Passive microwave satellite precipitation products exhibit highly variable errors which require quantification. These errors can be attributed in part to the ice contents of precipitating systems, which greatly affects cloud radiative properties in the microwave spectrum. Accessing information on atmospheric ice contents, however, is difficult due to the relative lack of measurement techniques and the limited spatiotemporal coverage of existing methods. Since precipitation systems are a product of the large-scale environment, this information should also provide information on the ice formation processes within these systems. This study seeks to establish a link between radar-derived ice content and large-scale meteorological conditions to characterize the effects of atmospheric ice on satellite precipitation errors. Seven years of spaceborne radar and precipitation measurements from the Global Precipitation Measurement (GPM) taken over three tropical land regions were obtained to investigate this hypothesis. Five ice content regimes, as defined by ice-rain ratio (IRR), were identified, with each regime describing environments controlled by system depth, convective capacity, and the source air mass. These arguments were coupled with convective available potential energy (CAPE), total column water vapor (TCWV), and column average temperature (Tavg) information from the fifth generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5). Using these variables, it was found that the IRR regimes can be reasonably described and their error characteristics reproduced, showing that knowledge of the environment can provide similar explanatory power for satellite precipitation errors as the ice content information itself. SIGNIFICANCE STATEMENT: Satellite rainfall errors are affected by the amount of ice present in a precipitating system. Identifying the amount of ice in a system, however, is difficult due to limited and inconsistent measurements from radar. This study aims to identify more widely available information which can be used to approximate the effects of ice contents on these errors. It was found that cloud ice content varies with system strength, depth, and moisture content, which can also be identified by reanalysis model data. These data have similar effects on precipitation errors, meaning that they can be used as substitutes for cloud ice content to assess these errors.

Accurately tracking the global distribution of precipitation is essential for both research and operational meteorology. Satellite observations remain the only means of achieving consistent, global precipitation monitoring. While machine learning has long been applied to satellite-based precipitation retrieval, the absence of a standardized benchmark dataset has hindered fair comparisons between methods. To address this, the International Precipitation Working Group has developed SatRain, the first AI benchmark dataset for satellite-based detection and estimation of rain. SatRain integrates multi-sensor satellite observations from the primary platforms used in precipitation remote sensing with high-quality reference precipitation estimates derived from gauge-corrected ground-based radar composites over the conterminous United States. It offers a standardized evaluation protocol and out-of-distribution testing data from Asia and Europe to enable robust and reproducible comparisons across machine learning approaches. In addition to algorithm evaluation, the diversity of sensors and inclusion of time-resolved geostationary observations make SatRain a valuable foundation for developing next-generation AI models to deliver more accurate global precipitation estimates.

Satellite precipitation retrieval algorithms whose measurement instruments are tilted to the zenith line are subject to a spatial mismatch between the theoretical ground coordinates and the coordinate pair corresponding to the cloud layers sending spectral signals to the satellite. This is the case of the precipitation retrievals of the Global Precipitation Measurement (GPM) Microwave Imager (GMI) on board the core satellite of the GPM that uses the Goddard profil-ing algorithm (GPROF). Currently, no geometrical correction is applied to GMI retrievals of surface precipitation, creating a horizontal displacement (or parallax mismatching) between the reported surface and the corrected coordinates corresponding to the cloud structures intersecting the fi field of view. GPROF precipitation retrievals over the conterminous United States are analyzed using the ground-validated Multi-Radar Multi-Sensor (GV-MRMS) system data and the fi fifth generation European Centre for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5) temperature profiles. Results applying this parallax correction scheme show improvements in the overall retrieval accuracy of GPROF, mainly during the summer months, for every precipitation type, when the freezing level (FL) is relatively high. The development of this new parallax correction algorithm for passive microwave radiometers will significantly improve the accuracy of remote sensing data by minimizing spatial distortions in atmospheric measurements, leading to more precise weather forecasting, climate monitoring, and environmental assessments.

