Skip to main content
ARS Home » Plains Area » Temple, Texas » Grassland Soil and Water Research Laboratory » Research » Publications at this Location » Publication #331758
Title: Assessment of input uncertainty by seasonally categorized latent variables using SWAT

Author
item YEN, HAW - Texas Agrilife Research
item SU, YU-WEN - Cheng Kung University
item WOLFE III, JUNE - Texas Agrilife Research
item CHEN, SHIEN-TSUNG - Cheng Kung University
item HSU, YU-CHAO - Cheng Kung University
item TSENG, WEN-HSIAO - Cheng Kung University
item BRADY, DAWN - Illinois College
item JEONG, JAEHAK - Texas Agrilife Research
item Arnold, Jeffrey

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/23/2015
Publication Date: 10/31/2015
Publication URL: https://handle.nal.usda.gov/10113/63229
Citation: Yen, H., Su, Y., Wolfe III, J.E., Chen, S., Hsu, Y., Tseng, W., Brady, D.M., Jeong, J., Arnold, J.G. 2015. Assessment of input uncertainty by seasonally categorized latent variables using SWAT. Journal of Hydrology. 531:685-695.

Interpretive Summary: Watershed models are commonly used to determine the impact of land management and climate change on water supply and quality. However, there is considerable uncertainty in model input and output. Techniques were developed to include uncertainty variables in model predictions. The study also showed that seasonal adjustments are required to improve model reliability. By understanding and reporting model uncertainty, we can make more informed land management decisions based on model output.

Technical Abstract: Watershed processes have been explored with sophisticated simulation models for the past few decades. It has been stated that uncertainty attributed to alternative sources such as model parameters, forcing inputs, and measured data should be incorporated during the simulation process. Among varying uncertainty sources, input uncertainty attributed to precipitation data exhibits a dominant role, as it is the source driving most hydrologically-related processes. In previous studies, latent variables (normally distributed random noise) have been implemented to explicitly incorporate input uncertainty from precipitation data. However, it may not be appropriate to apply the same set of latent variables throughout temporal series without considering seasonal effects. In this study, seasonally categorized latent variables were defined to investigate potential effects on model predictions and associated predictive uncertainty. Results show that the incorporation of seasonal latent variables resulted in better statistical solutions (NSE, Nash–Sutcliffe Efficiency coefficient) for both calibration (0.58[streamflow]/0.73[sediment]/0.59[ammonia]) and validation (0.57[streamflow] /0.45[sediment]/0.53[ammonia]) periods. Alternative definitions of Dry/Wet seasonality (two definitions are defined in this study) also affected model predictions. In addition, it was determined that predictive uncertainty can be enhanced by incorporating more latent variables during model calibration. The implementations of proposed seasonal latent variables have further substantiated the importance of incorporating seasonal effects when conducting comparable approaches. Applications of latent variables on future work should evaluate potential effects on model predictions before performing associated scientific studies or relevant decision making processes.