Improving Baseflow Simulations in a Distributed Hydrologic Model for Drought Seasons Using MODIS-based Evapotranspiration Products and Budyko's Dryness Index

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 6 February 2014: 11:30 AM
Room C209 (The Georgia World Congress Center )
Chengmin Hsu, NOAA, Boulder, CO; and R. Cifelli, L. Johnson, and R. J. Zamora

The Russian and Napa River watersheds (RNRW) in California define a region famed for grape-growing and winemaking. But to support this high-profit agriculture, the area requires significant water consumption in its growing seasons. Moreover, both watersheds have a large total population of around 480,000 and its rivers serve as habitats for the endangered salmon of Steelhead, Chinook, and Coho. While combating these competing water demands, water managers are now facing another problem: the watersheds' vulnerability to emerging erratic weather patterns in a time of extreme weather events. Such events have been witnessed in the RNRW in the form of the 2006 New Year's Day flood, the 2007-2009 droughts, and the unusually intense Spring 2008 frosts. To facilitate the establishment of wise water management strategies to fulfill the demands requested by all stakeholders, decision-makers have more expectations of the hydrologic models: particularly that they are capable of furnishing accurate baseflow information in addition to peak flow simulations. To that end, the National Oceanic Atmospheric Administration (NOAA) Physical Science Division Hydrometeorological Testbed (HMT) has been deploying the Hydrologic Laboratory – Research Distributed Hydrologic Model (HL-RDHM) in the RNRW. HL-RDHM is a hydrologic model developed by NOAA National Weather Service (NWS) Office of Hydrologic Development (OHD). In this study, the enhancement of the baseflow simulation capabilities in drought periods using integrated remotely-sensed land surface variables and a hydroclimatological relationship are demonstrated. The fractions of rainfall loss by evapotranspiration at a daily time scale in various sub-watersheds were the focus of this study. Three major tasks were implemented in this study.

The first task involved calibrating the model parameters against the flow observation at the USGS river gauge stations. The selected stations for calibration are located in the three sub-watersheds with the least impaired flow influenced. The streamflow simulation using the calibrated model results in Nash-Sutcliffe Efficiencies (NSE) higher than 0.85 in the calibration period (Feb.1, 2011 to Dec. 31, 2012) and higher than 0.8 for the validation period (Jan. 1, 2013 to Jun. 30, 2013), suggesting that the model behavior well represents the watersheds' hydrologic response. Nevertheless, cumulative flow comparison for the calibration period still demonstrates increasing deviation after the middle of the dry period. Traditionally, the calibration of these parameters depends on trial-and-error manual procedures. However, the subjectivity of the manual evaluations and the complex tradeoffs of the model's behaviors often result in several possible calibration answers (sets of parameter values) that generate equally “good” streamflow simulations. Consequently, an unrealistic estimation of evapotranspiration and deep percolation may be adopted into simulations and adversely influences baseflow forecasting in dry periods, even if the simulations for peak flow and interflow appear adequate. To resolve this problem, the second task involved developing a time series statistical analysis framework, allowing for the analysis of how precipitation interacts with soil and vegetation characteristics in influencing streamflow and soil moisture dynamics. This task began with deriving the mean monthly potential evapotranspiration (PE) from the 5-year Moderate Resolution Imaging Spectroradiometer (MODIS) MOD16 evapotranspiration (ET) product, which was then inserted into HL-RDHM as the parameters. Then, the Global Vegetation Moisture Index (GVMI) was derived from the MODIS adjusted reflectance product (MOD43B4) and Enhanced Vegetation Index (EVI) for the calibration and validation period. Generally, GVMI can provide information on vegetation water content for the growing seasons and on the separation fraction between surface water and bare soil for the winters. This vegetation-related information, together with the precipitation and soil information, was incorporated into the analytic process exploring the interactions among simulated and observed variables, such as actual ET, streamflow, soil moisture, latent heat, and Leaf Area Index (LAI), for the three selected sub-watersheds. The diagnosis indicates that the factors of vegetation, man-made features (e.g. farm ponds), and evapotranspiration loss rates are not well characterized in the a priori parameters for HL-RDHM, resulting in biases in baseflow simulation, especially for the drought periods. In the third task, two dimensionless numbers, which encapsulate the monthly climatology, soil, and vegetation conditions, were used to create the classical Budyko curves on the three selected sub-watersheds. The different curves were created to describe the partitioning pattern of precipitation into evapotranspiration and runoff, in addition to deep percolation for each sub-watershed. The Budyko curves relate evapotranspiration losses to the two dimensionless numbers, eventually enhancing the calibration of the PE parameters of HL-RDHM. Given that subsurface flow and surface runoff has been acceptably simulated in the model, the realization of ET losses in turn improves the calibration of the parameters related to percolation (e.g. the SIDE and ZPERC parameters in HL-RDHM).

This study provided a method to enhance the baseflow simulation during drought using the semi-empirical Budyko curve and the MODIS products. The results show that partitioning of rainfall into ET and runoff plus percolation based on the Budyko curve improves HL-RDHM calibration. Most importantly, the study elucidated the governing role of integrated vegetation, soil, and rainfall factors, specifically as described by the Budyko curve and the GVMI, in the space–time variability of hydrological response in the RNRW for the deep percolation and ET domains. Therefore, establishing baseflow and deep percolation parameters as a constant often abuses the natural mechanism which regulates baseflow, as plants and soil evolve with time. This study also found that deep percolation did not significantly correlate with mean monthly rainfall but rather correlated with the surficial soil properties of percent sand and percent silt. The link between areal extent of vegetation, available water capacity, and runoff production during drought was created in this study.