Spatial Characteristics of Global Strong Constant-Frequency Electromagnetic Disturbances from Electric-Field VLF Data of the CSES

Abstract

Ionospheric disturbances are mainly caused by solar and Earth surface activity. The electromagnetic data collected by the CSES (China Seismo-Electromagnetic Satellite, popularly known as the Zhangheng-1 satellite) can capture many space disturbances. Different spatial disturbances can exhibit distinctive shapes on spectrograms. Constant-frequency electromagnetic disturbances (CFEDs) such as artificially transmitted VLF radio waves, power line harmonics, and satellite platform disturbances can appear as horizontal lines on spectrograms. Therefore, we used computer vision and machine learning techniques to extract the frequency of global CFEDs and analyze their strong spatial signal characteristics. First, we obtained time-frequency spectrograms from CSES VLF electric-field waveform data using Fourier transform. Next, we employed an unsupervised clustering algorithm to automatically recognize CFED horizontal lines on spectrograms, merging horizontal lines from different spectrograms, to obtain the CFED horizontal-line frequency range. In the third stage, we verified the presence of CFEDs in power spectrograms, thus extracting their true frequency values. Finally, for strong CFED signals, we generated eight revisited periods, resulting in 10,230 power spectrograms for analyzing each CFED’s spatial characteristics using a combined periodic sequence and spatial region that included frequency offsets, frequency fluctuations, and signal non-observation areas. These findings contribute to enhancing the quality of CSES observational data and provides a theoretical basis for constructing global CFED spatial background fields and earthquake monitoring and early prediction systems.

1. Introduction

Since the 1980s, electromagnetic satellites have detected various spatial disturbances [1,2,3,4,5,6,7], including solar magnetic storms, substorms, lightning, atmospheric tides, artificial very-low-frequency (VLF) signals, power harmonics generated by power systems, and electromagnetic disturbances generated by satellite platforms themselves. These electromagnetic wave disturbances exhibit different shapes on spectrograms [8]. In order to meet the needs of navigation communication, ground-based VLF transmitting stations emit continuous electromagnetic waves at a constant frequency between 10 and 50 kHz. Due to their constant frequency, they are represented as a horizontal line on time-frequency diagrams [8]. The harmonic radiation generated by the ground power system is shown on spectrograms as several parallel lines that are significantly higher than the background intensity and are separated by 50/60 Hz [8,9], as shown in Figure 1. There is a weak current during the operation of the satellite in orbit, generated by the solar cell panel receiving sunlight, providing power to the satellite payload. This current flow produces low-frequency magnetic-field disturbances [10], such as changes in the magnetic field observed by DEMETER’s induction magnetometer—from 19.5 to 20 kHz in terms of power spectrum data [9]. Disturbance from DEMETER’s own body occurs mainly in the frequency range of 1 to 8 kHz, with smaller disturbances occurring below 19.5 Hz, such as at frequencies 7.33 Hz, 19.53 Hz, 39.06 Hz, and their harmonics [11]. Current flows are generated when the solar wings on solar cell panels provide power, and these noises mainly manifest in the form of machine harmonics at frequencies of 8 Hz, 13.25 Hz, 31 Hz, 77 Hz, and their harmonics [12]. These constant-frequency electromagnetic disturbances (CFEDs) are presented as spectral horizontal lines on the spectrogram, which are higher than the background intensity level.

Earthquakes are one of the most destructive natural disasters. Strong earthquakes often cause hundreds of thousands of deaths and property losses. A large number of studies have shown that electromagnetic anomalies are very sensitive to earthquakes [13,14,15,16]. When super-strong and shallow earthquakes occur, the energy of VLF and ultra-low-frequency (ULF) electromagnetic waves will become stronger, causing space ionospheric disturbances, which are used for short-term strong earthquake prediction [17,18]. One way to monitor ionospheric disturbances is to use VLF transmitters to transmit continuous VLF electromagnetic waves to space at different frequencies, which can penetrate the ionosphere. When the satellite flies over the transmitting station, the VLF signals are received synchronously by the satellite [19].

