Abstract. Rainfall monitoring is an important global issue in urban hydrological applications, such as flood warning and water resources management systems. Until the present time, rain gauges and weather radars have been widely used as sensors to provide rainfall information with a detailed resolution; most cities in the world however are inadequately equipped. Recently, commercial microwave links (MWL) have been proposed as a new means of monitoring space-time rainfall. A transmitted signal along such links is known to be attenuated by rainfall, hence the measurement of this signal attenuation could serve to estimate path-averaged rainfall intensity. The density of commercial MWL is typically high in most cities today, which raises new questions over the possibility of retrieving rainfall using signal attenuation data from multiple links. The objective of this article is to assess the feasibility of retrieving rainfall fields in urban areas using rain attenuation data from commercial MWL that are mainly operated by mobile phone companies. This work is based on a simulation framework applied to a real case study. The study area is the city of Nantes, France. Rainfall datasets containing 207 weather radar images recorded by the Météo-France Agency's C-band at high spatial (250 m × 250 m) and temporal (5 min) resolutions are first used to generate rain attenuation data over the existing mobile phone network, which combines 256 microwave links operating at 18, 23 and 38 GHz. The rain attenuation data generated are used as a real signal dataset. A novel retrieval algorithm is then proposed to convert the rain-induced attenuation data into a rainfall map. A priori knowledge introduced to initialize the algorithm heavily influences retrieval performance if the problem to be solved is under-determined, as is the case herein.

The capabilities as well as limitations of the retrieval algorithm, as regards capturing different rainfall variability, are evaluated. A detailed sensitivity analysis, carried out with respect to various parameters including a priori knowledge, decorrelation distance, and the retrieval performance of the algorithm depending on the density level of the MWL network is also evaluated in a light rain, a shower and amidst storm events. The conclusion, based on 200+ retrieval tests, states that the proposed algorithm is capable of capturing high rainfall variability in the presence of large measurement error sources according to the adopted methodology.