Abstract Accurate mapping of snow amount in the mountains is critical as mountain snowpacks are water supply for millions of people. Satellite remote sensing has been largely unable to reliably detect the amount of snowpack in these areas. Recently, C‐band Synthetic Aperture Radar (SAR) data from the Sentinel‐1 (S1) satellites have shown potential for measuring snow depth in the mountains. However, their spatiotemporal coverage is incomplete, and their evaluation with robust, aerial snow depth data is limited. Here, we evaluate two S1 snowpack datasets with some of the best available gridded snowpack data over the Colorado Rockies and Sierra Nevada mountains in the western US: the Airborne Snow Observatory (ASO) and the University of Arizona (UA) snowpack datasets. Compared to ASO and UA data, the S1 data are biased high when snow is shallow, and biased low when snow is deep (particularly later in spring when there is wet snow), though these biases are reduced for deep snow areas when wet snow pixels are removed. We then apply corrections based on machine learning that account for physiographic characteristics to improve the accuracy of the S1 data. Furthermore, we fill gaps in the S1 data by using snow persistence, but also account for potential snow accumulation and ablation, to generate temporally complete snow depth maps over mountainous areas. Corrected and gap‐filled S1 snow depth mapping could be especially important for snow monitoring in remote mountain areas where other techniques for snow mapping do not work or are logistically infeasible or cost‐prohibitive.

Several studies have demonstrated the ability of long short-term memory (LSTM) machine-learning-based modeling to outperform traditional spatially lumped process-based modeling approaches for streamflow prediction. However, due mainly to the structural complexity of the LSTM network (which includes gating operations and sequential processing of the data), difficulties can arise when interpreting the internal processes and weights in the model. Here, we propose and test a modification of LSTM architecture that is calibrated in a manner that is analogous to a hydrological system. Our architecture, called “HydroLSTM”, simulates the sequential updating of the Markovian storage while the gating operation has access to historical information. Specifically, we modify how data are fed to the new representation to facilitate simultaneous access to past lagged inputs and consolidated information, which explicitly acknowledges the importance of trends and patterns in the data. We compare the performance of the HydroLSTM and LSTM architectures using data from 10 hydro-climatically varied catchments. We further examine how the new architecture exploits the information in lagged inputs, for 588 catchments across the USA. The HydroLSTM-based models require fewer cell states to obtain similar performance to their LSTM-based counterparts. Further, the weight patterns associated with lagged input variables are interpretable and consistent with regional hydroclimatic characteristics (snowmelt-dominated, recent rainfall-dominated, and historical rainfall-dominated). These findings illustrate how the hydrological interpretability of LSTM-based models can be enhanced by appropriate architectural modifications that are physically and conceptually consistent with our understanding of the system.

Remote sensing-based precipitation products face several challenges in high latitudes and specifically over frozen surfaces (i.e., snow and ice). Consequently, precipitation estimates tend to be lower in quality over these regions, including Antarctica, the coldest continent on Earth. In this study, we developed a method for adjusting precipitation estimates over Antarctica by leveraging CloudSat's ability to capture snowfall compared to other satellite products over snow and ice surfaces. We addressed limitations of CloudSat, such as poor spatiotemporal sampling, noise, and incomplete coverage near the poles. We utilized the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis 5th Generation (ERA5) to guide development, particularly during the period when the TIROS Operational Vertical Sounder (TOVS) was contributing. We adjusted monthly and seasonal TOVS and Atmospheric Infrared Sounder – Infrared (AIRS-IR) precipitation biases over Antarctica at the pixel level, as these two products are the primary inputs for the Global Precipitation Climatology Project (GPCP) in high latitudes. We assessed the adjusted TOVS and AIRS-IR through analyses of geographical maps and time series of monthly and seasonal mean precipitation rates. The results are encouraging, indicating that the proposed approach could replace the current approach used in the GPCP for adjusting precipitation estimates from AIRS-IR and TOVS over Antarctica. Adjusting precipitation estimates from TOVS and AIRS-IR using the proposed approach improves the Kling-Gupta efficiency (KGE) over their entire period by 162% and 147%, respectively. Moreover, the proposed approach can be applied to adjust other precipitation products over Antarctica at the pixel level, including satellites and reanalysis products.

