Global prediction of extreme floods in ungauged watersheds.

Floods are one of the most common natural disasters, with a disproportionate impact in developing countries that often lack dense streamflow gauge networks1. Accurate and timely warnings are critical for mitigating flood risks2, but hydrological simulation models typically must be calibrated to long data records in each watershed. Here we show that artificial intelligence-based forecasting achieves reliability in predicting extreme riverine events in ungauged watersheds at up to a five-day lead time that is similar to or better than the reliability of nowcasts (zero-day lead time) from a current state-of-the-art global modelling system (the Copernicus Emergency Management Service Global Flood Awareness System). In addition, we achieve accuracies over five-year return period events that are similar to or better than current accuracies over one-year return period events. This means that artificial intelligence can provide flood warnings earlier and over larger and more impactful events in ungauged basins. The model developed here was incorporated into an operational early warning system that produces publicly available (free and open) forecasts in real time in over 80 countries. This work highlights a need for increasing the availability of hydrological data to continue to improve global access to reliable flood warnings.

In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance.

Building accurate rainfall-runoff models is an integral part of hydrological science and practice. The variety of modeling goals and applications have led to a large suite of evaluation metrics for these models. Yet, hydrologists still put considerable trust into visual judgment, although it is unclear whether such judgment agrees or disagrees with existing quantitative metrics. In this study, we tasked 622 experts to compare and judge more than 14,000 pairs of hydrographs from 13 different models. Our results show that expert opinion broadly agrees with quantitative metrics and results in a clear preference for a Machine Learning model over traditional hydrological models. The expert opinions are, however, subject to significant amounts of inconsistency. Nevertheless, where experts agree, we can predict their opinion purely from quantitative metrics, which indicates that the metrics sufficiently encode human preferences in a small set of numbers. While there remains room for improvement of quantitative metrics, we suggest that the hydrologic community should reinforce their benchmarking efforts and put more trust in these metrics.

The Efficiency of Data Assimilation.

Data assimilation is the application of Bayes' theorem to condition the states of a dynamical systems model on observations. Any real-world application of Bayes' theorem is approximate, and therefore we cannot expect that data assimilation will preserve all of the information available from models and observations. We outline a framework for measuring information in models, observations, and evaluation data in a way that allows us to quantify information loss during (necessarily imperfect) data assimilation. This facilitates quantitative analysis of tradeoffs between improving (usually expensive) remote sensing observing systems vs. improving data assimilation design and implementation. We demonstrate this methodology on a previously published application of the Ensemble Kalman Filter used to assimilate remote sensing soil moisture retrievals from AMSR-E into the Noah land surface model.

Caravan - A global community dataset for large-sample hydrology.

High-quality datasets are essential to support hydrological science and modeling. Several CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) datasets exist for specific countries or regions, however these datasets lack standardization, which makes global studies difficult. This paper introduces a dataset called Caravan (a series of CAMELS) that standardizes and aggregates seven existing large-sample hydrology datasets. Caravan includes meteorological forcing data, streamflow data, and static catchment attributes (e.g., geophysical, sociological, climatological) for 6830 catchments. Most importantly, Caravan is both a dataset and open-source software that allows members of the hydrology community to extend the dataset to new locations by extracting forcing data and catchment attributes in the cloud. Our vision is for Caravan to democratize the creation and use of globally-standardized large-sample hydrology datasets. Caravan is a truly global open-source community resource.

Bayesian analysis of the impact of rainfall data product on simulated slope failure for North Carolina locations.

