Multi-variable parameter estimation for a global hydrological model: Comparison and evaluation of three ensemble-based calibration methods for the Mississippi River basin

• DOI: 10.5194/hess-2023-18
• Publication Date: 2023-05-11T11:33:35Z
• Authors: Petra Döll; H. M. Mehedi Hasan; Kerstin Schulze; Helena Gerdener; Lara Börger; Somayeh Shadkam; Sebastian Ackermann; Seyed-Mohammad Ho...; Hannes Müller Sch...; Andreas Güntner; Jürgen Kusche
• Keywords: 13. Climate action; 0207 environmental engineering; 02 engineering and technology; 6. Clean water

Abstract. Global hydrological models enhance our understanding of the Earth system and support sustainable management water, food energy in a globalized world. They integrate process knowledge with multitude model input data (e.g., precipitation, land cover soil properties location extent surface water bodies) that describe state Earth. However, they do not fully utilize observations output variables streamflow storage) to decrease uncertainty by, e.g., parameter estimation. For pilot region Mississippi River basin, we assessed suitability three ensemble-based multi-variable calibration approaches for identifying both optimal behavioral sets global WaterGAP, utilizing (Q) total storage anomaly (TWSA). The common first steps all are 1) definition spatial units which parameters uniformly adjusted (CDA units), combined selection observation data, 2) identification potential their a-priori probability distributions 3) sensitivity analyses select most influential per CDA unit will be by calibration. In estimation uncertainty, considered uncertainties Q TWSA observations. We found Pareto-optimal (POC) approach, utilizes Borg multi-objective evolutionary search algorithm find sets, is best suited single “optimal” set each unit. This leads an improved fit monthly time series as compared standard WaterGAP variant, only calibrated against mean annual Q, can used compute estimate output. Generalized Likelihood Uncertainty Estimation (GLUE) approach less suitable than POC identify but enables due equifinality uncertainty. advantages ensemble Kalman filter assimilation (EnCDA) storages updated, could realized, likely high computational burden this limited EnCDA size 32, while 20,000 members evaluated case GLUE. Partitioning whole basin into five (sub-basins) instead one performance during validation periods. Very diverse were lead similarly good fits observations, range values narrowed Model structure uncertainties, particular regarding operation man-made reservoirs, small wetlands, (lacking) representation losing river conditions suspected main reasons low coverage bands GLUE-derived bands. also reason major trade-offs between TWSA. Calibration GRACE only, regions without may worsen simulation uncalibrated variant. plan add additional remotely-sensed suggest considering multi-model studies freshwater system.