Improving Water-Balance of a Coupled Vegetation-Hydrological Model through Bayesian Calibration to Satellite-Based Leaf Area Index
Abstract
Vegetation influences the water-balance in a watershed, for example, through its control on surface runoff and evapotranspiration. Introducing plant-growth in hydrological models has been shown to improve the accuracy of watershed models. For calibrating these plant-growth models, vegetation data is required in the watershed. Information on temporal vegetation characteristics like Leaf Area Index (LAI) is available over large scales through remote-sensing. Satellite-based remote-sensing data have associated uncertainties, in addition to those in model parameters. Accounting for these uncertainties is essential for robust model predictions. A Bayesian framework provides a suitable scheme for accounting for different sources of uncertainty. In this study, we calibrate a coupled plant-hydrological model in a Bayesian framework to satellite-based LAI. The HBV model in the Raven hydrological modelling framework is coupled to the SWAT-EPIC plant-growth model in the Robin vegetation modelling framework to simulate land-surface and hydrological processes of the Nith river in Ontario, Canada. Spatially distributed LAI estimates from Moderate Resolution Imaging Spectroradiometer (MODIS), gathered between 2011 and 2014 are used. In the first step, we calibrate the hydrological model using a parameter optimization technique to gauged stream-flow data with default parameters in the plant model. Keeping the optimized parameters of the hydrological model fixed, we then calibrate the plant model to LAI in a Bayesian framework. We define a suitable likelihood model which ensures that statistical assumptions are met and captures differences between the sub-basins of the watershed. We assess the water-balance of the calibrated model by comparing simulated soil-moisture to satellite-based estimates. A total of 35 parameters are used while calibrating the hydrological model. The stream-flow calibration resulted in a Nash-Sutcliffe Efficiency (NSE) of 0.71 for the calibrated gauge and 0.58 for the uncalibrated gauge. Vegetation-specific plant parameters for forests, grasslands, corn, soybean, and wheat, are used while calibrating the plant growth-model. Bayesian calibration to satellite-based LAI is expected to improve the soil-moisture estimates in the Nith watershed.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H15D1071V