The main characteristics of VLF electromagnetic waves emitted from artificial source transmitters are low energy loss and long transmission distance. They can propagate over long distances between the Earth and the ionospheric waveguide system and exhibit significant wave–particle interaction effects [20,21,22,23]. When a satellite passes over an artificial source transmitter, it can receive the artificial source signal within a specific stable frequency range [24]. However, when electromagnetic waves propagate through the ionosphere, many parameters, including velocity and phase, as well as refraction and scattering effects, change. Therefore, the various electromagnetic responses that are excited by artificial source VLF signals in the ionosphere have different spatiotemporal characteristics [25]. There are more than 40 artificial-source VLF and low-frequency radio-wave transmitters worldwide, widely used for long-distance navigation, maritime navigation, underwater communication navigation, and ionospheric disturbance detection [26]. During the propagation process of VLF/LF radio waves, anomalies are generated when the lower ionosphere above the propagation path is disturbed by various factors such as solar flares, magnetic storms, lightning discharges, and earthquakes.

The advancement of satellite detection technology has led to the utilization of electromagnetic-field detection as the primary scientific objective for exploring ionospheric environments. The identification of artificial VLF/LF signals from the ground is typically achieved through the use of electromagnetic detectors, which rely on probe potential detection and inductive magnetometers. Due to the proliferation of VLF/LF artificial transmitters worldwide, satellites can effectively record information at each station, serving as mobile spatial receiving stations that provide a platform for studying the detection of anomalous VLF/LF artificial source signals by satellites, including their use in seismic detection applications.

At present, the signal-to-interference-to-noise ratio (SNR) method is used for detecting satellite VLF radio wave signals to obtain earthquake-related disturbances. Studies have found that the SNR of the VLF radio wave signal decreases significantly before earthquakes, with recovery after the event and similar variations observed by multiple stations [27,28,29,30,31,32,33]. Similarly, the amplitude method used to detect satellite VLF wave signals can show a significant decrease or increase in the amplitude of the VLF wave signal before an earthquake [31,32,33,34,35]. Most of the current studies on electromagnetic response backgrounds are based on specific VLF transmitting stations and do not consider the global change characteristics of VLF constant-frequency electromagnetic disturbances (CFEDs) [36,37]. We attempt to utilize global CFEDs to study the anomalous changes before an earthquake. This necessitates an initial analysis and discussion of the spatial distribution characteristics of these CFEDs.

Therefore, this article uses CFED horizontal-line features on spectrograms [8,38] and integrates computer vision technology, unsupervised clustering methods, and statistical analysis. We have established a method for identifying CFEDs on time-frequency spectrograms, verifying and extracting their frequencies using power spectrograms. We then analyzed the spatial characteristics of each strong CFED using periodic sequences and spatial regions. This research provides a theoretical foundation for improving the quality of CSES observation data; studying and finding out what causes frequency fluctuation and frequency offset; investigating the characteristics of ionospheric anomalies potentially induced by earthquakes and the relationship between ionospheric disturbances and the occurrence of earthquakes; and targeting space-based earthquake monitoring and early warning systems.

2. Data Collection

On 2 February 2018, China launched its first satellite for the seismo-electromagnetic monitoring of the ionosphere over seismic-prone areas— the CSES (China Seismo-Electromagnetic Satellite, also known as the Zhangheng-1 satellite). Its goal is to obtain global electromagnetic field, ionospheric plasma, and high-energy particle observation data, to carry out real-time ionospheric dynamic monitoring and seismic-precursor tracking detection in China and surrounding areas, and to explore the mechanism of seismic ionospheric disturbance [39]. The CSES’s scientific mission is to monitor the variations in the main parameters of the topside ionosphere (electric and magnetic fields, plasma parameters, charge particle fluxes) caused by natural emitters, especially earthquakes or those that are artificial [40].