The water budget components of an atmospheric column are precipitation, evaporation, and horizontal water vapor divergence. This study finds that when precipitation from the Global Precipitation Climatology Project (GPCP), evaporation from the SeaFlux product, and water vapor divergence from ERA5 are employed, the degree of budget closure depends strongly on the location and time period. Variations in this error are not random, and this study seeks to better understand these biases as the climate system evolves. Errors are particularly significant over ocean regions in and near the west Pacific warm pool, where there are multiyear budget residuals of roughly 10% of the magnitude of precipitation. Biases in other tropical ocean basins are more seasonal and smaller in magnitude. Time-varying budget errors are strongly linked to variations in convective organization that vary with the large-scale environment; errors correlating with deep organized rain show coefficients of 0.62 and 0.56 in the west Pacific and central Pacific, respectively. Errors are linked to a lesser extent with cloud microphysical structures in the East Indian. Both factors affect rainfall retrievals through well-known precipitation bias mechanisms (beam-filling, convective/stratiform effects). Characteristics of the largescale environment that produce changes in convective organization and precipitable ice water content are explored.

Seasonal snow accumulation plays a fundamental role in the snow life cycle by modulating the magnitude and timing of snow ablation. Spaceborne passive microwave (PMW) retrieval algorithms offer a unique vantage point to estimate snowfall but have known issues in regions with complex terrain that lead to a poor description of snowfall within the mountains. First, using data from the western U.S. snow water equivalent (SWE) reanalysis (WUS-SR) product, we show the water year (WY) snowfall accumulation from the Multi-Radar Multi-Sensor (MRMS) system does not produce the expected magnitudes or spatial patterns of snowfall accumulation in the western United States. It was found that a domainwide scale factor of 1.5 must be applied to the MRMS to better represent WY snowfall accumulation. Second, three Goddard profiling algorithm (GPROF) models, including one Bayesian (GPROF V7) and two machine learning (GPROF-NN 1D and GPROF-NN 3D), are used to estimate snowfall. GPROF-NN 3D achieved the highest correlation of 0.62 and lowest RMSE of 0.063 mm h-1 compared to GPROF V7 with a correlation (RMSE) of 0.4 (0.072 mm h-1). The incorporation of snow climatology information into the retrieval database reduces the error in WY2016 snowfall accumulations by 16%, 34%, and 14% for GPROF V7, GPROF-NN 1D, and GPROF-NN 3D, respectively, compared to the WUS-SR. Additionally, the retrievals using the snowfall climatology show a reduction in the bias (RMSE) of up to 175 mm per WY (150 mm per WY) against independent SNOTEL observations. These results provide a positive outlook for PMW snowfall estimates in mountainous regions with a retrieval database concept that could be extended to global mountains.

Precipitation is a critical component of the terrestrial hydrological cycle. It plays a crucial role in shaping climate patterns and ecosystem dynamics. The quest for accurate measurements of global precipitation started over 150 years ago. Comprehensive evaluations of the estimates were rare until the 1970s, mainly due to the challenges of acquiring and maintaining up-to-date datasets. Nowadays, data availability is no longer an issue. However, in the face of the seemingly ever-growing number of available datasets, determining which one to rely upon poses a new obstacle, especially in the absence of global ground truth, further complicated by the interdependencies among datasets. Here, we identify the genealogy of multiple precipitation datasets and define multiplicity artifacts. Then, we compute an evaluation reference benchmark free of multiplicity artifacts to identify the dataset that best represents the artifact-free ensemble over different terrestrial spatial domains. These include countries, IPCC assessment report reference regions, major world river basins, land-cover types, elevation zones, biome categories, and K & ouml;ppen-Geiger climate classes. It should be noted that the datasets assessed herein had a monthly temporal scale, and our findings might not apply to the study of climate extreme events regardless of the terrestrial spatial domain. We repeatedly found GPM IMERG Final v07 to emerge as the most representative dataset over multiple domains. Furthermore, we found that the dataset's representativeness is largely influenced by how spatial domains are defined rather than their scale. SIGNIFICANCE STATEMENT: Over the past decades, we have amassed a vast array of precipitation datasets. While offering significant opportunities, this abundance of data creates a new challenge: which precipitation dataset should we use? In this work, we address this challenge by developing a method to identify and mitigate the impact of overlapping or dependent data sources, ensuring more reliable comparisons across diverse spatial domains. This is important because our results can help researchers select more reliable datasets and emphasize the need for a deeper understanding of the data used, thus avoiding misleading conclusions. Such efforts contribute to advancing hydrological sciences and the broader nexus fields, where reliable precipitation data serve as a cornerstone for modeling and projecting climate change impacts.

