Ionospheric Observations from the ISS: Overcoming Noise Challenges in Signal Extraction

This paper presents a statistical processing pipeline using a scaled Vecchia Gaussian process to overcome noise challenges in ISS-based EP-EE instrument data, enabling the recovery of ionospheric signals and increased data coverage during the Solar Cycle 25 maximum.

The Gist

The Big Picture: Listening to a Whisper in a Storm

Imagine you are trying to listen to a friend whispering a secret to you, but you are standing in the middle of a roaring stadium during a thunderstorm. The wind (solar activity) is howling, the crowd (the spacecraft's own electrical noise) is shouting, and your friend's voice (the actual data from space) is incredibly faint.

For years, scientists have been trying to hear that whisper using a special microphone called EPEE (Electric Propulsion Electrostatic Analyzer Experiment) attached to the International Space Station (ISS). The ISS orbits Earth, acting like a floating weather station for the "topside" of our atmosphere (the ionosphere).

The problem? The microphone is so sensitive that it picks up the "hiss" of the equipment itself. In the past, scientists had a rule: "If the signal is too quiet, throw it away." They would delete any data that looked too close to the background noise. This was like throwing away 98% of your friend's whispers because they were too quiet to hear over the wind. You ended up with huge gaps in your story, missing crucial moments of what was actually happening.

This paper introduces a new, smarter way to listen. Instead of throwing away the quiet whispers, the team built a statistical filter that learns the difference between the "hiss" of the microphone and the "voice" of the friend.

---

The Problem: The "Static" vs. The "Signal"

The EPEE instrument measures tiny electrical currents from charged particles in space.

• The Signal: When the ISS flies through a patch of space plasma, the current spikes. This tells us about the density and temperature of the space around us.
• The Noise: The instrument itself has a "noise floor" (a baseline hum). Sometimes, the real signal is so weak it looks just like that hum.

The Old Way:
If the reading was below a certain level (the noise floor), scientists assumed it was just garbage and deleted it.

• Analogy: Imagine trying to take a photo of a firefly in the dark. If the firefly is dimmer than the camera's sensor noise, the old method would say, "That's just a speck of dust; delete the photo." You'd end up with a photo album full of black holes where the fireflies should be.

The New Way:
The authors realized that the "noise" isn't random static; it has a shape. It's like a consistent hum. If you can map out exactly what that hum looks like, you can subtract it to reveal the firefly underneath.

---

The Solution: A Four-Step "Denoising" Pipeline

The team created a statistical pipeline (a recipe for cleaning data) that works in four main steps. Think of it like a high-tech laundry service for data.

Step 1: The "Smoothie" Maker (Gaussian Processes)

First, they take the messy, jagged data points and blend them into a smooth surface.

• Analogy: Imagine a bumpy road made of data points. Instead of driving over every single bump, they use a "Gaussian Process" (a fancy statistical tool) to draw a smooth, continuous line that represents the road's general shape. This helps them see the big picture without getting stuck on every tiny pothole.

Step 2: Finding the "Silent" Moments

They need to find the times when only the background noise is present, with no real signal.

• Analogy: They look for moments when the stadium is empty. They know that if the "friend" (the signal) is whispering, the current will spike in a specific energy range. If the current is flat and low across all ranges, it's likely just the equipment humming. They isolate these "silent" moments to learn exactly what the equipment's "voice" sounds like.

Step 3: Building the "Noise Mask" (The Baseline)

This is the magic trick. They take those "silent" moments and fit a mathematical curve to them. They use a mix of shapes (a "Richards curve" and a "parabola") to perfectly describe the shape of the noise.

• Analogy: They create a custom "noise mask." Imagine you have a piece of paper with a hole cut out of it that matches the shape of the background hum perfectly. This mask represents the "instrument baseline."

Step 4: Subtracting the Mask

Now, they take the original messy data and subtract this "noise mask."

• Analogy: You take the photo of the firefly in the dark, and you subtract the photo of the dark room (the noise). Suddenly, the firefly pops out clearly, even if it was very dim.
• The Result: They don't throw away the quiet data anymore. They clean it. This increased their usable data by over 98%.

---

Why Does This Matter?

1. Filling in the Gaps
By keeping the "quiet" data, they now have a continuous stream of information. Before, there were huge black holes in their timeline where they knew nothing. Now, they can see how the ionosphere changes second-by-second.

2. Better Space Weather Forecasts
The ionosphere affects GPS, radio communications, and satellite navigation. If we don't understand it, our GPS might drift, or radio signals might drop. This new method gives us a clearer, more complete picture of space weather, helping us predict when a solar storm might mess up our technology.

3. Safety for Astronauts
The ISS can build up an electrical charge (like static electricity on a balloon). If this charge gets too high, it can be dangerous for astronauts during spacewalks. By getting a better read on the plasma around the station, they can better predict and manage this charging, keeping the crew safe.

The Bottom Line

The authors took a problem where scientists were throwing away 98% of their data because it looked "too noisy." By using advanced statistics (specifically something called a Scaled Vecchia Gaussian Process) to learn the shape of the noise itself, they were able to subtract the noise and reveal the hidden signal.

It's the difference between saying, "I can't hear you, so I'm hanging up," and saying, "I know the wind is loud, but I've learned to tune it out so I can hear your whisper."

The result? A much clearer view of our planet's upper atmosphere, better predictions for space weather, and a lot more data to help keep our satellites and astronauts safe.