The CSES carries eight types of scientific payloads [41,42]. The space electric-field detection is completed by the electric-field detector (EFD), which provides data application services for seismic observation research. The detection frequency bands are divided into: ULF (0–16 Hz), ELF (6 Hz–2.2 kHz), VLF (1.8 kHz–20 kHz), and HF (18 kHz–3.5 MHz). The CSES’s full orbit is divided into ascending and descending orbits. The ascending orbit is from the south latitude to the north, and the descending orbit is from the north latitude to the south. There are 76 orbits in 1 revisited period, and the revisited period is 5 days. The flight time of each orbit is approximately 94.6 min [43]. The orbital spacing in one period is between 4.7° and 4.8° [44]. The working range is from 65° S to 65° N.

The data in this study are selected from the waveform data and power spectrum data of the Z components of the CSES electric-field VLF band. According to the CSES satellite data specification, the data structure is shown in

Table 1. EFD VLF level 2 data-structure description.

Name	Content	Type	Size	Attribute	Remark
VERSE_TIME	Relative time	64-bit int	N × 1	Unit: ms
UTC_TIME	Absolute time	64-bit int	N × 1	YYYYMMDD HHMMSSms
WORKMODE	Work mode	16-bit int	N × 1	1: Inspection 2: Detailed investigation −1: Invalid
A131_W	X	64-bit float	N × 2048	Unit: mV/m	X component of electric-field waveform in WGS84 coordinate system
A132_W	Y	64-bit float	N × 2048	Unit: mV/m	Y component of electric-field waveform in WGS84 coordinate system
A133_W	Z	64-bit float	N × 2048	Unit: mV/m	Z component of electric-field waveform in WGS84 coordinate system
A131_P	CH1	64-bit float	N × 1024	Unit: mV/m/Hz0.5	Probe ab direction power spectrum
A132_P	CH2	64-bit float	N × 1024	Unit: mV/m/Hz0.5	Probe cd direction power spectrum
A133_P	CH3	64-bit float	N × 1024	Unit: mV/m/Hz0.5	Probe ad direction power spectrum
ALTITUDE	Satellite orbit height	32-bit float	N × 1	Unit: km	The value in WGS84 spherical coordinate system
MAG_LAT	Geomagnetic latitude	32-bit float	N × 1	Unit: degree
MAG_LON	Geomagnetic longitude	32-bit float	N × 1	Unit: degree
GEO_LAT	Geographical latitude	32-bit float	N × 1	Unit: degree	The value in WGS84 spherical coordinate system
GEO_LON	Geographical longitude	32-bit float	N × 1	Unit: degree	The value in WGS84 spherical coordinate system
FREQ	Power spectrum frequency	32-bit float	1024 × 1
FLAG		32-bit int	N × 1		Data Quality Label

.

First, waveform data of the Z component of the VLF electric field are converted from the time domain to the frequency domain using Fourier transform. Figure 2 shows the time-frequency spectrogram obtained by transforming the waveform data from descending orbit No. 051760.

3. Methodology

The study methodology consists of three main steps, as illustrated by the operational flowchart in Figure 3.

In the first stage, we recognize the horizontal lines generated by CFEDs on the time-frequency spectrogram and obtain the frequency range of CFEDs through the statistical analysis of a significant number of spectrograms.

In the second stage, using the frequency range obtained in the first step, we generate power spectrograms to validate the presence of CFEDs and automatically extract their actual frequencies.

In the third stage, we generate power spectrograms for multiple revisited orbital periods of each CFED. By analyzing the single orbital sequence and multiple revisited orbital periods in the same spatial domain, we aim to uncover the spatial characteristics of CFEDs.

3.1. CFEDs Automatic Recognition

CFEDs are represented as horizontal lines on the time-frequency spectrogram. The recognition of these lines is performed automatically using the following steps.

3.1.1. Graying

Since line recognition is based on the brightness relative to the background color, color information is unnecessary. To simplify the matrix and improve the computational speed, we first convert the spectrogram to grayscale. Grayscale conversion can be achieved using various methods, and in this study, we utilize the blue channel [45]. Gray = RGB.B represents the spectrogram, where RGB is the original spectrogram and B represents the blue channel. The grayscale result is shown in Figure 4b.

