Enhanced Runoff Modeling by Incorporating Information from the GR4J Hydrological Model and Multiple Remotely Sensed Precipitation Datasets

7
Enhanced Runoff Modeling by Incorporating Information from the GR4J Hydrological Model and Multiple Remotely Sensed Precipitation Datasets


1. Introduction

Reliable runoff modeling is essential for hydro-energy exploitation [1], water resource utilization [2,3], and sustainable water resource strategy-making [4]. The first attempts for runoff modeling were based on correlation analysis between precipitation and runoff measurements that date back to the 19th century [5,6]. Since then, runoff modeling has been progressively developed through the empirical (or experimental) formulas of subsurface physical processes and the incorporation of key physical terms to develop hydrological models [7]. Particularly, as a typical representative of hydrological models, the GR4J (modèle du Génie Rural à 4 paramètres Journalier) was developed with a continuous improvement process over 15 years based on approximately 429 catchments [8]. It is a lumped precipitation-runoff model which has a simple structure and a small number of parameters. Meanwhile, the GR4J has been widely identified to produce satisfactory realism for hydrological behavior in different topographic conditions, including plateaus [9], mountains [10], and karst areas [11].
However, besides hydrological realism, a reliable precipitation dataset is also important for runoff modeling [12]. The widely used approach for precipitation estimation is based on point-scale gauge devices [13]. Unfortunately, the gauge measurements suffer from limited spatial coverage, potential incompleteness, and missing values. More importantly, gauge measurements usually take short periods [14] or are not even recorded in remote mountainous areas. These factors cause the true precipitation field to remain difficult to retrieve. As an alternative, remotely sensed precipitation datasets (RSPDs) can be used as precipitation approximations owing to their spatiotemporal continuity and large-scale coverage [15]. These datasets are typically produced using advanced retrieval algorithms and/or assimilating remotely sensed information from multiple satellites [16,17]. Some of the state-of-the-art RSPDs include climate hazards infrared precipitation with stations (CHIRPS) [18], integrated multi-satellite retrievals for GPM (IMERG) [19], climate prediction center morphing method (CMORPH) [20], global satellite mapping of precipitation (GSMaP) [21], and precipitation estimation from remotely sensed information using artificial neural networks (PERSIANN) [22], and multi-source weather (MSWX) [23]. These precipitation datasets provide consistent and continuous precipitation estimates that have been explored and applied in runoff modeling [24,25].
Nevertheless, errors exist within the RSPDs [26], rendering runoff modeling based on a single precipitation dataset highly uncertain [27,28]. Therefore, several studies have attempted to incorporate multiple precipitation datasets into runoff modeling to improve accuracy [29,30,31], and the widely used methods include machine learning [31], Bayesian model averaging [32], deep learning [31], Kalman filter model [33], and multi-objective optimization [34]. However, there has been little research on the incorporation explorations of up to six or more kinds of RSPDs in runoff modeling simultaneously, and the potential of RSPDs still needs to be investigated. More importantly, although the aforementioned methods have obtained better performance in some regions, they are strongly based on solid mathematical assumptions and suffer from various application limitations. For instance, Bayesian statistics assume that the prior distribution is known [35], but this is often not the case in practical applications. Machine learning is weak at capturing temporal correlations and is susceptible to data noise [36]. Kalman filtering requires the input data to obey a Gaussian distribution; however, runoff data is typically distributed as a Pearson type III [37]. The reliability of deep learning relies on a large number of observations as training data, which are difficult to collect in remote mountainous regions. These application limitations often make the aforementioned methods difficult to transfer to other watersheds with various hydrometeorological features.
An application-friendly approach for incorporating information from multiple sources is the ensemble average. This approach reduces application uncertainty by assigning equal weight to its incorporation members and summing those members. The approach has been successfully practiced in tropical cyclone tracking [38], drought process reconstruction [39], and hydrological modeling under model structure (and parameters) uncertain [40,41]. However, to the best of our knowledge, there are limited studies that attempted to apply the ensemble average to incorporate information from a hydrological model and multiple RSPDs. For example, Strauch et al. [42] found that the ensemble average with multiple precipitation inputs can provide reliable deterministic streamflow estimates. In addition, a clear feature of the ensemble average approach is that it may be highly effective when its information source biases are anisotropic (e.g., overestimation and underestimation). However, it tends to be like or even worse than its incorporation membership when the source biases are isotropic. Therefore, it remains a challenge to effectively incorporate information from a hydrological model and multiple RSPDs to enhance the accuracy of runoff modeling.
To address the aforementioned concerns, this study proposes a two-step approach combining ensemble average and cumulative distribution correction (i.e., EC) to incorporate information from the GR4J hydrological model and multiple RSPDs. In the EC approach, firstly, the ensemble average is applied to construct transitional fluxes using the reproduced runoff information, which is yielded by applying various remotely sensed precipitation datasets to drive the GR4J model. Subsequently, the cumulative distribution correction is applied to enhance the transitional fluxes to model runoff. The main objectives of this study are the following: (1) to analyze the error patterns of reproduced runoff information from the GR4J model and six remotely sensed precipitation datasets (CHIRPS, IMERG, CMORPH, GSMaP, PERSIANN, and MSWX; see detail in Section 2.2); (2) to compare the performance of the EC approach with those of the single precipitation dataset-based approaches and the ensemble average under different scenarios; (3) to identify the effectiveness and transferability of the EC approach in watersheds of different hydrometeorological features. This study is expected to explore the potential of RSPDs and improve the accuracy of runoff modeling.