Technical Summary

1. Problem Statement

The Electric Propulsion Electrostatic Analyzer Experiment (EPEE), deployed on the International Space Station (ISS) in March 2023, measures ion energy and current to derive plasma parameters critical for understanding space weather and spacecraft charging. However, the data presents significant challenges:

• Instrument Noise Floor: The instrument has a noise floor of approximately 0.15 nA. When true ionospheric signals are weak, measurements hover near this floor, making it difficult to distinguish between actual signal and noise.
• Data Rejection: Traditional processing methods discard data points where the current falls below a specific threshold or where the "peak" current occurs at high energy bins (indicating non-physical spacecraft potentials). This results in massive data loss (over 98% of low-current observations were previously rejected).
• Irregular Sampling & Artifacts: The data suffers from irregular temporal sampling and artifacts caused by the ISS environment, leading to erroneous spacecraft potential calculations when standard peak-finding algorithms are applied to noisy distributions.
• Calibration Gaps: The loss of data creates gaps that hinder cross-calibration with the Floating Potential Measurement Unit (FPMU), a gold-standard instrument on the ISS, limiting the ability to validate plasma density and temperature models.

2. Methodology

The authors propose a four-step statistical processing pipeline that replaces simple thresholding with a sophisticated Gaussian Process (GP) framework and iterative baseline modeling.

Step 1: Smooth Current Surface Estimation

• Model: Instead of treating raw data points as independent, the authors model the current surface $I(t, E)$ as a function of time ($t$) and energy ($E$) using a Gaussian Process (GP).
• Approximation: To handle the computational cost of inverting large covariance matrices ($O(n^3)$), they employ a Scaled Vecchia approximation. This approximates the joint density of the latent surface by conditioning each observation on a fixed-size set of neighbors ($m=30$).
• Kernel: An anisotropic Matérn 3/2 kernel is used, allowing for different correlation lengths in the time and energy dimensions.
• Output: This produces a smooth, continuous estimate of the current surface ($I_S$), filling in gaps caused by irregular sampling.

Step 2: Background Identification and Baseline Profiling

• Selection: The algorithm identifies "background-dominated" timestamps (likely noise) by looking for periods where the maximum current occurs in high energy bins (50–100) or where the current profile lacks a distinct peak.
• Iterative Fitting: For these candidate timestamps, the smooth current profile is modeled as a sum of three components:
  1. Richards Curve: A generalized logistic function modeling the monotonic rise and plateau of the instrument background.
  2. Parabolic Term: A second-order polynomial to capture low-order curvature not accounted for by the Richards curve.
  3. Gaussian Peak: A localized peak representing the true ionospheric signal.
• Sequential Strategy: The components are fitted sequentially (Richards $\to$ Gaussian on residuals $\to$ Parabolic on remaining residuals) to prevent the flexible Richards curve from absorbing the true signal. The Gaussian component is used only to protect the baseline fit and is not included in the final background profile.
• Consolidation: A single, conservative instrument baseline profile ($N^*$) is selected from the candidates based on the smallest integrated magnitude over low-energy bins.

Step 3: Signal Extraction

• The learned baseline $N^*$ is subtracted from the smoothed GP surface ($I^* = I_S - N^*$).
• The maximum current value ($I^*_{0,t}$) and its corresponding energy ($E^*_{0,t}$) are identified from this cleaned surface.
• These values are used to calculate the spacecraft potential ($\phi_t$) using the drifted Maxwellian distribution relation.

Step 4: Post-Processing

• Filters are applied to remove physically implausible results (e.g., energy values > 45 eV) and sudden artifacts, while retaining flagged data points for transparency.

3. Key Contributions

• Statistical Denoising Framework: The paper introduces a novel method to extract signals from near-noise-floor data without discarding observations, utilizing a Scaled Vecchia GP for scalable surface fitting.
• Adaptive Baseline Modeling: The development of a Richards-Parabolic-Gaussian decomposition allows for the separation of complex instrument background behavior from true ionospheric signals in a way that standard thresholding cannot.
• Data Recovery: The method successfully recovers data points that were previously deemed "untrustworthy," increasing the usable dataset significantly.
• Physical Insight: The authors demonstrate that "untrustworthy" data is not random noise but is spatially concentrated, suggesting it may be indicative of specific physical phenomena (e.g., the Equatorial Ionization Anomaly).

4. Results

• Data Retention: The new methodology reduced the number of rejected data points by 98.2% (from 2,144 rejected points to only 38 in the test period).
• Cross-Calibration: The cleaned EPEE data shows significantly improved coherence with FPMU measurements (Wide-Sweeping Langmuir Probe and Floating Potential Probe). The smoothed estimates preserve the expected periodic structure of spacecraft charging and fill gaps that previously hindered interpolation.
• Correlation: The cleaned currents exhibit increased correlation and reduced calibration residual variance compared to the original thresholded data.
• Spatial Clustering: The few remaining "untrustworthy" points (n=38) are highly concentrated in a specific latitude range, whereas the original thresholding method scattered these points randomly. This suggests the new method effectively isolates true anomalies rather than random noise.

5. Significance

• Mission Safety: By providing more continuous and accurate estimates of spacecraft potential, the method enhances the ability to monitor and mitigate hazardous spacecraft charging, protecting astronauts and ISS systems.
• Space Weather Monitoring: The increased data coverage enables better monitoring of ionospheric variability, particularly during the Solar Cycle 25 maximum, which is critical for satellite navigation and communication systems.
• Model Validation: The high-quality, continuous data stream allows for rigorous validation and potential calibration of the International Reference Ionosphere (IRI) model, especially during anomalous solar conditions where empirical models often fail.
• Generalizability: While specific to EPEE, the framework offers a generalizable approach for signal extraction in noisy sensor data across various scientific domains, particularly where data is sparse or near the detection limit.