In the past decades, many different approaches have been developed in the literature to quantify the load-carrying capacity and geotechnical stability (or the Factor of Safety, F s) of variably saturated hillslopes. Much of this work has focused on a deterministic characterization of hillslope stability. Yet, simulated F s values are subject to considerable uncertainty due to our inability to characterize accurately the soil mantle's properties (hydraulic, geotechnical and geomorphologic) and spatiotemporal variability of the moisture content of the hillslope interior. This is particularly true at larger spatial scales. Thus, uncertainty-incorporating analyses of physically based models of rain-induced landslides are rare in the literature. Such landslide modeling is typically conducted at the hillslope scale using gauge-based rainfall forcing data with rather poor spatiotemporal coverage. For regional landslide modeling, the specific advantages and/or disadvantages of gauge-only, radar-merged and satellite-based rainfall products are not clearly established. Here, we compare and evaluate the performance of the Transient Rainfall Infiltration and Grid-based Regional Slope-stability analysis (TRIGRS) model for three different rainfall products using 112 observed landslides in the period between 2004 and 2011 from the North Carolina Geological Survey database. Our study includes the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis Version 7 (TMPA V7), the North American Land Data Assimilation System Phase 2 (NLDAS-2) analysis, and the reference 'truth' Stage IV precipitation. TRIGRS model performance was rather inferior with the use of literature values of the geotechnical parameters and soil hydraulic properties from ROSETTA using soil textural and bulk density data from SSURGO (Soil Survey Geographic database). The performance of TRIGRS improved considerably after Bayesian estimation of the parameters with the DiffeRential Evolution Adaptive Metropolis (DREAM) algorithm using Stage IV precipitation data. Hereto, we use a likelihood function that combines binary slope failure information from landslide event and 'null' periods using multivariate frequency distribution-based metrics such as the False Discovery and False Omission Rates. Our results demonstrate that the Stage IV-inferred TRIGRS parameter distributions generalize well to TMPA and NLDAS-2 precipitation data, particularly at sites with considerably larger TMPA and NLDAS-2 rainfall amounts during landslide events than null periods. TRIGRS model performance is then rather similar for all three rainfall products. At higher elevations, however, the TMPA and NLDAS-2 precipitation volumes are insufficient and their performance with the Stage IV-derived parameter distributions indicate their inability to accurately characterize hillslope stability.

Benchmarking NLDAS-2 Soil Moisture and Evapotranspiration to Separate Uncertainty Contributions.

Model benchmarking allows us to separate uncertainty in model predictions caused by model inputs from uncertainty due to model structural error. We extend this method with a "large-sample" approach (using data from multiple field sites) to measure prediction uncertainty caused by errors in (i) forcing data, (ii) model parameters, and (iii) model structure, and use it to compare the efficiency of soil moisture state and evapotranspiration flux predictions made by the four land surface models in the North American Land Data Assimilation System Phase 2 (NLDAS-2). Parameters dominated uncertainty in soil moisture estimates and forcing data dominated uncertainty in evapotranspiration estimates; however, the models themselves used only a fraction of the information available to them. This means that there is significant potential to improve all three components of the NLDAS-2 system. In particular, continued work toward refining the parameter maps and look-up tables, the forcing data measurement and processing, and also the land surface models themselves, has potential to result in improved estimates of surface mass and energy balances.

The plumbing of land surface models: is poor performance a result of methodology or data quality?

The PALS Land sUrface Model Benchmarking Evaluation pRoject (PLUMBER) illustrated the value of prescribing a priori performance targets in model intercomparisons. It showed that the performance of turbulent energy flux predictions from different land surface models, at a broad range of flux tower sites using common evaluation metrics, was on average worse than relatively simple empirical models. For sensible heat fluxes, all land surface models were outperformed by a linear regression against downward shortwave radiation. For latent heat flux, all land surface models were outperformed by a regression against downward shortwave, surface air temperature and relative humidity. These results are explored here in greater detail and possible causes are investigated. We examine whether particular metrics or sites unduly influence the collated results, whether results change according to time-scale aggregation and whether a lack of energy conservation in flux tower data gives the empirical models an unfair advantage in the intercomparison. We demonstrate that energy conservation in the observational data is not responsible for these results. We also show that the partitioning between sensible and latent heat fluxes in LSMs, rather than the calculation of available energy, is the cause of the original findings. Finally, we present evidence suggesting that the nature of this partitioning problem is likely shared among all contributing LSMs. While we do not find a single candidate explanation for why land surface models perform poorly relative to empirical benchmarks in PLUMBER, we do exclude multiple possible explanations and provide guidance on where future research should focus.