3.1.2. Horizontal Feature Enhancement

To enhance the edge of horizontal lines and improve line recognition accuracy, we employ a horizontal convolution kernel [46]. The convolution kernel is described by Equation (1).

$$\mathrm{kernel}=[1,0,-1]$$

(1)

The convolution operation is performed as described in Equation (2), where Gray represents the grayscale spectrogram and Cov_dst represents the convolution result. The convolution result is shown in Figure 4c.

$$\mathrm{cov}\_\mathrm{dst}(x,y)={\displaystyle \sum _{\begin{array}{l}0\le x^{\prime }<\mathrm{kernel}.\mathrm{cols},\\ 0\le y^{\prime }<\mathrm{kernel}.\mathrm{rows}\end{array}}\mathrm{kernel}(x^{\prime },y^{\prime })\times \mathrm{gray}(x+x^{\prime }-\mathrm{anchor}.x,y+y^{\prime }-\mathrm{archor}.y)}$$

(2)

3.1.3. Binarization

Binarization is used to convert each pixel into black or white, simplifying the image and helping the algorithm to accurately recognize lines. The binarization operation is described by Equation (3), where i and j represent pixel coordinates. In this experiment, we set max = 255 and tresh = 10. Figure 4d illustrates the binarization result.

$$\mathrm{bi}\_\mathrm{map}(i,j)=\left\{\begin{array}{ll}max\hfill & if \mathrm{cov}\_\mathrm{dst}(i,j)>tresh\\ 0\hfill & \mathrm{otherwise}\hfill \end{array}\right.$$

(3)

3.1.4. K-Means Clustering

The K-means clustering algorithm is an unsupervised clustering analysis technique in machine learning. Initially, the data are divided into K groups, and K objects are randomly chosen as the initial clustering centers μ_j (j = 1, 2 … k). The distance between each object x⁽ⁱ⁾ and each seed clustering center μ_j is then calculated, and each object is assigned to the nearest clustering center using Equation (4). The cluster center and assigned objects represent a cluster.

$$c^{(i)}:=\arg \underset{j}{\min }||x^{(i)}-{\mu }_j|{|}^2$$

(4)

For each assigned sample, the cluster center is recalculated according to Equation (5). This process is repeated until the termination condition is satisfied. In this study, K is set to 2 [45], enabling unsupervised clustering of the binary graph in both linear and nonlinear ways. The experiment was terminated after 10 iterations. Figure 4e shows the result of the line clustering, marked by red dots on the spectrogram.

$${\mu }_{\mathrm{i}}:=\frac{{\displaystyle {\sum }_{i=1}^{m}1\{c^{(i)}=k\}{x}^{(i)}}}{{\displaystyle {\sum }_{i=1}^{m}1\{c^{(i)}=j\}}}$$

(5)

3.1.5. Calculation of Recognized Line Frequency

Calculating the recognized line frequency involves establishing a proportional relationship between the height of the time-frequency spectrogram and the frequency range. Figure 5 illustrates this relationship, and Equation (6) describes the column calculation. In this equation, line_freq represents the CFED frequency corresponding to the recognized line, max_freq represents the maximum frequency depicted in the time-frequency spectrogram, map_height represents the height of the time-frequency spectrogram, and line_height represents the height of the recognized line on the time-frequency spectrogram. When affected by other space electromagnetic waves, the line shape on the time-frequency spectrogram undergoes changes such as thinning, thickening, interruption, or even disappearance, due to signal coupling and background enhancement [28]. Therefore, while the line frequency does not accurately represent the real frequency, it approximates the actual value. To accurately recognize, merge, and extract the existence and frequency range of CFEDs corresponding to the line, a significant number of time-frequency spectrograms must be analyzed.

$$line\_freq=\max \_fre-\frac{max\_freq\times line\_h\mathrm{e}ight}{map\_h\mathrm{e}ight}$$

(6)

3.1.6. Combine the Lines to Determine the CFED Frequency Range

After recognizing lines on many spectrograms, each spectrogram has its own set of horizontal lines. Subtle changes in the shape of these lines result in different line frequencies. To determine whether the recognized lines on each spectrogram belong to the same CFEDs, lines with intersecting linear frequencies between any two spectrograms are merged using a union operation. Finally, all recognized lines on the spectrogram are merged in pairs using a merging algorithm to obtain the frequency range for each CFED.

