Rain Estimation Over a Region Using Cyclegan

Sergey Timinsky, Hai Victor Habi, Jonatan Ostrometzky

Abstract: In the last couple of years, supervised machine learning (ML) methods have shown state-of-the-art results for near-ground rain estimation. Information is usually obtained from two kinds of sensors - rain gauges, which measure rain rate, and commercial microwave links (CMLs) which measure attenuation. These data sources are paired to create a dataset on which a model is trained. The arising problem of such methods of training is in the need for the datasets to be constructed with a CML-rain gauge pairing relation. In this paper, we propose a novel approach for rain estimation using a training method that does not require a matching between a CML and a rain gauge. Our goal is to infer the relation between CML measurements to rain rate values, with a data-driven approach using an unpaired dataset. We achieve this by inducing two cycle-consistency losses that capture the intuition that if we translate from attenuation measurements to rain rate observations and back again - we should arrive at where we started. Moreover, we learn two mapping functions translating between A (attenuation) and R (rain-rate), denoted by $G: \mathcal{A} \rightarrow \mathcal{R}$ and $F: \mathcal{R} \rightarrow \mathcal{A}$. No information is provided as to which sample in, $\mathcal{A}$ matches which sample in $\mathcal{R}$. We demonstrate our results using estimated accumulated rain predictions and validate them with a nearby rain gauge station.