Physically based distributed hydrological model calibration based on a short period of streamflow data: case studies in four Chinese basins

Abstract

Abstract. Physically based distributed hydrological models are widely used for hydrological simulations in various environments. As with conceptual models, they are limited in data-sparse basins by the lack of streamflow data for calibration. Short periods of observational data (less than 1 year) may be obtained from fragmentary historical records of previously existing gauging stations or from temporary gauging during field surveys, which might be of value for model calibration. However, unlike lumped conceptual models, such an approach has not been explored sufficiently for physically based distributed models. This study explored how the use of limited continuous daily streamflow data might support the application of a physically based distributed model in datasparse basins. The influence of the length of the observation period on the calibration of the widely applied soil and water assessment tool model was evaluated in four Chinese basins with differing climatic and geophysical characteristics. The evaluations were conducted by comparing calibrations based on short periods of data with calibrations based on data from a 3-year period, which were treated as benchmark calibrations of the four basins, respectively. To ensure the differences in the model simulations solely come from differences in the calibration data, the generalized likelihood uncertainty analysis scheme was employed for the automatic calibration and uncertainty analysis. In the four basins, contrary to the common understanding of the need for observations over a period of several years, data records with lengths of less than 1 year were shown to calibrate the model effectively, i.e., performances similar to the benchmark calibrations were achieved. The models of the wet Jinjiang and Donghe basins could be effectively calibrated using a shorter data record (1 month), compared with the dry Heihe and upstream Yalongjiang basins (6 months). Even though the four basins are very different, when using 1-year or 6-month (covering a whole dry season or rainy season) data, the results show that data from wet seasons and wet years are generally more reliable than data from dry seasons and dry years, especially for the two dry basins. The results demonstrated that this idea could be a promising approach to the problem of calibration of physically based distributed hydrological models in datasparse basins, and findings from the discussion in this study are valuable for assessing the effectiveness of short-period data for model calibration in real-world applications.

Parameter uncertainty analysis for an operational hydrological model using residual based and limits of acceptability approaches

“…30 -40 % of a 35 hydrologic year, was able to effectively identify behavioural models; and this result is consistent with findings of other studies dealing with the effect of observation size on constraining model parameters (e.g. Seibert and Beven, 2009;Liu and Han, 2010;Sun et al, 2017). The information content of the input and observation datasets is more important than the length of the datasets especially in continuous rainfall-runoff modelling.…”

Factors controlling alterations in the performance of a runoff model in changing climate conditions

“…However, they require more input data, which is often not available, and have more parameters. Furthermore, the higher complexity of these models demands a longer computation time (Sun, et al, 2017). While conceptual models like HBV have minimal data requirements, require minimal computing time, they may not be well suited under changing conditions.…”

Prediction of hydrographs and flow-duration curves in almost ungauged catchments: Which runoff measurements are most informative for model calibration?

“…Moreover, event based sampling strategies resulted in better model performances than strategies with measurements at fixed time intervals (McIntyre and Wheater, 2004;Juston et al, 2009;Seibert and McDonnell, 2013). Model calibration with a limited number of runoff measurements performed best in relatively wet catchments (Perrin et al, 2007;Sun et al, 2017), which is a common observation in rainfall runoff modelling even when long continuous time series are available, or when runoff samples are selected during a wet period (Yapo et al, 1996;Vrugt et al, 2006;Kim and Kaluarachchi, 2009;Melsen et al, 2014;Correa et al, 2016). In addition, the consideration of hydrological variability and of hydrologically important processes was found to be essential for the calibration process and the resulting simulation uncertainty (Harlin, 1991;Vrugt et al, 2006;Konz and Seibert, 2010;Singh and Bárdossy, 2012).…”