2015年-世界发展银行全球_Calibration_Approaches_for_Distributed_Hydrologic_Models_in_Poorly_Gaged_Basins___Implication_for_Streamflow_Projections_under_Climate_Change_28页_6mb

Summary of "Calibration approaches for distributed hydrologic models in poorly gaged basins: implication for streamflow projections under climate change"

Core Content

This study investigates the performance and uncertainty of calibration strategies for a distributed hydrologic model in a poorly gaged watershed, specifically the Kabul River basin. The goal is to improve model accuracy and better understand prediction uncertainty at interior ungaged sites, which is critical for climate change impact assessments in data-scarce regions.

Main Questions and Objectives

The study addresses four key research questions:

1. How does the calibration procedure using multisite data affect the accuracy and uncertainty of distributed models for streamflow predictions at ungaged sites?
2. What effects does increasing parameter complexity have on model calibration and prediction?
3. How much degradation in model accuracy and uncertainty can be expected when using only outlet gage data for calibration?
4. How do different calibration formulations alter streamflow projections under climate change conditions?

Key Findings

• Multisite Calibration: Simultaneous use of multisite data during model calibration significantly improves the accuracy and reduces uncertainty compared to a stepwise approach. Calibration based on the basin outlet alone performs poorly, leading to substantial deviations in mid-21st century streamflow projections.
• Parameter Complexity: Increased parameter complexity does not necessarily lead to higher uncertainty in streamflow projections, even though parameter equifinality (multiple parameter sets producing similar results) is present. This suggests that structural uncertainty may offset the effects of increased parametric complexity.
• Climate Model Uncertainty: The largest source of uncertainty in future streamflow projections comes from variations in climate model outputs, which far outweighs the uncertainty introduced by the calibration process.
• Calibration Caution: Calibration using only the basin outlet may not be reliable for predicting interior flow, especially in basins with significant spatial variability in physiographic properties. This is particularly relevant in developing countries where data is limited.

Methodology

• The study uses a distributed version of the HYMOD hydrologic model (HYMOD_DS) applied to the Kabul River basin.
• Data Sources: Gridded daily precipitation and temperature data (0.25° resolution) from 1961 to 2007 were used, along with monthly streamflow data from seven stations in the basin (1960–1981). The data were bias-corrected using the UD dataset.
• Glacier and Snow Modeling: Glacier coverage was extracted using the RGI 3.2 dataset, and glacier melt was modeled using a degree day factor (DDF) approach. A parameter $ r > 1 $ was introduced to reflect the higher melt rate of glaciers compared to snow.
• Model Parameters: The HYMOD_DS model includes 15 parameters, such as $ C_{\text{max}} $, $ B $, $ \alpha $, and others related to snowmelt, glacier storage, and routing processes.
• Optimization Technique: A genetic algorithm (GA) was employed for model calibration, with the Nash-Sutcliffe Efficiency (NSE) as the primary performance metric. A multisite average of NSE was used to evaluate model performance across multiple locations.
• Alternative Metric: The Kling-Gupta Efficiency (KGE) was also used as an alternative performance metric, which accounts for model bias, variance, and correlation with observations.

Implications for Climate Change Impact Studies

• Calibration strategies have significant implications for streamflow projections under climate change. In particular, the use of multisite data and appropriate parameter complexity is crucial for improving model reliability in data-scarce regions.
• The study highlights the importance of considering both calibration uncertainty and structural uncertainty when assessing model performance in poorly gaged basins.
• The results emphasize the need for caution when using distributed models calibrated only at the basin outlet, as this may lead to inaccurate predictions at interior ungaged sites.

Conclusion

The study underscores the value of multisite calibration and the role of parameter complexity in improving distributed hydrologic model performance. It also highlights the limitations of using only outlet gage data in poorly gaged basins and the critical influence of climate model uncertainty on future streamflow projections. These findings are particularly relevant for climate change impact assessments in regions with limited hydrological data.