5. Discussion

A unique aspect of this study is the investigation of six RSPDs in runoff modeling. Therefore, we further discussed the error characteristics and value of RSPDs. The study found that the single precipitation dataset-based approaches tended to underestimate runoff, which is consistent with references [54,55]. This phenomenon may be attributed to error propagation. Specifically, RSPDs suffer from an underestimation of heavy precipitation [56], and this error propagates through the precipitation-runoff process. An indirect evidence to support the above attribution is the significant similarity between the errors in the reproduced runoff during the calibration and calibration stages (Figure 4 and Figure 5). Interestingly, this study found that the single precipitation dataset-based approaches exhibited outstanding performance in reproducing seasonal fluctuations, indicating their great value in supporting long-term applications. In addition, we found that the single precipitation dataset-based approaches have limited transferability (Figure 8). The possible reason is that there are huge regional performance differences in RSPDs [26,57]. For example, IMERG has satisfactory performance in plains but is highly uncertain in mountainous regions [57]. PERSIANN tends to be better and worse than GSMaP in China and CONUS, respectively [57,58]. The incorporation approaches (i.e., ensemble averaging and EC) exhibited better overall performance and transferability than the single precipitation dataset-based approaches (Table 4 and Figure 8). Therefore, the results of this study highlight the value of the incorporation of multiple precipitation datasets in runoff modeling.
The key novelty of this study lies in the development of a two-step approach (i.e., EC) to incorporate information from the GR4J hydrological model and multiple remotely sensed precipitation datasets. Compared to ensemble averaging, the EC method exhibited superior transferability and performance (Figure 8). The possible reason is that a distribution correction process is introduced in the EC method (Equation (2)). Specifically, all the single precipitation dataset-based approaches suffer from a similar underestimation, which might make the performance of ensemble averaging methods unstable (Figure 5 and Table 4). The distribution correction is effective in reducing systematic errors, which in turn may make the EC method perform better than the ensemble average. In addition to runoff modeling, the EC method can also be extended to near real-time runoff forecasts with short lag times consistent with precipitation datasets. Meanwhile, the EC method only relies on reproduced runoff information without increasing observed hydrogeographic data. Therefore, this methodology is valuable and replicable for other regions that are characterized by different hydrometeorological features.
Although this study provides some new insights into the hydrological application of RSPDs and the proposed EC approach has achieved reasonable performance and transferability, some limitations still need to be discussed. The assumptions within EC cover the consistency of hydrometeorological behavior across calibration and validation stages. This assumption has also been adopted in a large number of precipitation-runoff modeling studies [12,50]. However, with global industrialization and geographic resource exploitation, non-stationarity in hydrometeorological behavior has been progressively observed [59]. Severe environmental disturbances can disrupt the statistical distribution of hydrometeorological elements and the consistency of the precipitation-runoff process, which in turn may make the EC no longer applicable. Therefore, it is not realistic to extrapolate the findings of this study to other regions with severe hydrometeorological disturbances. In addition, only global precipitation datasets were considered in the study owing to the need to support transferability testing. However, regional precipitation datasets, e.g., the China Meteorological Forcing Dataset [60], tended to be more accurate in their coverage areas than global precipitation products. Therefore, if the conditions permit, more precipitation datasets can be included to further enhance runoff modeling for a special application.

Disasters Expo USA, is proud to be supported by Inergency for their next upcoming edition on March 6th & 7th 2024!

The leading event mitigating the world’s most costly disasters is returning to the Miami Beach

Convention Center and we want you to join us at the industry’s central platform for emergency management professionals.
Disasters Expo USA is proud to provide a central platform for the industry to connect and
engage with the industry’s leading professionals to better prepare, protect, prevent, respond
and recover from the disasters of today.
Hosting a dedicated platform for the convergence of disaster risk reduction, the keynote line up for Disasters Expo USA 2024 will provide an insight into successful case studies and
programs to accurately prepare for disasters. Featuring sessions from the likes of The Federal Emergency Management Agency,
NASA, The National Aeronautics and Space Administration, NOAA, The National Oceanic and Atmospheric Administration, TSA and several more this event is certainly providing you with the knowledge
required to prepare, respond and recover to disasters.
With over 50 hours worth of unmissable content, exciting new features such as their Disaster
Resilience Roundtable, Emergency Response Live, an Immersive Hurricane Simulation and
much more over just two days, you are guaranteed to gain an all-encompassing insight into
the industry to tackle the challenges of disasters.
By uniting global disaster risk management experts, well experienced emergency
responders and the leading innovators from the world, the event is the hub of the solutions
that provide attendees with tools that they can use to protect the communities and mitigate
the damage from disasters.
Tickets for the event are $119, but we have been given the promo code: HUGI100 that will
enable you to attend the event for FREE!

So don’t miss out and register today: https://shorturl.at/aikrW

And in case you missed it, here is our ultimate road trip playlist is the perfect mix of podcasts, and hidden gems that will keep you energized for the entire journey

-

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More