11th Workshop of the International Precipitation Working Group What: Over 130 people from 16 countries, 3 continents, and over 70 different affiliations, including government agencies, research centers, universities, and the commercial space sector discussed the current and future status of satellite-based global precipitation estimation, including applications, research, algorithms, data products, and observational systems. When: 15-18 July 2024 Where: Tokyo Institute of Technology, Tokyo, Japan

The Global Precipitation Measurement (GPM) mission was launched in February 2014 as a joint mission between JAXA from Japan and NASA from the United States. By applying the insight provided by the GPM radars, the program has contributed enormously to the quality of the passive microwave radiometer time series that now spans almost 40 years. This talk will examine the long time series of precipitation from 3 approaches. The first is an uncertainty analysis based upon first principles. It shows that time series can be homogenized, but that potential changes in convective organization over annual time scale must be included as a source of uncertainty in order to homogenize the time series of different satellites. This is verified with the second approach that focuses on closing the water budget on regional scales. While not as direct, it also hints strongly at the fact that our current time series overestimate precipitation when convection is better organized into large Mesoscale Convective Complexes. The final approach seeks to correlate biases with large scale meteorological conditions to also show that biases due to convective organization are predictable. While not applied in any product yet, this insight may serve as a blueprint for gaining confidence in our time series of precipitation where even a 1% change/decade in global precipitation is more than currently expected from observed warming trends.

Satellite passive microwave (PMW) sensors rely on the scattering signature from ice and snow particles at high frequencies (>89 GHz). The local atmospheric conditions dictate the ice particle sizes, shapes, and distribution that ultimately add uncertainty to snowfall retrieval algorithms. The Goddard Profiling Algorithm (GPROF), a widely used precipitation retrieval algorithm, uses the Multi-Radar Multi-Sensor (MRMS) system snowfall product as the reference snowfall in its a priori databases. The MRMS snowfall is estimated using a static reflectivity–snowfall ( Z – S ) relationship that does not account for local meteorological conditions. First, snowfall rates estimated from the Global Precipitation Measurement (GPM) Microwave Imager (GMI) are compared using three types of GPROF algorithms: one Bayesian (GPROF V7) and two neural network (GPROF-NN 1D and GPROF-NN 3D). GPROF-NN 3D performed the best of the three models across snowfall detection and quantitative metrics with Heidke skill scores (HSS) as high as 60% and a probability of detection of up to 70%. It is also shown that artificial biases can be introduced when evaluating model performance using a snowfall threshold to determine snowing versus nonsnowing pixels. Second, investigation into the MRMS Z – S is performed by incorporating in situ snowfall data from the U.S. Climate Reference Network (USCRN). A large range of snowfall rates was found for a given reflectivity value, and temperature is able to constrain the corresponding Z – S relationships. This dynamic temperature-dependent Z – S is found, when applied to the MRMS reflectivity, to slightly improve annual biases between MRMS and USCRN as well as GPROF snowfall detection metrics.

Accurate observations of maritime low clouds are important for air and sea transportation, understanding boundary layer processes, and measuring Earth's radiation budget. The nighttime maritime low cloud extent is often determined in meteorological satellite imagery using the brightness‐temperature difference between the longwave infrared (e.g., 11 μm) and shortwave infrared (e.g., 3.9 μm) window bands. However, this nighttime low‐cloud detection has been previously shown to be contaminated by clear‐sky false low cloud (FLC) signals associated with warm and moist air over cold regions of water. We use numerical model data and radiative transfer to quantitatively estimate the global extent and intensity of FLC signals. Insights from this research can help forecasters and researchers determine which regions and conditions are prone to nighttime FLCs and thus may require either refinement of detection algorithms or independent sensor assessments.