Abstract Snowfall forecasting has historically been an area of difficulty for operational meteorologists, particularly in regions of complex terrain, such as the western United States. Attempts at improving forecasts have been made, but skill is still poor, with snowfall routinely overpredicted. A major reason for this overprediction has been the failure to accurately predict snow–liquid ratios (SLR) ahead of major events. This research proposes, develops, and tests multiple machine learning methods for dynamic SLR prediction for the Sky Islands of southeast Arizona by objectively comparing a multiple linear regression (MLR) against several more complex and flexible machine learning methods. Input parameters for each method were chosen based on variables found by previous studies to have a regression‐based relationship with SLR, with a focus on the lower mid‐levels of the troposphere. These parameters were also used to construct the MLR model, and its performance was compared objectively with the machine learning methods. When tested on historical events, a very high percentage of the network‐predicted SLR values fall within the margin of error of observed SLRs, which were calculated using gridded snow depth and snow water equivalent (SWE) data from the University of Arizona daily 4‐km SWE, SD, and SCE dataset (UASnow). A support vector machine (SVM), a k‐nearest neighbor (KNN) algorithm, and a random forest also showed high accuracies when tested on the dataset, and each showed a significant gain in skill compared with the MLR model, with skill being evaluated by multiple metrics.

The Global Precipitation Climatology Project (GPCP) Version 3.2 Precipitation Analysis provides globally complete analyses of surface precipitation on a 0.5 degrees 3 0.5 degrees latitude-longitude grid at both monthly and daily time scales, covering from 1983 to the present and from June 2000 to the present, respectively. These merged products continue the GPCP heritage of incorporating precipitation estimates from low-orbit satellite microwave data, geosynchronous-orbit sat-ellite infrared data, sounder-based estimates, and surface rain gauge observations emphasizing the strengths of various in-puts and striving for time and space homogeneity. Furthermore, these analyses incorporate modern algorithms, refined intercalibrations among sensors, climatologies of recent high-quality satellite precipitation data, and fine-scale multisatellite estimates. New data fields have been introduced to better characterize the precipitation, including the fraction of the pre-cipitation that is liquid (rain) in both the monthly and daily products, and a quality index for the monthly product. Com-pared to the operational GPCP Version 2.3 Monthly, the Version 3.2 Monthly product provides a more reasonable climatology in the Southern Ocean and increases the estimated global average precipitation by about 4.5%, which is similar to estimates from recent global water budget assessments. Global and regional trends for 1983-2020 with this new Monthly dataset are very similar to those computed from Version 2.3. Compared to the operational One-Degree Daily (Version 1.3) product, the new Version 3.2 Daily is designed to better represent the histogram of precipitation rates, particularly at high values and shifts the start of less-certain high-latitude estimates from 40 degrees to 58 degrees latitude in each hemisphere.SIGNIFICANCE STATEMENT: Studies of Earth's climate require long-term global datasets based on observations to show how the climate functions and to validate numerical climate models. This study describes an important upgrade to the monthly and daily precipitation (rain and snow) products computed by the Global Precipitation Climatology Project. We use modern analysis schemes, add new sources of data, and deliver results on a finer-scale 0.5 degrees 3 0.5 degrees latitude- longitude grid [roughly 55 km (34 mi) on a side at the equator]. The new data show improved agreement with other studies and depict more reasonable behavior in the Southern Ocean. The daily product shows improved estimates of how often different intensities of precipitation occur around the world, particularly the high amounts that drive floods and landslides.

