Leveraging opportunistic rainfall sensors to improve hydrological flood modelling in a peri-urban catchment

Accurate flood simulations necessitate rainfall inputs with fine spatiotemporal resolution, especially if semi- or fully-distributed hydrological models are used. Rainfall data are commonly obtained from rain gauges and/or weather radars, each with their associated uncertainties and challenges, especially with capturing heavy, localised events, and with high implementation- and maintenance costs [1]. This further translates into high costs of hydrological modelling of flood events [2].An interesting alternative to rain gauges and radars are the rainfall data gathered from opportunistic sensors, such as Commercial Microwave Links (CMLs). CML data come at no infrastructure cost as they are generated by the network management system of mobile networks to monitor link performance. Furthermore, CMLs cover a large part of the world. Their strong potential to providing near-surface, fine-resolution rainfall fields has been demonstrated in many studies [3]. However, their usage for hydrological modelling has been little investigated so far. CML data have been mostly used for fully-distributed models in small catchments with an area of few square kilometres [1], with isolated examples of application in large catchments and/or with semi-distributed models [1],[4].In this study, we analyse the impact of various modelling decisions about application of CML rainfall data on simulated flood hydrographs. Specifically, selection of (i) the approach to pre-processing CML signals to obtain hyetographs [3], (ii) CML data usage as a standalone input or in a combination with conventional datasets, and (iii) the way to calculate sub-catchment-averaged rainfall, are analysed. Different rainfall inputs are created accordingly, and used to force a semi-distributed model of the pre-alpine, peri-urban Lambro catchment in northern Italy notorious for intensive, tightly-localised events that trigger floods [4]. The simulated hydrographs of twelve flood events are compared to the observed ones in terms of the Nash-Sutcliffe coefficient, relative errors in peak magnitudes and runoff volumes, and timing of peak occurrence. Based on our analyses, specific recommendations are provided, with the ultimate goal to promote a wider application of CML data for hydrological modelling. AcknowledgmentsThe authors would like to thank the “OpenSense” COST Action (CA20136) for supporting their collaboration through the STSM program.References[1]           J. Olsson et al., ‘How close are opportunistic rainfall observations to providing societal benefit?’, Journal of Hydrometeorology, Aug. 2025, doi: 10.1175/JHM-D-25-0043.1.[2]           J. Seibert, F. M. Clerc‐Schwarzenbach, and H. J. (Ilja) Van Meerveld, ‘Getting your money’s worth: Testing the value of data for hydrological model calibration’, Hydrological Processes, vol. 38, no. 2, p. e15094, Feb. 2024, doi: 10.1002/hyp.15094.[3]           S. C. Doshi, C. De Michele, G. Cazzaniga, and R. Nebuloni, ‘A Framework for Minimizing the Impact of Wet Antenna Attenuation on Rainfall Estimates Provided by Commercial Microwave Links’, IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing, vol. 19, pp. 421–437, 2026, doi: 10.1109/JSTARS.2025.3632933.[4]           G. Cazzaniga, C. De Michele, M. D’Amico, C. Deidda, A. Ghezzi, and R. Nebuloni, ‘Hydrological response of a peri-urban catchment exploiting conventional and unconventional rainfall observations: the case study of Lambro Catchment’, Hydrol. Earth Syst. Sci., vol. 26, no. 8, pp. 2093–2111, Apr. 2022, doi: 10.5194/hess-26-2093-2022.