3.2. Confirmed CFEDs and Extract Its True Frequencies

By performing the above operations, we obtain frequency ranges for many lines. However, these ranges do not represent the true frequency values of CFEDs. To confirm the existence of CFEDs within these ranges, we use power spectrograms. Using the power spectrum data, we generate 8 discontinuous revisited periods, totaling 10,230 power spectrograms, according to frequency ranges. These power spectrograms help us determine the presence and true frequency of CFEDs. Figure 6 illustrates one of the power spectrograms.

3.3. CFED Spatial Characteristics Statistics

To analyze the spatial characteristics of CFEDs, we use the following methods.

Generate multiple revisited orbital-period CFED power spectrograms.

Perform statistical analyses of ascending and descending orbits to explore features and differences.

Analyze each CFED power spectrogram over a revisited orbital period to examine spectral variation characteristics.

Analyze CFED spatial distribution characteristics in the same spatial domain by examining multiple revisited orbital periods.

4. Experimental Results and Analysis

4.1. Experimental Environment

For this experiment, we used Matlab2020 to generate time-frequency and power spectrograms. We used the Cartopy library in Python 3.7 to draw geographic information and used CV2 and SKLearn libraries to recognize lines on time-frequency spectrograms.

4.2. Experimental Data

We randomly selected data from eight discontinuous revisited periods of CSES VLF data observed in 2019 and 2020. The data were preprocessed to retain only the full orbital period where the latitude range was (−64, 64) or (64, −64). Please refer to

Table 2. Experimental data of eight discontinuous revisited periods.

Orbital Period	Start and End Time	The Time-Frequency Spectrogram Number	Descending Number	Ascending Number
Period 1	6 January 2019–10 January 2019	126	67	59
Period 2	20 July 2019–24 July 2019	126	65	61
Period 3	25 July 2019–29 July 2019	116	59	57
Period 4	30 July 2019–4 August 2019	142	74	68
Period 5	1 June 2020–5 June 2020	130	69	61
Period 6	26 June 2020–30 June 2020	131	69	62
Period 7	1 July 2020–5 July 2020	131	69	62
Period 8	22 July 2020–26 July 2020	121	62	59
SUM		1023	534	489

for further details.

4.3. Strong CFED Signals

A strong CFED signal is a clearly identifiable CFED on the power spectrogram. The CFED shown in Figure 7a is considered a strong CFED signal, while the one presented in Figure 7b requires careful identification. In this paper, we focus only on clearly identifiable CFEDs. We provide a detailed description of the spatial characteristics of 10 strong CFED signals in Section 4.5.5.

4.4. Comparison of Ascending and Descending Signals

A complete orbit of the CSES includes ascending and descending orbits. The ascending orbit moves from the south to the north latitude, while the descending orbit moves from the north to the south latitude.

Table 3. Comparison of descending and ascending orbits.

CFEDs Frequency (kHz)	The Spectrograms Tiling of Descending Orbits in a Period	The Spectrograms Tiling of Ascending Orbits in a Period
6.05, 5.95
12.05
11.8
18.1

shows the power spectrograms of the ascending and descending orbits for four CFEDs over a revisited period. Through comparative analysis, we observe differences between the power spectrograms of ascending and descending orbits. There is too much interference in the power spectrum of the ascending orbit, which has a great influence on data analysis. Due to significant interference, the subsequent analysis is focused solely on descending orbits.

4.5. Spatial Characteristics of Strong CFEDs

Before analyzing the spatial characteristics of CFEDs, we provide two types of frequency variations: frequency fluctuation and frequency offset.