Access to spatiotemporal distribution of precipitation is needed in many hydrological applications. However, gauges often have spatiotemporal gaps. To mitigate this, we considered three main approaches: (i) using remotely sensing and reanalysis precipitation products; (ii) machine learning-based approaches; and (iii) a gap-filling software explicitly developed for filling the gaps of daily precipitation records. This study evaluated all approaches over a sparsely gauged basin in East Africa. Among the examined precipitation products, PERSIANN-CDR outperformed other satellite products in terms of root mean squared error (7.3 mm), and correlation coefficient (0.46) while having a large bias (50%) compared to the available in situ precipitation records. PERSIANN-CDR also demonstrates the highest skill in distinguishing rainy and non-rainy days. On the other hand, Random Forest outperformed all other approaches (including PERSIANN-CDR) with the least relative bias (-2%), root mean squared error (6.9 mm), and highest correlation coefficient (0.53).

ECMWF Reanalysis (ERA), one of the most widely used precipitation products evolved over time from ERA-40 to ERA-20CM, ERA-20C, ERA-Interim, and ERA5. Studies evaluating the performance of individual ERA products cannot adequately assess the evolution in the products. We compared the performance of all ERA precipitation products using daily, monthly, and annual data (1980-2018) from more than 2100 Iran precipitation gauges. Results indicated that ERA-40 performed worst, followed by ERA-20CM, which showed only minor improvements over ERA-40. ERA-20C considerably outperformed its predecessors, benefiting from assimilation of observational data. Although several previous studies have reported full superiority of ERA5 over ERA-Interim, our results revealed several shortcomings compared with ERA-Interim in ERA5 estimates. Both ERA-Interim and ERA5 performed best overall, with ERA-Interim showing better statistical and categorical skill scores, and ERA5 performing better in estimating extreme precipitations. These results suggest that the accuracy of ERA precipitation products improved from ERA-40 to ERA-Interim, but not consistently from ERA-Interim to ERA5. This study employed a grid-grid comparison approach by firstly creating a gridded reference dataset through spatial aggregation of point source observations, however, the results from a point-grid approach show no change in the overall ranking of products (despite the existence of slight changes in the error index values). These findings are useful for model development at global scale and for hydrological applications in Iran.

The Global Precipitation Climatology Project (GPCP) product is a popular combined satellite-gauge precipitation data set in which the long-term standards of consistency and homogeneity are underlined. Here we discuss various high latitude analyses considered in the recently released GPCP V3.2 monthly and daily products. Satellite data are used over land and ocean and obtained from the Special Sensor Microwave Imager (SSMI), Special Sensor Microwave Imager/Sounder (SSMIS), geostationary imagers and polar-orbiting infrared sounders. GPCP uses the Global Precipitation Climatology Centre (GPCC) over land, as its in situ component, but prior to combination with satellite data GPCC estimates are adjusted for gauge undercatch. Advanced sensors aboard the Tropical Rainfall Measuring Mission (TRMM), CloudSat, and Global Precipitation Measurement (GPM) mission have enabled more accurate estimation of rain and snowfall rates in recent years. Starting with GPCP V3.1 these observations are integrated into GPCP through the development of the Tropical Combined Climatology (TCC) used at lower latitudes and the Merged CloudSat, TRMM, and GPM (MCTG) climatology used over the extratropics and higher latitudes. Improved calibrations of Television-Infrared Operational Satellite (TIROS) Operational Vertical Sounder (TOVS) and Advanced Infrared Sounder (AIRS) precipitation are used outside 60ºN-S, where inside this zone the Goddard Profiling (GPROF) algorithm retrievals from SSMI/SSMIS is used to calibrate geostationary IR based precipitation estimate at monthly scale. The Gravity Recovery and Climate Experiment (GRACE) mass change observations are used to determine snowfall accumulations over frozen land and arctic basins and to assess gauge undercatch corrections. Observations of snow on sea ice from NASA’s Operation IceBridge (OIB) flights are utilized as an additional tool for snowfall assessment over sea ice. GPCP V3.2 has a higher spatial resolution (0.5ox0.5o) than earlier versions (e.g., 2.5ox2.5o in V2.3) over both land and ocean, going back to 1983. Version 3 Daily product uses the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (GPM) mission (IMERG) Final Run V06 estimates as well as rescaled TOVS/AIRS data in high-latitude areas, all calibrated to the GPCP V3.2 Monthly estimate. GPCP V3.2 shows about 5.5% increase in global oceanic precipitation and about 4 % increase over global land and ocean compared to the previous version (V2.3), some major changes occur over the ocean and around 40oS and 60 oS. We will discuss other important changes of GPCP V3.2, compared to the earlier versions, and our future plans. This includes a discussion of some challenges that the team had to deal with, such as consistencies in inter-annual variations of satellite precipitation products and modification of gauge undercatch correction methods.

