Tropospheric Ducting Prediction based on GFS Model Data at Inyarrimanha Ilgari Bundara, the CSIRO Murchison Radio-astronomy Observatory

Abstract

Tropospheric ducting transports terrestrial RFI over hundreds of kilometres into the ARQZWA at Inyarrimanha Ilgari Bundara, the CSIRO Murchison Radio-astronomy Observatory. We present a ducting prediction method based on vertical refractivity profiles derived from GFS data, validated against seven years of continuous spectrum monitoring between 88 MHz and 2.68 GHz. Per-transmitter basic transmission loss to the site is computed with Rec ITU-R P.452-16 above 100 MHz and Rec ITU-R P.1812-6 below it, over Copernicus GLO-90 terrain, driven at each timestep by the GFS-derived refractivity lapse along the path. Skill is scored as a stratified AUC, computed within month and three-hour strata. The median stratified AUC is 0.724 and 24 of 25 confidence intervals exclude chance. A time-percentage calibration derived from four pilot months transfers without retuning to the full record, and skill is flat through three days of forecast lead. A pooled GBM on the same physical features, augmented by a duct-corridor connectivity metric, reaches a mean test AUC of 0.806 over all 25 channels and 0.799 over the 16 on which the analytical predictor is itself skilled. In-situ LTE cell decoding confirms attribution: 86-97% of over-the-horizon cell detections fall within flagged event hours, the range spanning the four monitored LTE channels, and the decoded network identities match the licensing records. Ship AIS receptions at 162 MHz, each a self-located 12 W transmitter at a known sea position, provide VHF ground truth. The FM channels retain only weak skill, far below the ducting channels, and we attribute their events to near-threshold local sources after excluding ducting, sporadic-E and aircraft scatter. The method enables adaptive scheduling of observations away from frequencies affected by forecast ducting, and drives a live forecast service at the observatory. (Abstract abridged for arxiv)