Abstract Meteorological satellite imagery is a critical asset for observing and forecasting weather phenomena. The Joint Polar Satellite System (JPSS) Visible Infrared Imaging Radiometer Suite (VIIRS) Day–Night Band (DNB) sensor collects measurements from moonlight, airglow, and artificial lights. DNB radiances are then manipulated and scaled with a focus on digital display. DNB imagery performance is tied to the lunar cycle, with the best performance during the full moon and the worst with the new moon. We propose using feed-forward neural network models to transform brightness temperatures and wavelength differences in the infrared spectrum to a pseudo-lunar reflectance value based on lunar reflectance values derived from observed DNB radiances. JPSS NOAA-20 and Suomi National Polar-Orbiting Partnership (SNPP) satellite data over the North Pacific Ocean at night for full moon periods from December 2018 to November 2020 were used to design the models. The pseudo-lunar reflectance values are quantitatively compared to DNB lunar reflectance, providing the first-ever lunar reflectance baseline metrics. The resulting imagery product, Machine Learning Nighttime Visible Imagery (ML-NVI), is qualitatively compared to DNB lunar reflectance and infrared imagery across the lunar cycle. The imagery goal is not only to improve upon the consistent performance of DNB imagery products across the lunar cycle, but ultimately to lay the foundation for transitioning the algorithm to geostationary sensors, making global continuous nighttime imagery possible. ML-NVI demonstrates its ability to provide DNB-derived imagery with consistent contrast and representation of clouds across the full lunar cycle for nighttime cloud detection. Significance Statement This study explores the creation and evaluation of a feed-forward neural network to generate synthetic lunar reflectance values and imagery from VIIRS infrared channels. The model creates lunar reflectance values typical of full moon scenes, enabling quantifiable comparisons for nighttime imagery evaluations. Additionally, it creates imagery that highlights low clouds better than its infrared counterparts. Results indicate the ability to create visually consistent nighttime visible imagery across the full lunar cycle for the improved nighttime detection of low clouds. Wavelengths chosen are available on both polar and geostationary satellite sensors to support the utilization of the algorithm on multiple sensor platforms for improved temporal resolution and greater simultaneous geographic coverage over polar orbiters alone.

Meteorological satellite imagery is a critical asset for observing and forecasting weather phenomena. The Joint Polar Satellite System (JPSS) Visible Infrared Imaging Radiometer Suite (VIIRS) Day-Night Band (DNB) sensor collects measurements from moonlight, airglow, and artificial lights. DNB radiances are then manipulated and scaled with a focus on digital display. DNB imagery performance is tied to the lunar cycle, with the best performance during the full moon and the worst with the new moon. We propose using feed-forward neural network models to transform brightness temperatures and wavelength differences in the infrared spectrum to a pseudo-lunar reflectance value based on lunar reflectance values derived from observed DNB radiances. JPSS NOAA-20 and Suomi National Polar-Orbiting Partnership (SNPP) satellite data over the North Pacific Ocean at night for full moon periods from December 2018 to November 2020 were used to design the models. The pseudo-lunar reflectance values are quantitatively compared to DNB lunar reflec-tance, providing the first-ever lunar reflectance baseline metrics. The resulting imagery product, Machine Learning Nighttime Visible Imagery (ML-NVI), is qualitatively compared to DNB lunar reflectance and infrared imagery across the lunar cycle. The imagery goal is not only to improve upon the consistent performance of DNB imagery products across the lunar cycle, but ultimately to lay the foundation for transitioning the algorithm to geostationary sensors, making global continuous nighttime imagery possible. ML-NVI demonstrates its ability to provide DNB-derived imagery with consistent contrast and representation of clouds across the full lunar cycle for nighttime cloud detection. SIGNIFICANCE STATEMENT: This study explores the creation and evaluation of a feed-forward neural network to generate synthetic lunar reflectance values and imagery from VIIRS infrared channels. The model creates lunar reflectance values typical of full moon scenes, enabling quantifiable comparisons for nighttime imagery evaluations. Additionally, it creates imagery that highlights low clouds better than its infrared counterparts. Results indicate the ability to create visually consistent nighttime visible imagery across the full lunar cycle for the improved nighttime detection of low clouds. Wavelengths chosen are available on both polar and geostationary satellite sensors to support the utilization of the algorithm on multiple sensor platforms for improved temporal resolution and greater simultaneous geographic coverage over polar orbiters alone.

