This study presents the first evaluation of using commercial microwave link (CML) data for rainfall measurements in Australia, with the test site being the greater Melbourne Metropolitan area. More than 100 CMLs with microwave frequency ranging between 10 and 40 GHz have been used for the rainfall retrieval. The 15-minute received signal levels (RSLs) for each CML based on two sampling strategies (average and minimum/maximum) collected for 2 years provided a unique dataset to compare performances of rainfall retrievals. The open source algorithm RAINLINK was used for deriving rainfall from the 15-minute RSL data. From two years of data, a subset of 30 rainy days distributed across this period were used for calibrating the RAINLINK parameters, with the remaining data used for validation. For this study, only path-averaged rainfall intensities were validated based on a gauge-adjusted radar product serving as the reference. The result of the wet-dry classification showed that the minimum and maximum RSL data performed better, with lower probability of false detection and higher Matthews correlation coefficient than average RSL data. For the rainfall retrieval, both datasets showed similar correlation with the gauge adjusted radar product. However, based on other statistics (RMSE, bias and CV) minimum and maximum RSL data outperformed average for the rainfall retrieval. Overall, this study highlights the robust accuracy of commercial microwave links for rainfall retrieval while using only minimum and maximum RSL data.

Commercial microwave links (CMLs) have proven useful for providing rainfall information close to the ground surface. However, large uncertainties are associated with these retrievals, partly due to challenges in the type of data collection and processing. In particular, the most common case is when only minimum and maximum received signal levels (RSLs) over a given time interval (hereafter 15 min) are stored by mobile network operators. The average attenuation and the corresponding rainfall rate are then calculated based on a weighted average method using the minimum and maximum attenuation. In this study, an alternative to using a constant weighted average method is explored, based on a machine learning model trained to produce actual attenuation from minimum/maximum values. A rainfall retrieval deep learning model was designed based on a long short-term memory (LSTM) model architecture and trained with disdrometer data in a form that is comparable to the data provided by mobile network operators. A first evaluation used only disdrometer data to mimic both attenuation from a CML and corresponding rainfall rates. For the test data set, the relative bias was reduced from 5.99% to 2.84% and the coefficient of determination (R²) increased from 0.86 to 0.97. The second evaluation used this disdrometer-trained LSTM to retrieve rainfall rates from an actual CML located nearby the disdrometer. A significant improvement in the overall rainfall estimation compared to existing microwave link attenuation models was observed. The relative bias reduced from 7.39% to −1.14% and the R² improved from 0.71 to 0.82.

Knowledge of the full rainfall drop size distribution (DSD) is critical for characterising liquid water precipitation for applications such as rainfall retrievals using electromagnetic signals and atmospheric model parameterisation. Southern Hemisphere temperate latitudes have a lack of DSD observations and their integrated variables. Laserbased disdrometers rely on the attenuation of a beam by falling particles and are currently the most commonly used type of instrument to observe the DSD. However, there remain questions on the accuracy and variability in the DSDs measured by co-located instruments, whether identical models, different models or from different manufacturers. In this study, raw and processed DSD observations obtained from two of the most commonly deployed laser disdrometers, namely the Parsivel1 from OTT and the Laser Precipitation Monitor (LPM) from Thies Clima, are analysed and compared. Four co-located instruments of each type were deployed over 3 years from 2014 to 2017 in the proximity of Melbourne, a region prone to coastal rainfall in south-eastern Australia. This dataset includes a total of approximately 1.5 million recorded minutes, including over 40 000 min of quality rainfall data common to all instruments, equivalent to a cumulative amount of rainfall ranging from 1093 to 1244mm (depending on the instrument records) for a total of 318 rainfall events. Most of the events lasted between 20 and 40 min for rainfall amounts of 0.12 to 26.0 mm. The colocated LPM sensors show very similar observations, while the co-located Parsivel1 systems show significantly different results. The LPM recorded 1 to 2 orders of magnitude more smaller droplets for drop diameters below 0.6mm compared to the Parsivel1, with differences increasing at higher rainfall rates. The LPM integrated variables showed systematically lower values compared to the Parsivel1. Radar reflectivity- rainfall rate (ZH-R) relationships and resulting potential errors are also presented. Specific ZH-R relations for drizzle and convective rainfall are also derived based on DSD collected for each instrument type. Variability of the DSD as observed by co-located instruments of the same manufacturer had little impact on the estimated ZH-R relationships for stratiform rainfall, but differs when considering convective rainfall relations or ZH-R relations fitted to all available data. Conversely, disdrometer-derived ZH-R relations as compared to the Marshall-Palmer relation ZH D 200R1:6 led to a bias in rainfall rates for reflectivities of 50 dBZ of up to 21.6mmh1. This study provides an open-source high-resolution dataset of co-located DSD to further explore sampling effects at the micro scale, along with rainfall microstructure.