Fade Dynamics, Availability and Measurement Limits of a Tropical Ku-Band Earth–Space Link: What a Sensor Record Supports

What does a short sensor record support, and what do its faults cost? We retained 762 h of a 919-h wet-season window (40 calendar dates) on a 12-m Ku-band dish in Bangkok; Ku-band carries Internet-of-Things backhaul. The novelty is quantifying, on this record, the limits deciding which statistics hold: sampling uncertainty, the clear-sky reference, noise floor, receiver ceiling, time base, and the exclusion of 51% of gauge-wet against 15% of gauge-dry time. Claim-bearing statistics carry day–block bootstrap intervals, with four named exceptions. Level uncertainty is 22–52%, and ceiling-censored 0.1–0.2% levels are lower bounds. A key null result: over 16 rain days, ridge regression and gradient boosting on twelve weather-station and time-of-day inputs did not improve on a rain-rate-only power law; paired 95% intervals exclude gains above 0.58 dB. Fades above 3 dB last from one 30-s sample to ≥87 min. A 3-dB margin is met for 98.4% of retained time and 81.6–98.7% of wall-clock time, depending on how excluded time is counted. Late-afternoon 3-dB exceedance tops overnight by 3.85 percentage points [1.34, 6.91]. Time base, level uncertainty and censoring leave rain-attenuation prediction error unmeasurable. Practitioners get planning inputs for comparable Ku-band gateways; absolute levels need adjustment.