Frequency fluctuation: Within a single orbit’s data spectrogram, the actual frequency of the CFED signal periodically increases or decreases relative to the expected frequency. Figure 8 shows examples of frequency fluctuations. In Figure 8a, the CFED with a frequency of 12.05 kHz exhibits frequency fluctuations in a latitude range of approximately (64.90, 15). Similarly, in Figure 8b, the CFED with a frequency of 18.1 kHz shows fluctuations in a latitude range of about (64.90, 15).

Frequency offset: Frequency offset refers to the difference between the actual frequency of an electromagnetic wave and its expected or theoretical value. Figure 9 illustrates the frequency offset within the same CFED (the same constant-frequency electromagnetic disturbance) and spatial domain. In Figure 9a, the CFED has a center frequency of about 11.80 kHz, while in Figure 9b, the CFED has a center frequency of about 11.85 kHz.

Next, we delve into the spatial and frequency characteristics of strong CFEDs from various perspectives.

4.5.1. CFED with a Frequency of 12.05 kHz

This CFED, represented in Figure 10, exhibits strong characteristics with a frequency of 12.05 kHz.

The CFED signal shows fluctuations within the descending orbit spectrogram, as shown in Figure 11.

Furthermore, Figure 12 shows the frequency offset between consecutive orbits of the same CFED in the same spatial domain, highlighting the changes in frequency.

In the identical spatial domain, the power spectrogram reveals consistent discontinuity characteristics for six revisited periods, as shown in Figure 13.

Figure 14 shows the approximate region, in descending orbits, where the 12.05 kHz CFED cannot be observed—at the latitude range of about (−58, −22) and the longitude range of about (31, 122).

4.5.2. CFED with a Frequency of 10 kHz

The CFED signal with a frequency of 10 kHz is strong, as shown in Figure 15.

This CFED exists throughout the orbit with a stable signal, characterized by no frequency offset, fluctuation, or discontinuity in certain areas, as shown in Figure 16.

4.5.3. CFED with Frequencies of 6.05, 5.95 kHz

This CFED consists of two frequencies that coexist or undergo simultaneous discontinuity in the spectrogram, as shown in Figure 17.

Minor frequency fluctuations are observed in the descending orbit power spectrograms, as shown in Figure 18.

Figure 19 shows the frequency offset between two orbits of the same CFED in the same spatial domain.

The spectrogram in Figure 20 shows consistent discontinuity characteristics for six revisited periods in the same latitude and longitude spatial domain.

Similar to the previous CFED, Figure 21 shows regions in the descending orbits where CFEDs with frequencies of 6.05 kHz and 5.95 kHz cannot be observed, specifically the latitude range (−58, −22) and longitude range (31, 122).

4.5.4. CFED with a Frequency of 20.5 kHz

The CFED signal with a frequency of 20.5 kHz appears strong, as shown in Figure 22.

This CFED exists throughout the orbit with a stable signal, characterized by no frequency fluctuation, offset, or discontinuity in certain areas, as shown in Figure 23.

The four CFEDs mentioned above are analyzed in detail, while the others including the four above are summarized in

Table 4. Spatial characteristics of strong CFEDs.

Frequency kHz	Spectrogram	Descending Orbit Power Spectrograms of Period 1	Fluctuation	Off Set	Discontinuity	Unobserved Areas
12.05			Yes	Yes	Yes	Lat (−58, −22) Lon (31, 122)
10			No	No	No	No
6.05 5.95			Yes	Yes	Yes	Lat (−58, −22) Lon (31, 122)
20.5			No	No	No	No
15.58			No	No	No	No
14.5			No	No	No	No
18.1			Yes	Yes	Yes	Lat (−58, 0) Lon (34, 130)
11.8			No	Yes	Yes	Lat (−50, −36) Lon (117, 122); Lat (−58, −22) Lon (104, 120); Lat (−58, 0) Lon (71, 120); Lat (−58, −22) Lon (38, 78)
17.7			No	Yes	Yes	Lat (−58, −22) Lon (33, 119)
23.7			No	Yes	Yes	Lat (−50, −36) Lon (117, 122); Lat (−58, −22) Lon (104, 120); Lat (−58, 0) Lon (71, 120); Lat (−58, −22) Lon (38, 78)

.

