The majority of stations in this dataset use the purpose-built Grape V1, a low-cost receiver described by . This is a low- to intermediate-frequency receiver optimized for Doppler measurements.

It is also possible to use the software components with other hardware: as noted in the nodelist.csv file in the software repository, which is included in abridged form in Table , some of the registered nodes collect data using commercial off-the-shelf (COTS) amateur radio receivers which are capable of accepting an external frequency input from, e.g., GPS-disciplined oscillators. Citizen scientists from the amateur radio and shortwave listening communities can therefore leverage their existing hardware to contribute to the PSWS network at no additional cost and with no licensure requirement. The data processing framework of the Personal Space Weather Station network is robust to the addition and modification of new nodes and to data outages. Data collected up to 1 June 2022 are represented in the data inventory shown in Fig. .

The process of data curation is depicted in Fig. . Each station collects 24 h datasets according to an established standard and uploads them on a daily basis to a central FTP server. Test files, corrupted files, and spurious uploads are eliminated, and the data are consolidated into a single *.zip file, which is posted to the data repository . While the size of the final *.zip file varies according to the number of stations collecting data, the efficiency of compression, and other factors, it is on the order of a few gigabytes. The updated dataset can then be downloaded from Zenodo to a subdirectory in the code repository and used to create updated versions of the visualizations discussed in this paper.

An example file is shown in Appendix , which shows a file with the corresponding filename 2020-07-09TT002940Z_N0000001 _G1_EN91fh_FRQ_WWV5.csv. This filename includes, in order, the date the data were collected and the UTC time at which that collection began; the node number, corresponding to the list in Table ; the type of radio being used (e.g., “G1” indicates Grape version 1); the Maidenhead grid square in which the data were recorded; and the time standard station being measured. A detailed description of the file format and upload process is available from . Metadata at the beginning of each file record station information, including room for comments. The main table has three columns: UTC time, estimated frequency, and received power.

A map of stations to date is shown in Fig. . Stations were chosen on a volunteer basis, with some (e.g., Node 18 in California) specifically recruited to improve coverage. Clusters of stations are evident around universities involved in the project: Case Western Reserve University in Cleveland, the University of Scranton in Pennsylvania, and the New Jersey Institute of Technology each have a collection of nodes belonging to researchers. An additional cluster is generated by volunteers of the New England amateur radio community. There are also nodes close to WWV in the Fort Collins area, Colorado (e.g., Node 13), which are within the transmitter's radio horizon and can be used to confirm that trends in the data originate with the ionosphere and not the radio transmitters.

Several stations are registered as nodes but do not have data included in the dataset reported at the time of writing. This may be for one of three reasons: first, the station may have data recorded but not uploaded to the FTP server; second, the station may be in the process of installing a node; and third, the station may be used for experimentation with new data collection methods, including spectrum sampling and other frequency analysis algorithms. A central aspect of this work is its architecture as a living dataset, i.e., a dataset into which new stations and historic data may be easily incorporated.

Figure  shows the data collected by each node over time. The network is modular: new stations can easily be added, and data analysis procedures are tolerant of outages and changes in frequency for each node.

As illustrated in Fig. , electron density in the ionosphere increases during the day as a result of photoionization and decreases at night due to recombination , producing a recognizable trend in Doppler plots. Confirming this diel variation (i.e., checking for a sunrise peak) is recommended by as a benchmark for an operator to ensure that the trends observed in their station's data are geophysical in nature.

The plotting routine automatically computes the local sunrise and sunset for a given station location and shades the background accordingly. An example of data collected by Node 1 is shown in Fig. . The output produces two plots: Doppler shift on the top and amplitude on the bottom. In each case, the raw data, scatter-plotted in blue, undergo filtering to produce the filtered result, which is overlaid in yellow. By default, the data processing uses a sixth-order Butterworth low-pass filter with a cutoff frequency of 5 mHz (Tc≈3.33 min). A similar plotting routine is used to provide station maintainers with daily feedback, as described and depicted in Figs. 10–13 of . Diel fluctuations vary with local conditions but are distinct in long-term data, as discussed in Sect. . Figure  also shows a Doppler flash, which is discussed in Sect. .

Long-term trends in the data of Node 7 are shown in the time–date–parameter plots of Fig. . Two plots are shown: Doppler shift in hertz using a red–blue divergent color map (see Appendix ) and received power in decibels. Each day is represented by a column of pixels, with corresponding solar mean time lined up across the plots horizontally and time's arrow running from bottom to top. On the horizontal axis, time's arrow runs left to right, covering a span from mid-2020 to spring 2022. This is consistent with the computed sun graph of Fig. . Several observations, both geophysical and instrumentation-related in nature, can be gleaned from these two plots. First, the seasonal movement of sunrise and sunset is clearly visible at the bottom and top of the frequency plot respectively. The amplitude plot on the bottom demonstrates that reception from WWV to this station's location in the Cleveland area is much better during the nighttime, when the F2 layer of the ionosphere allows a propagation path to open up between the two locations. Vertical stripes toward the left side of both plots indicate changes or gaps in instrumentation, which are also reflected in the metadata for the affected time period. In this station's case, the station maintainer recorded a change of antenna at their station on 26 August 2021, when they switched from an off-center-fed (OCF) dipole to a magnetic loop antenna with a preamplifier. This change produced an overall increase of received power, which is clearly visible in the power plot. The lack of a corresponding change in the frequency plot above it indicates that the frequency estimation algorithm was able to function well with either antenna.

