Does your latency measurement account for dropped packets? Rain-caused attenuation could cause more packets to be dropped, increasing latency if you measure it as an average over total packets sent, instead of just across packets that get delivered?