A wide range of aerosol effects on precipitation have been proposed, from the scale of individual clouds to that of the globe.This presentation, based on the findings of an expert workshop under the umbrella of the GEWEX Aerosol Precipitation initiative, reviews the evidence and scientific consensus behind these effects and the underlying set of physical mechanisms, categorised into i) radiative effects via modification of radiative fluxes and the energy balance and ii) microphysical effects via modification of cloud droplets and ice crystals.There exists broad consensus and strong theoretical evidence that, because global mean precipitation is constrained by energetics and surface evaporation, aerosol radiative effects (aerosol-radiation interactions and aerosol-cloud interactions) act as drivers of precipitation changes. Likewise, aerosol radiative effects cause well-documented shifts of large-scale precipitation patterns, such as the Inter-Tropical Convergence Zone (ITCZ). The extent to which aerosol effects on precipitation are applicable at smaller scales and driven or buffered by compensating microphysical and dynamical mechanisms and budgetary constraints is less clear. Although there exists broad consensus and strong evidence that suitable aerosol perturbations increase cloud droplet numbers, reducing the efficiency of warm rain formation across cloud regimes, the overall aerosol effect on cloud microphysics and dynamics as well as the subsequent impact on local, regional and global precipitation is less constrained.This presentation provides a review of the physical mechanisms of aerosol effects on precipitation backed up by evidence from recent cloud-resolving and global modelling simulations as well as from satellite observations.

The Goddard Profiling Algorithm (GPROF) is used operationally for the retrieval of surface precipitation and hydrometeor profiles from the passive microwave (PMW) observations of the Global Precipitation Measurement (GPM) mission. Recent updates have led to GPROF V7, which has entered operational use in May 2022. In parallel, development is underway to improve the retrieval by transitioning to a neural-network-based algorithm called GPROF-NN. This study validates retrievals of liquid precipitation over snow-free and non-mountainous surfaces from GPROF V7 and multiple configurations of GPROF-NN against ground-based radar measurements over the conterminous United States (CONUS) and the tropical Pacific. GPROF retrievals from the GPM Microwave Imager (GMI) are validated over several years, and their ability to reproduce regional precipitation characteristics and effective resolution is assessed. Moreover, the retrieval accuracy for several other sensors of the constellation is evaluated. The validation of GPROF V7 indicates that the retrieval produces reliable estimates of liquid precipitation over the CONUS. During all four assessed years, annual mean precipitation is within 8 % of gauge-corrected radar measurements. Although biases of up to 25 % are observed over sub-regions of the CONUS and the tropical Pacific, the retrieval reliably reproduces each region's diurnal and seasonal precipitation characteristics. The effective resolution of GPROF V7 is found to be 51 km over the CONUS and 18 km over the tropical Pacific. GPROF V7 also produces robust precipitation estimates for the other sensors of the GPM constellation. The evaluation further shows that the GPROF-NN retrievals have the potential to significantly improve the GPM PMW precipitation retrievals. GPROF-NN 1D, the most basic neural network implementation of GPROF, improves the mean-squared error, mean absolute error, correlation and symmetric mean absolute percentage error of instantaneous precipitation estimates by about 20 % for GPROF GMI while the effective resolution is improved to 31 km over land and 15 km over oceans. The two GPROF-NN retrievals that are based on convolutional neural networks can further improve the accuracy up to the level of the combined radar–radiometer retrievals from the GPM core observatory. However, these retrievals are found to overfit on the viewing geometry at the center of the swath, reducing their overall accuracy to that of GPROF-NN 1D. For the other sensors of the constellation, the GPROF-NN retrievals produce larger biases than GPROF V7 and only GPROF-NN 3D achieves consistent improvements compared to GPROF V7 in terms of the other assessed error metrics. This points to shortcomings in the hydrometeor profiles or radiative transfer simulations used to generate the training data for the other sensors of the GPM constellation as a critical limitation for improving GPM PMW retrievals.