One category of ionospheric phenomena of particular interest for the PSWS network is medium-scale traveling ionospheric disturbances (MSTIDs), defined by as wavelike perturbations of ionospheric plasma with wavelengths of hundreds of kilometers, phase velocities of hundreds of meters per second, and periods between 10 min and 1 h. While MSTIDs may be associated with either atmospheric gravity waves (AGWs) from the neutral atmosphere e.g., or from electrodynamic processes e.g.,, the source of MSTIDs is still not well understood due to their ubiquitous nature and the complexities of atmosphere–ionosphere coupling.

and developed a technique to estimate TID period, speed, propagation direction, and velocity from a network of AM broadcast band Doppler receivers described by ; at the time of writing, this technique is being developed for use with HF data from the PSWS network by .

Figure a is generated using the same standard processing as Fig. . Next, the data are interpolated and filtered with a 0.5–1.2 mHz (T=14–33 min) bandpass to isolate the dominant MSTID, similar to the approach used in Sect. 3.1.2 of . The output is then separated into 4 h bins with a 90 % overlap and plotted as a spectrogram in Fig. b, which shows signatures consistent with MSTIDs.

Figure  shows the response of the network to solar flares on 28 October 2021, providing an example of a multi-instrument measurement which demonstrates how the PSWS may augment and be validated by existing professional networks. The top plot gives the X-ray irradiance as measured by NOAA's GOES-17 spacecraft, with two flares marked: an X-class solar flare with a maximum at 15:35 UTC and a smaller, C-class flare about 2 h later. Figure b shows the Doppler shift for eight stations from the day in question, color-mapped by station longitude; below it, Fig. c shows the relative received power for the same stations. A longitude-dependent Doppler flash is observed in the frequency plot in conjunction with each flare, and a radio blackout following the X1 flare is observed in the power plot. (The lone exception, Node 13, is the groundwave station near WWV.) This Doppler flash was also measured by the SuperDARN at Fort Hays, KS, as shown in Fig. , albeit at a slower cadence than the Grape measurements.

By default, no scaling is applied in the received power plot of Fig. c. As discussed in Sect. , received signal strength varies with the antenna but may not impact the accuracy of the estimated frequency. Additionally, the PSWS nodes which use commercial off-the-shelf (COTS) hardware rather than Grape receivers (see Table ) may have an automatic gain control (AGC) which impacts the utility of the power measurement. Therefore, users are encouraged to begin by examining the raw data from an event of interest before applying scaling.

WWV's transmitter is well characterized and inherently accurate, with a measured carrier stability below one part in 1012 . Allan deviation analysis by demonstrates that the Grape V1 receiver recovers frequency with an upper bound of 2 ppb (2×10-10). Further, performs calibration of the Leo Bodnar GPSDO recommended by and demonstrates that, with a frequency stability of one part in 1012 over a 1 d interval, it contributes no discernible measurement uncertainty.

Between transmitter and receiver, the received power varies according to location, antenna gain, and atmospheric attenuation. For example, in Fig. , the antenna replacement which took place at that station in August 2021 distinctly impacts the power plot but has relatively little impact on the frequency estimation. Because the frequency and power are logged together in the raw data, the end user may elect to discard or replace frequencies where the logged power is below a threshold of their choosing. Even a low-amplitude signal may yield viable frequency estimation data, however, as shown in Fig. .

Trends observed by the network may therefore reasonably be considered to be of geophysical origin, albeit the result of multiple causes. Quantifying these ionospheric propagation effects is beyond the scope of this paper. Instead, this paper will allow investigations of these effects in the future through comparisons with other instruments and data–model comparisons.

We have provided comparison to prior products e.g.,, event-based validation from alternate sources (GOES-17 and SuperDARN, per Figs.  and respectively), comparison of measured diel variation to model outcome (Figs.  and ), and initial records of sensors (see Figs. , ).

The definition of citizen science has evolved over time, and the PSWS network, which invites significant personal investment and involvement from network participants, hews more closely to the “co-created” models of intensive citizen science projects, in which participants play an active role in shaping all levels of the work, than it does to “contributory” models which emphasize crowdsourced data collection .

The Personal Space Weather Station exemplifies the key elements of citizen science projects identified by . They emphasize that participants in citizen science projects are primarily not project-relevant scientists; indeed, the majority of nodes in Table  are maintained by volunteers with no financial or academic connection to the project. also note that citizen science projects actively engage participants, engage those participants with data, and enable those participants to derive benefit from their participation: these aspects of the PSWS are attested to by . Finally, they note that citizen science projects use a systematic approach to producing reliable knowledge, help advance science, and communicate results, all aspects which are supported by this paper.

The indispensable participation of the amateur radio and shortwave listening communities in these efforts is part of a citizen science legacy in those communities which dates back to the dawn of radio and continues to the present day .