The usefulness of satellite multisensor precipitation products such as NASA’s 30-minute, 0.1° Integrated Multi-satellitE Retrievals for the Global Precipitation Mission (IMERG) is hindered by their associated errors. Reliable estimates of uncertainty would mitigate this limitation, especially in near-real time. Creating such estimates is challenging, however, due both to the complex discrete-continuous nature of satellite precipitation errors and to the lack of “ground truth” data precisely in the places—including complex terrain and developing countries—that could benefit most from satellite precipitation estimates. In this work, we use swath-based precipitation products from the Global Precipitation Mission (GPM) Dual-frequency Precipitation Radar (DPR) as an alternative to ground-based observations to facilitate IMERG uncertainty estimation. We compare the suitability of two DPR derived products, 2ADPR and 2BCMB, against higher-fidelity Ground Validation Multi-Radar Multi-Sensor (GV-MRMS) ground reference data over the contiguous United States. 2BCMB is selected to train mixed discrete-continuous error models based on Censored Shifted Gamma Distributions. Uncertainty estimates from these error models are compared against alternative models trained on GV-MRMS. Using information from NASA’s Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis, we also demonstrate how IMERG uncertainty estimates can be further constrained using additional precipitation-related predictors. Though several critical issues remain unresolved, the proposed method shows promise for yielding robust uncertainty estimates in near-real time for IMERG and other similar precipitation products at their native resolution across the entire globe.

Existing global FT records, derived from the Soil Moisture Active Passive (SMAP), the Advanced Scanning Microwave Radiometer (AMSR), and the Special Sensor Microwave Imager (SSM/I) produce relatively course spatial resolution (25-36km) binary FT classifications. These classifications can vary widely depending on the microwave bands used, topography, and land cover, leading to a somewhat ambiguous definition of ‘frozen’ and ‘thawed’ states. In this study, we assess the relationship between satellite observation derived FT products over North America compared to modeled near-surface temperatures and land surface temperature (LST) from the Geostationary Operational Environmental Satellite system (GOES). Utilizing the higher spatial resolution of these products (~4.5km), sub-grid scale variability and its relationship to courser microwave FT classifications was assessed. Through an analysis of spatial variability and uncertainty across North America, five focus study pixels each representing unique FT profiles were examined. These included pixels in: (1) Southern Plains (36, -97), (2) Tundra (61, -76), (3) Northern Forest (47, -74), (4) Northern Plains (52, -103), and Mountainous (38.9, -107.9). The model ensemble adequately captured near surface temperatures as they relate to FT classifications in Tundra, the Northern Plains, and Northern Forest regions. On average, 85.3% to 99.6% of sub-grid cells were below freezing when FT products classified the associated pixels as frozen. GOES - LST observations were shown to have the highest proportion of sub-grid cells below freezing on average, when classified as frozen by FT products (97.3% - 100%) across the same 3 focus locations. However, we also find that fractional FT products utilizing higher resolution data inputs, such as LST, would provide a considerable improvement in mountainous regions with high inter-grid cell heterogeneity, in regions characterized by ephemeral FT events (Southern Plains), as well as during freeze and thaw onset periods. These locations showed a significant reduction in the average temperature product frozen proportion associated with frozen classifications (as low as 5.8%). This study provides insight to improving representation of FT state and providing a clearer meaning of what constitutes a ‘frozen’ classification.

In this study, we have developed a hyper-resolution land-surface forcing dataset (temperature, pressure, humidity, wind speed, incident longwave and shortwave radiation) from coarse resolution products using a physically-based downscaling approach. These downscaling techniques rely on correlations with landscape variables, such as topography, temperature lapse rate corrections, surface roughness and land cover. A proof-of-concept has been implemented over the Oklahoma domain, where high-resolution observations are available for validation purposes. The hourly NLDAS (North America Land Data Assimilation System) forcing data at 0.125° have been downscaled to 500m resolution over the study area during 2015. Results show that correlation coefficients between the downscaled forcing dataset and ground observations are consistently higher and biases are lower than the ones between the NLDAS forcing dataset at their native resolution and ground observations. Results are therefore encouraging as they demonstrate that the 500m forcing dataset has a good agreement with the ground information and can be adopted to force the land surface model for land state estimation. The Noah-MP land surface model is then forced with both the native resolution NLDAS dataset and the downscaled one to simulate surface and root zone soil moisture. Model outputs are compared with in situ soil moisture observations and SMAP (Soil Moisture Active Passive Mission) products at different spatial resolutions. This work will result in a radical improvement over the current state-of-the-art forcing data and will move into the era of hyper-resolution land modeling.