Current satellite precipitation retrievals like GPROF assume that brightness temperature is sufficient to constrain rainfall. This information, however, often represents multiple rain states, resulting in rainfall estimate uncertainties. These uncertainties, while dominated by random variability, can also exhibit substantial regional biases, complicating the use of traditional ground validation techniques which seek to understand these uncertainties. This study aims to characterize the physical contributors to these biases for use in uncertainty quantification. To do this, coincident GPROF Version 7, GMI, and GPM Combined observations were examined over three tropical land regions, the Amazon, Congo, and Southeast Asia, which are known to exhibit distinct biases relative to one another when comparing GPROF with GPM Combined. Rain intensity and ice-rain ratio were identified as the primary sources of GPROF regional biases. By incorporating the information from these sources, the self-similarity between these three regions was brought from within 13 percent to within 7 percent, reducing the interregional bias by half. Including a third constraint based on the polarization-corrected 37-GHz brightness temperature further improved this self-similarity to within 4 percent by accounting for second-order hydrometeor profile differences which were underutilized by GPROF. Comparing the effects of these three constraints between GPROF Version 7 with the 1D version of GPROF-NN showed similar improvements, indicating the utility of this uncertainty quantification and adjustment method across precipitation products. With these constraints, regional GPROF biases can be made more consistent, improving the fidelity of the precipitation climate data records and operational precipitation products which utilize this information.

Organized deep convective activity has been routinely monitored by satellite precipitation radar from the Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Mission (GPM). Organized deep convective activ-ity is found to increase not only with sea surface temperature (SST) above 27 degrees C, but also with low-level wind shear. Precip-itation shows a similar increasing relationship with both SST and low-level wind shear, except for the highest low-level wind shear. These observations suggest that the threshold for organized deep convection and precipitation in the tropics should consider not only SST, but also vertical wind shear. The longwave cloud radiative feedback, measured as the tropo-spheric longwave cloud radiative heating per amount of precipitation, is found to generally increase with stronger orga-nized deep convective activity as SST and low-level wind shear increase. Organized deep convective activity, the longwave cloud radiative feedback, and cirrus ice cloud cover per amount of precipitation also appear to be controlled more strongly by SST than by the deviation of SST from its tropical mean. This study hints at the importance of non-thermodynamic factors such as vertical wind shear for impacting tropical convective structure, cloud properties, and associated radiative energy budget of the tropics. SIGNIFICANCE STATEMENT: This study uses tropical satellite observations to demonstrate that vertical wind shear affects the relationship between sea surface temperature and tropical organized deep convection and precipitation. Shear also affects associated cloud properties and how clouds affect the flow of radiation in the atmosphere. Although how verti-cal wind shear affects convective organization has long been studied in the mesoscale community, the study attempts to apply mesoscale theory to explain the large-scale mean organization of tropical deep convection, cloud properties, and radiative feedbacks. The study also provides a quantitative observational baseline of how vertical wind shear modifies cloud radiative effects and convective organization, which can be compared to numerical simulations.

Several decades of continuous improvements in satellite precipitation algorithms have resulted in fairly accurate level-2 precipitation products for local-scale applications. Numerous studies have been carried out to quantify ran-dom and systematic errors at individual validation sites and regional networks. Understanding uncertainties at larger scales, however, has remained a challenge. Temporal changes in precipitation regional biases, regime morphology, sampling, and observation-vector information content, all play important roles in defining the accuracy of satellite rainfall retrievals. This study considers these contributors to offer a quantitative estimate of uncertainty in recently produced global precipitation climate data record. Generated from intercalibrated observations collected by a constellation of passive microwave (PMW) radiometers over the course of 30 years, this data record relies on Global Precipitation Measurement (GPM) mis-sion enterprise PMW precipitation retrieval to offer a long-term global monthly precipitation estimates with corresponding uncertainty at 5 & DEG; scales. To address changes in the information content across different constellation members the study de-velops synthetic datasets from GPM Microwave Imager (GMI) sensor, while sampling-and morphology-related uncertain-ties are quantified using GPM's dual-frequency precipitation radar (DPR). Special attention is given to separating precipitation into self-similar states that appear to be consistent across environmental conditions. Results show that the variability of bias patterns can be explained by the relative occurrence of different precipitation states across the regions and used to calculate product's uncertainty. It is found that at 5 & DEG; spatial scale monthly mean precipitation uncertainties in tropics can exceed 10%.