4.5.5. Show Strong CFEDs in Table

Table 4 summarizes the spatial characteristics of ten strong CFEDs, including the four previously discussed CFEDs. The table provides information on frequency fluctuations, offsets, signal discontinuity, and unobserved areas for each CFED. Lat represents latitude, Lon represents longitude.

Through experimental analysis and statistics, we observe that these strong CFEDs possess distinct spatial characteristics. Some are present throughout the period orbits, while others exhibit frequency fluctuations or offsets and cannot be observed in specific areas. However, they also share common characteristics. CFEDs with frequency offsets tend to be unobserved in certain spatial domains. These unobserved areas are mainly concentrated in the southeastern hemisphere, as shown in Figure 24. Further discussion is needed to explore the shared and distinctive characteristics of these strong CFEDs.

5. Discussion

In this paper, regarding the horizontal-line characteristics of CFEDs in spectrograms, we used computer vision techniques to recognize these lines and extracted the corresponding CFED frequency values. The statistical analysis of the spatiotemporal characteristics of global strong CFED events revealed the presence of frequency offsets, frequency fluctuations, and data differences between ascending and descending orbits observed on the CSES. Aiming at these spatiotemporal characteristics of CFEDs, we will now proceed to discuss each of them separately.

5.1. Frequency Offset

Frequency offset is the difference between the actual frequency of a signal and the expected frequency. During signal generation or transmission, various disturbances or errors can cause the actual frequency to deviate from the set or expected frequency [47]. There are many reasons for frequency offset, such as temperature changes that can alter the physical characteristics of electronic components and circuits, leading to frequency offset [48]. Over time, electronic components and circuits can be affected by aging and degradation, resulting in frequency drift [49]. Issues with the quality and stability of the power supply can cause frequency offset [50]. Environmental conditions such as humidity, air pressure, and altitude can also affect the performance of electronic devices and circuits, leading to frequency offset [51], among other factors.

Frequency offset can lead to a decrease in communication quality, signal distortion, increased error rate, data loss, and damage. It has a significant impact on the correct reception, decoding, and processing of signals [52,53,54,55]. Therefore, studying and understanding the characteristics and causes of frequency offset is crucial for effective signal transmission and processing. Based on the statistical analysis of the experiments mentioned above, several strong carrier-frequency-estimation devices exhibit noticeable frequency offsets.

In this paper, these CFEDs include 12.05 kHz, 6.05–5.95 kHz, 18.1 kHz, 11.8 kHz, 17.7 kHz, 23.7 kHz, etc. (We use the approximate center value of the frequency to represent a CFED). These frequency offsets have a significant impact on the quality of data analysis. Therefore, future research is needed to investigate the causes of these frequency offsets and explore methods to reduce them.

However, frequency offset quantification analysis can also be used to detect and monitor the stability and consistency of electromagnetic waves, as well as for applications such as spectrum analysis and signal processing. When studying earthquake precursors or other related phenomena, the occurrence of frequency offset may indicate changes or abnormal conditions in the electromagnetic-wave propagation environment [56,57,58,59,60]. Future research can be conducted based on the phenomenon of frequency offset to explore phenomena related to earthquake precursors.

5.2. Frequency Fluctuation

Frequency fluctuation is the periodic increase or decrease in the actual frequency of a signal relative to the expected frequency over a certain period of time [61]. Frequency fluctuations can be observed and quantified by the frequency-spectrum analysis of the signal. There can be various reasons for frequency fluctuations. For example, temperature variations can cause changes in the physical characteristics of electronic components and circuits, leading to frequency fluctuations [62]. Over time, the aging and degradation of electronic components and circuits can also cause frequency fluctuations [63]. Quality and stability issues in the power supply can lead to frequency fluctuations [64]. Environmental conditions such as humidity, air pressure, and altitude can affect the performance of electronic devices and circuits, leading to frequency fluctuations [65].

