploying ApRES systems in environments with lots of surface meltwater use higher attenuation settings, or a mix of attenuation and gain settings, to limit clipping and the artificially bright reflections that can result. I think some of the other data anomalies like radio frequency interference are unfortunately harder to plan against, because these artifacts are more dependent on activities of other researchers or programs near the test site that can generate signals that interfere with the data.

      -Akua

  3. Hi Akua,

    This is a great presentation, and it seems like you have some very interesting results. I’m curious how you could adapt such an anomaly detection scheme to different kinds of data – have you thought about expanding your approach to fit a range of inputs? I can imagine this kind of processing would be very useful for many other research projects.

  4. Hey Akua,
    Very cool presentation! I was just wondering if you could elaborate on how large variations in surface meltwater creates the artifacts in the data that was collected by the ApRES.

Leave a Reply

Your email address will not be published.

css.php