Multimodel uncertainty changes in simulated river flows induced by human impact parameterizations

Global hydrological models (GHMs) increasingly incorporate human impacts like irrigation and reservoirs to simulate river flows. Prior research has shown that anthropogenic factors can dominate natural variability in some regions. However, the added complexity of human impact parameterizations may increase between-model uncertainty. No prior work had resolved how much additional uncertainty these parameterizations introduce. Existing studies have not quantified the signal-to-noise ratio (SNR) differences between models with and without human impacts. This gap motivated an investigation into how human impact representations affect simulation consistency. The uncertainty in model outputs is a known challenge in climate science. Yet, the specific contribution of human impact parameterizations to this uncertainty remains unclear. This study addresses that uncertainty by comparing simulations with and without human impacts.

This study aimed to quantify how human impact parameterizations affect between-model uncertainty in global hydrological simulations. The specific problem is that human activities like irrigation and reservoirs are often included in models, but their effect on uncertainty is not well understood. The motivation is to assess whether these parameterizations increase or decrease model agreement. The authors sought to compare simulations with and without human impacts using four GHMs. They focused on the signal-to-noise ratio (SNR) of river flows during 1971–2000. The study also aimed to identify which regions are most affected by increased uncertainty. Another goal was to determine if irrigation area fractions correlate with uncertainty changes. Lastly, the authors intended to suggest directions for improving GHM parameterizations.

The study used four global hydrological models (GHMs) to simulate river flows with and without human impact parameterizations. Two experiments were conducted: one with human impacts (VARSOC) and one without (NOSOC). The signal-to-noise ratio (SNR) was calculated for annual and high-flow simulations. The SNR measures the ratio of modeled signal to natural variability. The models were run for the period 1971–2000 to capture long-term trends. Basins were categorized by region and irrigation area fraction. Statistical comparisons were made between the two experiments to assess uncertainty changes. The study also analyzed how irrigation and reservoir regulation parameterizations differ across models. The results were mapped to show regional variations in uncertainty.

The VARSOC simulations showed higher between-model uncertainty than NOSOC simulations. The signal-to-noise ratio (SNR) differences were about 2% globally and varied regionally. The largest uncertainty increases were observed in Asia and northern Mediterranean regions. The SNR differences were mostly negative, indicating higher uncertainty in VARSOC simulations. Annual flow simulations showed SNR differences ranging from -20% to 5%. High-flow simulations had SNR differences from -20% to 20%. The uncertainty changes were strongly related to the fraction of irrigation areas in each basin. The study found that irrigation and reservoir regulation parameterizations differ significantly between models.

The authors concluded that human impact parameterizations increase between-model uncertainty in global hydrological simulations. The signal-to-noise ratio (SNR) differences suggest that these parameterizations introduce additional uncertainty. The largest uncertainty increases were observed in regions with high irrigation activity. The study highlights the need for better understanding of human impact parameterizations in GHMs. The authors propose that differences in irrigation and reservoir regulation approaches contribute to uncertainty. They suggest that statistical methods could help reduce between-model uncertainty. Calibration of GHMs is emphasized for improving historical simulations and future projections. The findings suggest that model development should focus on refining human impact representations.