We examine the ability of machine learning (ML) and deep learning (DL) algorithms to infer surface/ground exchange flux based on subsurface temperature observations. The observations and fluxes are produced from a high-resolution numerical model representing conditions in the Columbia River near the Department of Energy Hanford site located in southeastern Washington State. Random measurement error, of varying magnitude, is added to the synthetic temperature observations. The results indicate that both ML and DL methods can be used to infer the surface/ground exchange flux. DL methods, especially convolutional neural networks, outperform the ML methods when used to interpret noisy temperature data with a smoothing filter applied. However, the ML methods also performed well and they are can better identify a reduced number of important observations, which could be useful for measurement network optimization. Surprisingly, the ML and DL methods better inferred upward flux than downward flux. This is in direct contrast to previous findings using numerical models to infer flux from temperature observations and it may suggest that combined use of ML or DL inference with numerical inference could improve flux estimation beneath river systems.

Assessment of the precipitation products with ground-based data is essential to building confidence in these datasets. Precipitation products tend to have large errors in semi-arid regions such as the Southwest United States, where accurate precipitation quantification is critical for water resource management and flood mitigation measures. Therefore, this region, with its high density of ground-based data, is important for the evaluation of the products. The Southwest United States is also interesting due to its monsoonal precipitation pattern, in which changes in circulation patterns that bring tropical moisture to the region yield roughly 50% of the region's precipitation between the months of June–September. In the present study, the performance of precipitation products was evaluated over Arizona and New Mexico for the monsoon seasons of the 2002–2021 period, with an emphasis on the recent extreme years of 2020 and 2021. Results indicate that all satellite products tend to capture interannual variations of precipitation rate but struggled to capture high-intensity events. IMERG Final notably has better performance than other products, with the lowest root mean square error and highest correlation with Stage IV, which was 3–60 percent better than other products. IMERG Final had the best detection capacity for rainy days as well. ERA5-Land performed well in capturing the average monsoon precipitation rate; however, showed limited skill in the detection of trace, light, and extreme precipitation events. IMERG Late and PDIR-Now showed difficulty detecting light precipitation events and overestimated extreme events. This study shows the importance of gauge adjustment for satellite products (e.g., IMERG) as well as the need for improvement of reanalysis products over arid regions, and better representation of orographic precipitation.

•Global land precipitation can be underestimated by up to ∼ 10% without applying correction factors compensating for systematic errors in gauges.•There is a ∼ 4 % difference in the global land precipitation estimate based on the method used for gauge-undercatch correction.•Differences between CFs can be traced through environmental conditions such as temperature and relative humidity.