Frequency fluctuations can cause a decrease in communication quality, signal distortion, increased error rates, as well as data loss and corruption [66,67,68,69]. However, frequency fluctuations also have significance. For example, they can be used to monitor and analyze the propagation characteristics and environmental changes of electromagnetic waves, which is important for communication networks, radar systems, and more [70]. More importantly, the occurrence of frequency fluctuations can indicate the possibility of earthquake precursors or other related phenomena, providing important clues for earthquake research [71]. Signal-processing and frequency-spectrum analysis: by processing and analyzing frequency fluctuations, useful information can be extracted from the signal and used for frequency-spectrum analysis and signal-processing applications [72].

In this experiment, we observed frequency fluctuations in several CFEDs on the CSES (Chinese Seismo-Electromagnetic Satellite), such as 12.05 kHz, 6.05–5.95 kHz, and 18.1kHz, etc. Our future work will focus on investigating the causes of these CFEDs’ frequency fluctuations, as well as further studying the extent and temporal-spatial domains of fluctuations to prepare for seismic-precursor prediction research.

5.3. Data Difference between Ascending and Descending Orbits on the CSES

Based on experimental statistics and analysis, we found that there are some differences in the quality of ascending and descending orbital data from the CSES satellite. These differences manifest themselves as clearer, and there is less background disturbance in power spectrograms generated during descending orbits compared to ascending orbits. One of the reasons for the differences between ascending and descending orbit data is due to their different orbital positions and inclinations, which result in changes in the related physical environment and observation conditions [73,74]. Satellites ascend from lower orbital altitudes to higher ones during ascending orbits, while they descend from higher altitudes to lower ones during descending orbits, leading to different physical conditions in different altitude ranges. Secondly, the ascending and descending orbits pass through different times and locations on Earth, resulting in the observation of different physical phenomena and environmental conditions at different times, thereby causing differences in the data between ascending and descending orbits [75,76].

To address the differences in ascending and descending orbit data, it is common to treat them as separate data sets, allowing for a better understanding and interpretation of the characteristics and trends within each data set [77,78]. Additionally, based on factors such as orbital altitude and local time, segment analysis or specific analysis methods can be employed for the ascending orbit data [79,80]. In summary, when dealing with satellite ascending and descending orbit data, it is important to recognize their differences and employ appropriate data analysis and processing methods to fully understand and utilize the characteristics and information within these two data sets.

5.4. Method Improvement

At present, we use artificial statistical methods to process data, but as the amount of data increases, this operation is not accurate enough, and the workload is very large. The next stage of work, based on computer vision technology, uses machine learning methods to automatically identify CFEDs on the power spectrogram and accurately calculate the discontinuity area. Data accuracy is very important for clearing waveform data or background fields.

In addition, we only processed 10 strong CFEDs signals. Using machine learning methods, we can improve the ability of data processing not only for weak CFED signals but also for CFEDs with a small geographical coverage.

6. Conclusions

Seismic ionospheric disturbances exhibit complex and diverse characteristics. To enhance the early prediction of earthquakes using CSES satellite observation data, it is essential to understand the spatial and temporal distribution of other electromagnetic disturbances. Therefore, we used computer vision technology to extract the frequency ranges of global CFEDs from time-frequency spectrograms obtained by the Fourier transformation of waveform data. By analyzing a dataset of 1023 full-orbit waveform samples and 10,230 strong CFED power spectrograms captured during 8 revisited CSES EFD VLF periods, we identified their spatial characteristics in both ascending and descending orbits, including frequency fluctuations, frequency offsets, and unobserved areas. Through this analysis, we discovered both shared and distinctive spatiotemporal features among these CFEDs. These findings contribute to enhancing the quality of CSES observational data. In addition, they hold significant value for studying the disturbance characteristics of other space electromagnetic waves, earthquake monitoring, early prediction efforts, and waveform data cleaning through waveform suppression.

Moving forward, our research will focus on three main aspects: investigating the factors contributing to these spatial features; exploring patterns of change in CFEDs and other spatial disturbances—in particular, the variation in CFEDs before earthquakes; and implementing CFED-suppression techniques to clean waveform data.