Evaluation of multi- and many-objective optimization techniques to improve the performance of a hydrologic model using evapotranspiration remote sensing data

Authors: HERMAN, M.R. - Michigan State University; NEJADHASHEMI, A.P. - Michigan State University; HERNANDEZ-SUAREZ, J.S. - Michigan State University; KROPP, I. - University Of Michigan; SADEGHI, A.M. - US Department Of Agriculture (USDA)

Submitted to: Journal Hydrologic Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/11/2019
Publication Date: N/A
Citation: N/A

Interpretive Summary: This study explores two different calibration approaches; multi-objective and many-objective to test model calibration, using different ET/Remote Sensing products. In general, the best model performances were obtained from the multi-objective calibrations. The ensemble dataset utilized in this study showed the best fit with the Soil and Water Assessment Tool (SWAT) model and outperformed the individual actual evapotranspiration (Eta) products. Meanwhile, when considering the many-objective calibration, ETa performance was found to be satisfactory. It is important to note that this study was performed for only one watershed in Michigan. Future studies should expand this work using different hydrologic models with different physiographic and climatological zones. This would serve to test the robustness of the techniques implemented in this study.

Technical Abstract: In order to address global issues such as water security, it has become increasingly important to monitor the movement of water across the Earth. This has traditionally been accomplished through the use of land-based monitoring stations. However, high-resolution, large-scale monitoring is often not feasible due to a lack of monitoring stations across the globe. One solution to this issue is the use of hydrological models, which are less accurate than in-situ monitoring but can simulate the hydrological conditions in large areas. One way to improve the accuracy associated with models is the use of calibration and validation techniques. This has led to the development of a variety of different techniques such as single objective and multi-objective calibrations. However, the performance of a hydrological model using different calibration approaches and remote sensing products has not been evaluated, which is the main focus for this paper. Specifically, we explore the use of different multi- and many-objective calibration approaches in hydrological modeling when considering both observed streamflow and remotely sensed actual evapotranspiration (ETa). Eight remotely sensed ETa products were used here, namely: the USGS Simplified Surface Energy Balance (SSEBop), the USDA/NASA Atmosphere-Land Exchange Inverse (ALEXI), the MODIS Global Evapotranspiration Project (MOD16A2) 500m, the MOD16A2 1 km, the North American Land Data Assimilation Systems 2 Evapotranspiration (NLDAS-2) Mosaic, the NLDAS-2 Noah, the NLDAS-2 VIC, and finally TerraClimate. In addition to these datasets, an Ensemble was also developed and used in the Honeyoey Creek-Pine Creek Watershed, located in Michigan, USA. Regarding the calibration process, the Unified- Non-dominated Sorting Genetic Algorithm III (U-NSGA-III) was integrated with the Soil and Water Assessment Tool (SWAT). The first nine calibrations utilized a multi-objective approach with two objectives, one being streamflow and the other being one of the remotely sensed Eta products/Ensemble. The tenth calibration was a many-objective calibration with nine objective functions that represented observed streamflow and all eight of the remotely sensed evapotranspiration datasets. Nash- Sutcliffe efficiency (NSE), percent bias (PBIAS), and root mean squared error-observations standard deviation ratio (RSR) were used as the statistical calibration criteria and a measure of the overall model performance. Results showed that the multi-objective calibrations were able to successfully calibrate both streamflow and ETa. However, the highest model performances were achieved using the Ensemble ETa product (NSE = 0.79, PBIAS = 6.21%, and RSR = 0.46 for flow and NSE = 0.95, PBIAS = 1.82%, and RSR = 022 for ETa). Meanwhile, the required computational time for the many-objective calibration is significantly higher than the multi-objective calibration while also not successful in calibrating streamflow. However, the overall performance of many-objective method can be improved by considering weighting factors and constraining the search space.