The Global Precipitation Measurement (GPM) mission measures global precipitation at a temporal resolution of a few hours to enable close monitoring of the global hydrological cycle. GPM achieves this by combining observations from a spaceborne precipitation radar, a constellation of passive microwave (PMW) sensors, and geostationary satellites. The Goddard Profiling Algorithm (GPROF) is used operationally to retrieve precipitation from all PMW sensors of the GPM constellation. Since the resulting precipitation rates serve as input for many of the level 3 retrieval products, GPROF constitutes an essential component of the GPM processing pipeline. This study investigates ways to improve GPROF using modern machine learning methods. We present two neural-network-based, probabilistic implementations of GPROF: GPROF-NN 1D, which (just like the current GPROF implementation) processes pixels individually, and GPROF-NN 3D, which employs a convolutional neural network to incorporate structural information into the retrieval. The accuracy of the retrievals is evaluated using a test dataset consistent with the data used in the development of the GPROF and GPROF-NN retrievals. This allows for assessing the accuracy of the retrieval method isolated from the representativeness of the training data, which remains a major source of uncertainty in the development of precipitation retrievals. Despite using the same input information as GPROF, the GPROF-NN 1D retrieval improves the accuracy of the retrieved surface precipitation for the GPM Microwave Imager (GMI) from 0.079 to 0.059 mm h−1 in terms of mean absolute error (MAE), from 76.1 % to 69.5 % in terms of symmetric mean absolute percentage error (SMAPE) and from 0.797 to 0.847 in terms of correlation. The improvements for the Microwave Humidity Sounder (MHS) are from 0.085 to 0.061 mm h−1 in terms of MAE, from 81 % to 70.1 % for SMAPE, and from 0.724 to 0.804 in terms of correlation. Comparable improvements are found for the retrieved hydrometeor profiles and their column integrals, as well as the detection of precipitation. Moreover, the ability of the retrievals to resolve small-scale variability is improved by more than 40 % for GMI and 29 % for MHS. The GPROF-NN 3D retrieval further improves the MAE to 0.043 mm h−1; the SMAPE to 48.67 %; and the correlation to 0.897 for GMI and 0.043 mm h−1, 63.42 %, and 0.83 for MHS. Application of the retrievals to GMI observations of Hurricane Harvey shows moderate improvements when compared to co-located GPM-combined and ground-based radar measurements indicating that the improvements at least partially carry over to assessment against independent measurements. Similar retrievals for MHS do not show equally clear improvements, leaving the validation against independent measurements for future investigation. Both GPROF-NN algorithms make use of the same input and output data as the original GPROF algorithm and thus may replace the current implementation in a future update of the GPM processing pipeline. Despite their superior accuracy, the single-core runtime required for the operational processing of an orbit of observations is lower than that of GPROF. The GPROF-NN algorithms promise to be a simple and cost-efficient way to increase the accuracy of the PMW precipitation retrievals of the GPM constellation and thus improve the monitoring of the global hydrological cycle.

Satellite-based oceanic precipitation estimates, particularly those derived from the Global Precipitation Measurement (GPM) satellite and CloudSat, suffer from significant disagreement over regions of the globe where warm rain processes are dominant. GPM estimates of average rain rate tend to be lower than CloudSat estimates, due in part to GPM being less sensitive to shallow and/or light precipitation. Using coincident observations between GPM and CloudSat, we find that the GPM_2BCMB product misses about two-thirds of total accumulated warm rain compared to the CloudSat 2C-RAIN-PROFILE product. This difference becomes much smaller when products are compared at 1000 m above the surface (mitigating surface clutter issues) and when forcing the frequency of rain from CloudSat to match the frequency from GPM (mitigating sensitivity issues). However, even then a gap of about 25% remains. Using an optimal estimation retrieval algorithm on the underlying data, we retrieve a similar result, but find that the remaining difference between the GPM and CloudSat retrieved rain rates can be almost entirely accounted for by inconsistent assumptions about the shape of the drop size distribution (DSD) that are made in the two retrievals. We conclude that DSD assumptions contribute significantly to the relative underestimation of warm rain by GPM compared to CloudSat. Because the choice of DSD model has such a large effect on retrieved rain rates, more work is needed to determine whether the DSD models assumed by either the GPM_2BCMB or 2C-RAIN-PROFILE algorithms are actually appropriate for warm rain.