An accurate characterization of the global hydrologic cycle is essential not only to study and forecast climate variations, but also for extreme event mitigation and agricultural planning. Since precipitation is the major driving force of the hydrological cycle, current and future satellite missions are critical to estimate precipitation globally. Error estimates associated with satellite precipitation retrievals are crucial to allow inferences about the reliability of such products in their operational applications. However, evaluating satellite precipitation error characteristics is challenging because of the inherent temporal and spatial variability of precipitation, measurement errors, and sampling uncertainties, especially at fine temporal and spatial resolutions. This study proposes to use a stochastic error model – PUSH (Probability Uncertainty in Satellite Hydrology) – for estimating uncertainties associated with fine resolution satellite precipitation products. The framework is tested on the daily IMERG (Integrated Multi-satellitE Retrievals for GPM) infrared-only (IR) precipitation component using a satellite-based radar product (the Level-3 Dual-frequency Precipitation Radar, 3DPRD) as reference. PUSH decomposes the error into four components and employs different modeling approaches for each case: correct no-precipitation detection; missed precipitation; false alarm; hit bias. PUSH is calibrated globally over land for different climatological regions. The calibrated parameters are validated using an independent period to verify whether they can be applied to estimate uncertainties associated with future IR retrievals without degrading the model performance. The four error components are then investigated as a function of climate region to study their spatial variability.

The accurate representation of the local-scale variability of precipitation plays an important role in understanding the hydrological cycle and land-atmosphere interactions in the High Mountain Asia region. Therefore, the development of hyper-resolution precipitation data is of urgent need. In this study, we propose a statistical framework to downscale the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) precipitation product using the random forest classification and regression algorithm. A set of variables representing atmospheric, geographic, and vegetation cover information are selected as model predictors, based on a recursive feature elimination method. The downscaled precipitation product is validated in terms of magnitude and variability against a set of ground- and satellite-based observations. Results suggest improvements with respect to the original resolution MERRA-2 precipitation product and comparable performance with gauge-adjusted satellite precipitation products.

The usefulness of satellite multi-sensor precipitation (SMP) and other satellite-informed precipitation products in water resources modeling can be hindered by substantial errors which vary considerably with spatiotemporal scale. One approach to cope with these errors is by combining SMPs with ensemble generation methods, such that each ensemble member reflects one plausible realization of the true—but unknown—precipitation. This requires replicating the spatiotemporal autocorrelation structure of SMP errors. The climatology of this structure is unknown for most locations due to a lack of ground reference observations, while the unique anisotropy and nonstationarity within any particular precipitation system limit the relevance of this climataology to the depiction of error in individual storm systems. Characterizing and simulating this autocorrelation across spatiotemporal scales has thus been called a grand challenge within the precipitation community. We introduce the Space-Time Rainfall Error and Autocorrelation Model (STREAM), which combines anisotropic and nonstationary SMP spatiotemporal correlation structures with a pixel-scale precipitation error model to stochastically generate ensemble precipitation fields that resemble “ground truth” precipitation. We generate STREAM precipitation ensembles at high resolution (1-hour, 0.1˚) with minimal reliance on ground-reference data, and evaluate these ensembles at multiple scales. STREAM ensembles consistently “bracket” ground-truth observations and replicate the autocorrelation structure of ground-truth precipitation fields. STREAM is compatible with pixel-scale error/uncertainty formulations beyond those presented here, and could be applied globally to other precipitation sources such as numerical weather predictions or “blended” products. In combination with recent work in SMP uncertainty characterization, STREAM could be run without any ground data.

This first paper of the two-part series focuses on demonstrating the predictability of a hyper-resolution, offline terrestrial modeling system used for the High Mountain Asia (HMA) region. To this end, this study systematically evaluates four sets of model simulations at point scale, basin scale, and domain scale obtained from different spatial resolutions including 0.01 degree (∼ 1-km) and 0.25 degree (∼ 25-km). The assessment is conducted via comparisons against ground-based observations and satellite-derived reference products. The key variables of interest include surface net shortwave radiation, surface net longwave radiation, skin temperature, near-surface soil temperature, snow depth, snow water equivalent, and total runoff. In the evaluation against ground-based measurements, the superiority of the 0.01 degree estimates are mostly demonstrated across relatively complex terrain. Specifically, hyper-resolution modeling improves the skill in meteorological forcing estimates (except precipitation) by 9% relative to coarse-resolution estimates. The model forced by downscaled forcings in its entirety yields the highest predictability skill in model output states as well as precipitation, which improves the skill obtained by coarse-resolution estimates by 7%. These findings, on one hand, corroborate the importance of employing the hyper-resolution versus coarse-resolution modeling in areas characterized by complex terrain. On the other hand, by evaluating four sets of model simulations forced with different precipitation products, this study emphasizes the importance of accurate hyper-resolution precipitation products to drive model simulations.