Accurate and timely estimation of precipitation is critical for issuing hazard warnings (e.g., for flash floods or landslides). Current remotely sensed precipitation products have a few hours of latency, associated with the acquisition and processing of satellite data. By applying a robust nowcasting system to these products, it is (in principle) possible to mitigate this latency and improve their applicability, value, and impact. However, the development of such a system is complicated by the chaotic nature of the atmosphere, lack of sufficient knowledge about the evolution of precipitation systems based on previous observations, and the consequent rapid changes that can occur in the structures of precipitation systems. In this work, we develop two approaches (hereafter referred to as NowCasting-nets ) that use recurrent and convolutional deep neural network (DNN) structures to address the challenge of precipitation nowcasting. A total of five models are trained using global precipitation measurement (GPM) Integrated MultisatellitE Retrievals for GPM (IMERG) precipitation data over the Eastern contiguous United States (CONUS) and then tested against independent data for the Eastern and Western CONUS. The models were designed to provide forecasts with a lead time of up to 1.5 h, and by using a feedback loop approach, the ability of the models to extend the forecast time to 4.5 h was also investigated. The performance of the models was compared against the random forest (RF) and linear regression (LR) machine learning (ML) methods, a persistence benchmark (BM) that uses the most recent observation as the forecast, and optical flow (OF). Independent IMERG observations were used as a reference, and experiments were conducted to examine both overall statistics and case studies involving specific precipitation events. Overall, the forecasts provided by the NowCasting-net models are superior, with the convolutional NowCasting-net (CNC) achieving 42%, 24%, 18%, and 16% improvement on the test set mean squared error (MSE) over the BM, LR, RF, and OF models, respectively, for the Eastern CONUS. Results of further testing over the Western CONUS (which was not part of the training data) are encouraging and indicate the ability of the proposed models to learn the dynamics of precipitation systems without having explicit access to motion vectors and other auxiliary features and then to generalize to different hydro-geo-climatic conditions.

We confirm that energy dissipation weighting provides the most accurate approach to determining the effective hydraulic conductivity (K-eff) of a binary K grid. A deep learning algorithm (UNET) can infer K-eff with extremely high accuracy (R-2 > 0.99). The UNET architecture could be trained to infer the energy dissipation weighting pattern from an image of the K distribution, although it was less accurate for cases with highly localized structures that controlled flow. Furthermore, the UNET architecture learned to infer the energy dissipation weighting even if it was not trained directly on this information. However, the weights were represented within the UNET in a way that was not immediately interpretable by a human user. This reiterates the idea that even if ML/DL algorithms are trained to make some hydrologic predictions accurately, they must be designed and trained to provide each user-required output if their results are to be used to improve our understanding of hydrologic systems.

Precipitation retrieval is a challenging topic, especially in high latitudes (HL), and current precipitation products face ample challenges over these regions. This study investigates the potential of the Advanced Very High-Resolution Radiometer (AVHRR) for snowfall retrieval in HL using CloudSat radar information and machine learning (ML). With all the known limitations, AVHRR observations should be considered for HL snowfall retrieval because (1) AVHRR data have been continuously collected for about four decades on multiple platforms with global coverage, and similar observations will likely continue in the future; (2) current passive microwave satellite precipitation products have several issues over snow and ice surfaces; and (3) good coincident observations between AVHRR and CloudSat are available for training ML algorithms. Using ML, snowfall rate was retrieved from AVHRR’s brightness temperature and cloud probability, as well as auxiliary information provided by numerical reanalysis. The results indicate that the ML-based retrieval algorithm is capable of detection and estimation of snowfall with comparable or better statistical scores than those obtained from the Atmospheric Infrared Sounder (AIRS) and two passive microwave sensors contributing to the Global Precipitation Measurement (GPM) mission constellation. The outcomes also suggest that AVHRR-based snowfall retrievals are spatially and temporally reasonable and can be considered as a quantitatively useful input to the merged precipitation products that require frequent sampling or long-term records.

