Short-Term Variability Associated with 20 Meter Sequential Matched WSPR Observations: A Statistical Exploratory Study

Automated amateur radio networks such as the Reverse Beacon Network and WSPRnet record details about hundreds of millions of radio contact contacts that investigators can use to study and ultimately predict HF propagation and its relationship to solar phenomena. However, before researchers can undertake such investigations, it is crucial to understand and document the variability inherent in the measurements provided by these networks. Here, we investigated the short-term variability associated with the signal-to-noise(SNR) reports from WSPRnet. Specifically, we analyzed 2,286,311 pairs of 20 meter WSPR SNR reports observed between Jan 2017 and July 2021. Each pair consisted of two sequential WSPR observations between the same two stations, i.e., the paired observations were separated by a single WSPR time slot of two minutes.  To describe the SNR variability, we present the SNR distributional characteristics and use Generalized Linear Models (GLMs) to explore the influence of the time of day, the month of the year, and the azimuth between the stations. The models predicted the absolute SNR difference between the sequential observations. Model errors were adjusted to account for multiple observations of pairs of stations. To account for the non-gaussian data distribution, the GLMs assumed a gamma distribution with a log link. Because this study was exploratory, we included all three covariates as categorical variables rather than imposing a particular model form. The three models reported here consist of a fully specified two-way interaction between two of the three covariates, i.e., both main effects and interaction.   Computing resource limitations limited the complexity of the models investigated. Based upon the predicted model averages, two sequential WSPR reports typically vary by 6 dB. Deviations from this average are apparent by month, hour, and azimuth between the reporting stations, and we show those graphically. Future research should increase the complexity of the models to incorporate other covariates, e.g., distance or latitude, ultimately tying these data to solar and atmospheric phenomena.