The study began with WSPR reports transmitted and received by roughly 20,000 active amateur stations worldwide. Each report supplies the time, transmitter and receiver coordinates, operating frequency, great-circle distance and signal-to-noise ratio, enabling researchers to infer where a signal was reflected in the ionosphere. The researchers retained only reports on the 28 and 50 MHz bands whose paths extended between 750 and 2,500 km and whose signal-to-noise ratio exceeded −15 dB. These thresholds were designed to exclude ground-wave and tropospheric-scatter propagation, multi-hop reflections and reports decoded from the noise floor. Of the approximately 830 million raw records, about 1.49 million passed the filters: around 1.48 million at 28 MHz and 10,000 at 50 MHz. The relatively low F2-layer critical frequency expected in 2021 also made ordinary F-layer skywave propagation unlikely to explain the selected 28 and 50 MHz signals.

For each qualifying report, the investigators estimated an Es reflection point by taking the midpoint of the great-circle path between transmitter and receiver. This simplification assumes a single, specular reflection, whereas tilted Es structures, lateral scattering and magnetic ion birefringence can shift the true reflection point away from the geometric center. The resulting location error can reach roughly 150–300 km. To identify dense groups of reflection points despite that uncertainty and the irregular shape of Es clouds, the team used the Density-Based Spatial Clustering of Applications with Noise algorithm, or DBSCAN. The baseline analysis used a 2-degree spatial radius, equivalent to about 200 km at mid-latitudes, required at least five reports in a neighborhood, and placed reports into non-overlapping 30-minute time windows. The time interval was selected because Es structures commonly persist from about 30 minutes to several hours and drift at roughly 50–100 metres per second.

The baseline pipeline produced 27,253 window-level detections. Because one physical Es process may generate reports in several adjacent windows, these numbers are not an absolute tally of discrete events; instead, they provide a consistent relative measure of occurrence intensity on the selected temporal grid. Sensitivity tests altered the minimum cluster population, spatial radius, window length and window overlap. Requiring eight rather than five reports reduced the count to 17,292, while lowering the threshold to three increased it to 46,784. A 60-minute window yielded 22,431 detections, a 15-minute window yielded 32,495, and overlapping 30-minute windows advanced every 15 minutes yielded 54,526. Across these configurations, the summer-to-winter ratio remained close to three and the same three mid-latitude hotspots persisted. The overlapping analysis was equivalent to approximately 13,600 temporally independent processes, but the paper used the baseline window-level convention throughout.

Defining Northern Hemisphere summer as May through August and winter as November through February, the baseline analysis recorded 19,002 summer detections, or 69.7% of the annual total, compared with 6,146 winter detections, or 22.6%. The resulting ratio of 3.09 matched the well-established tendency for mid-latitude Es activity to peak in summer. July produced the largest monthly count, with 6,597 detections, followed by June with 5,410 and May with 3,800. The winter pattern was less symmetrical: December and January each contributed more than 2,500 detections, whereas February and November contributed fewer than 600. The researchers attributed much of this contrast to the Southern Hemisphere summer, which falls during those months in the Northern Hemisphere.

A separate analysis of the 25–35° south mid-latitude belt supported that interpretation. It found 2,540 detections from December through February, compared with 454 during June through August, yielding a Southern Hemisphere summer-to-winter ratio of about 5.6 to one. The authors associated the elevated counts over Australia with the Southern Hemisphere warm season, although they cautioned that the regional sample was much smaller than the Northern Hemisphere sample and carried greater relative uncertainty because fewer amateur stations operate south of the equator. The study also interpreted the patchy clustering, with horizontal scales of roughly 100–200 km, and the seasonal maximum as qualitatively consistent with the wind-shear model. In that framework, neutral-wind shears concentrate meteorically supplied metallic ions into thin, electron-rich layers near 90–120 km altitude.