Precipitation gauges are critical for measuring precipitation rates at regional and global scales and are often used to calibrate precipitation rates estimated from other instruments such as satellites. However, precipitation measured at the gauges is affected by gauge-undercatch that is often larger for solid precipitation. In the present work, two popular gauge-undercatch correction factors are assessed: one utilizes a dynamic correction model and is used in the Global Precipitation Climatology Centre (GPCC) Monitoring product and the other one employs a fixed climatology and is used in the Global Precipitation Climatology Project (GPCP) product. How much the choice of correction factors can impact the total estimate of precipitation was quantified over land at seasonal, annual, regional, and global scales. The correction factors are also compared as a function of the environmental variables used in their development, among those are near-surface air temperature, relative humidity, wind speed, elevation, and precipitation intensity. Results show that correction factors can increase the annual precipitation rate based on the gauges by ~9.5 % over the global land (excluding Antarctica), although this amount can vary from ~6.3% (in boreal summer) to more than 10% (in boreal winter), depending on the season and the method used for gauge-undercatch correction. Annual variations of correction factors can also be large, so the use of the fixed climatology correction factors requires caution. Given their magnitudes and differences, selection of appropriate correction factors can have important implications in refining the water and energy budget calculations.

The combination of snowfall, snow water equivalent (SWE), and precipitation rate measurements from 39 snow telemetry (SNOTEL) sites in Alaska were used to assess the performance of various precipitation products from satellites, reanalysis, and rain gauges. Observation of precipitation from two water years (2018–2019) of a high-resolution radar/rain gauge data (Stage IV) product was also utilized to give insights into the scaling differences between various products. The outcomes were used to assess two popular methods for rain gauge undercatch correction. It was found that SWE and precipitation measurements at SNOTELs, as well as precipitation estimates based on Stage IV data, are generally consistent and can provide a range within which other products can be assessed. The time-series of snowfall and SWE accumulation suggests that most of the products can capture snowfall events; however, differences exist in their accumulation. Reanalysis products tended to overestimate snow accumulation in the study area, while the current combined passive microwave remote sensing products (i.e., IMERG-HQ) underestimate snowfall accumulation. We found that correction factors applied to rain gauges are effective for improving their undercatch, especially for snowfall. However, no improvement in correlation is seen when correction factors are applied, and rainfall is still estimated better than snowfall. Even though IMERG-HQ has less skill for capturing snowfall than rainfall, analysis using Taylor plots showed that the combined microwave product does have skill for capturing the geographical distribution of snowfall and precipitation accumulation; therefore, bias adjustment might lead to reasonable precipitation estimates. This study demonstrates that other snow properties (e.g., SWE accumulation at the SNOTEL sites) can complement precipitation data to estimate snowfall. In the future, gridded SWE and snow depth data from GlobSnow and Sentinel-1 can be used to assess snowfall and its distribution over broader regions.

We develop a simple Quantile Spacing (QS) method for accurate probabilistic estimation of one-dimensional entropy from equiprobable random samples, and compare it with the popular Bin-Counting (BC) and Kernel Density (KD) methods. In contrast to BC, which uses equal-width bins with varying probability mass, the QS method uses estimates of the quantiles that divide the support of the data generating probability density function (pdf) into equal-probability-mass intervals. And, whereas BC and KD each require optimal tuning of a hyper-parameter whose value varies with sample size and shape of the pdf, QS only requires specification of the number of quantiles to be used. Results indicate, for the class of distributions tested, that the optimal number of quantiles is a fixed fraction of the sample size (empirically determined to be ~0.25–0.35), and that this value is relatively insensitive to distributional form or sample size. This provides a clear advantage over BC and KD since hyper-parameter tuning is not required. Further, unlike KD, there is no need to select an appropriate kernel-type, and so QS is applicable to pdfs of arbitrary shape, including those with discontinuous slope and/or magnitude. Bootstrapping is used to approximate the sampling variability distribution of the resulting entropy estimate, and is shown to accurately reflect the true uncertainty. For the four distributional forms studied (Gaussian, Log-Normal, Exponential and Bimodal Gaussian Mixture), expected estimation bias is less than 1% and uncertainty is low even for samples of as few as 100 data points; in contrast, for KD the small sample bias can be as large as −10% and for BC as large as −50%. We speculate that estimating quantile locations, rather than bin-probabilities, results in more efficient use of the information in the data to approximate the underlying shape of an unknown data generating pdf.