Spatial normalization for the number of active transmitters still left three mid-latitude belts clearly visible: Europe at 35–55° north, East Asia at 25–45° north and North America at 30–50° north. The correction divided detections in each grid cell by the number of unique transmitting stations located there, reducing prominent sampling artifacts without claiming to remove them completely. Europe accounted for 35.4% of detections in absolute terms, North America for 27.9% and East Asia for only 1.8%. East Asia’s low share does not necessarily indicate weak physical activity. The 750-km minimum path length excluded many short routes within Japan, while sparse station coverage in mainland China further reduced sampling. Normalized rates in parts of South America and Africa rose, but remained below those of the principal mid-latitude regions.

Signal frequency provided a second, indirect indication of Es intensity. Detections containing only 28 MHz reports were classified as moderate, while clusters containing 50 MHz reports were treated as strong. About 87.6% of all 50 MHz reports occurred during May through August, indicating that the higher-frequency reflections were concentrated in the Northern Hemisphere summer. The monthly average was roughly 250,000 reports at 28 MHz during summer, compared with about 1,500 at 50 MHz. Monthly counts for the two bands had a Pearson correlation of −0.06, which the authors interpreted as a difference in seasonal sensitivity rather than physical anticorrelation. Frequency composition cannot serve as a calibrated measure of critical frequency or electron density, because transmitter power, receiver performance, antennas and path geometry also vary throughout the crowdsourced network. It therefore distinguishes relative classes of event rather than measuring Es intensity continuously.

Cross-validation with ionosondes demonstrated both the promise and the difficulty of comparing a wide-area radio network with point instruments. At the Juliusruh station in Germany, using June and July 2021 data, an 800-km spatial match window and a 3 MHz foEs threshold, the WSPR method had a probability of detection of 57%, a false-alarm rate of 51% and a critical success index of 4%. For stronger events under the study’s higher-threshold criterion, probability of detection rose to 59%, while the false-alarm rate remained 51% and the critical success index was 1%. At Wakkanai, Japan, probability of detection was 12% at both thresholds, with false-alarm rates of 15% and 13% and critical success indices of 12% and 11%. The authors emphasized that an ionosonde measures Es directly overhead, whereas a WSPR detection may locate an event several hundred kilometres away. A genuine event within the radio method’s broader footprint can therefore be counted as a miss by the point instrument, making the low critical success index and high Juliusruh false-alarm rate measures of coverage mismatch rather than intrinsic detection error.

The results support WSPR as a practical supplement to, rather than a replacement for, ionosondes. Ionosondes provide precise local measurements of foEs, while the crowdsourced network offers much broader spatial sampling and continuous temporal coverage at little direct infrastructure cost. The method remains constrained by station density, midpoint-location uncertainty and its reliance on relative event counts. Its current band selection also applies most cleanly to the early ascending phase of solar cycle 25, represented by 2021’s annual mean sunspot number of 29.6 and F10.7 flux of 81.8 s.f.u. Applying the same pipeline to the more active years 2022–2025 produced sharply higher detection totals that the authors attributed to F-layer contamination once the maximum usable frequency exceeded 28 MHz. Extending the technique into those years would require additional F-layer filtering or a different primary band, while broader ionosonde comparisons, station-sensitivity models and machine-learning methods could improve its validation and geographic representativeness.

Subject of Research: Crowdsourced WSPR monitoring of sporadic E ionospheric activity during 2021

Article Title: Detection of the sporadic E layer using crowdsourced WSPR signal reports during the ascending phase of cycle 25

Article References: Liu, Y., & Dai, L. (2026). Detection of the sporadic E layer using crowdsourced WSPR signal reports during the ascending phase of cycle 25. Frontiers in Astronomy and Space Sciences, 13, Article 1946062. https://doi.org/10.3389/fspas.2026.1946062

Image Credits: AI Generated

DOI: 10.3389/fspas.2026.1946062

Keywords: sporadic E layer, WSPR, amateur radio, ionosphere, VHF propagation, DBSCAN, crowdsourced science, ionosonde validation, solar cycle 25, space weather, Detection, sporadic