The efficacy of calibrating hydrologic model using remotely sensed evapotranspiration and soil moisture for streamflow prediction
Introduction
Significant research has been done in the past to develop efficient calibration algorithms to attain reliable streamflow predictions at gauged catchments. Calibration of hydrologic/land surface models is carried out usually using streamflow since it gives time-integrated information about water output from the catchment. However, lack of streamflow observations and other ground data in the vast majority of areas makes hydrological model calibration a difficult task. Particularly, accurate estimation of runoff at ungauged catchments is a growing concern for the hydrologic community (Sivapalan, 2003, Wagener and Montanari, 2011).
Alternative approaches to the conventional streamflow-based calibration for ungauged locations include estimation of parameters from prior information on catchment physical characteristics (e.g., soil hydraulic properties and vegetation properties) (Atkinson et al., 2003, Koren et al., 2003) and regionalization. For example, in parameter regionalization, a model is calibrated for a number of gauged watersheds and the model parameters are derived from a regression relationship between the watershed characteristics and parameters (Abdulla and Lettenmaier, 1997, Jakeman et al., 1992, Parajka et al., 2007, Post et al., 1998, Sefton and Howarth, 1998, Viney et al., 2009, Wagener and Wheater, 2006, Wagener et al., 2004, Zhang et al., 2011). Yadav et al. (2007) modified the approach by regionalizing flow characteristics and by incorporating uncertainty in the regressed estimates. However, the drawbacks of these approaches have been reviewed previously (Beven, 2000, Wagener and Montanari, 2011, Wagener and Wheater, 2006): uniqueness in topography, geology, vegetation features and anthropogenic modification for the watersheds within similar climatic region can make individual watersheds respond to input in widely different manners (Beven, 2000). Wagener and Wheater (2006) stated that through regionalization, model structural uncertainty is transferred to the optimized parameters, which can result in biased calibration. To explore methods that can overcome these issues associated with the regionalization, we investigate the efficacy of RS data in calibration and its utility as an alternative source of observational data to constrain our model parameter estimates for ungauged locations.
Typically, RS data provides spatially distributed land surface parameters with a regular temporal repeat across much of the globe, and the volume and quality of data have greatly increased over the past decades providing extensive datasets for hydrologic modelling. The RS data has been widely used in land use and land cover classification (e.g., soil and land use pattern) and the estimation of vegetation indices, surface SM and evapotranspiration (ET) (Asner et al., 2003, Chen and Cihlar, 1996, Congalton, 1991, Gupta et al., 2008, Kustas et al., 1994, Ottlé et al., 1989, Owe et al., 2008, Sucksdorff and Ottle, 1990). Resulting land surface products have enhanced overall capability of hydrologic modelling from local to global scales. For example, several studies demonstrated that annexation of RS land surface temperature, SM, leaf area index (LAI) and fraction vegetation cover data improves hydrologic modelling (Andersen et al., 2002, Corbari and Mancini, 2013, Crow and Ryu, 2009, Garcia-Quijano and Barros, 2005, Parajka et al., 2006, Renzullo et al., 2008, Silvestro et al., 2015, Sutanudjaja et al., 2014, Zhang and Wegehenkel, 2006). LAI estimated by the MODerate resolution Imaging Spectrometer (MODIS) onboard Terra and Aqua satellites can be used with the Penman–Monteith equation to produce 8-day composite ET (Cleugh et al., 2007, Leuning et al., 2008, Zhang et al., 2008). Satellite-derived products, such as ET, SM and vegetation biomass, have become important parts of hydrological modelling.
ET retrieved from AVHRR (Advanced Very High Resolution Radiometers) (Nemani and Running, 1989, Taconet et al., 1986) and MODIS (Cleugh et al., 2007, Guerschman et al., 2009, Leuning et al., 2008, Zhang et al., 2008) satellites and microwave SM retrievals from the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR‐E) (Njoku et al., 2003, Owe et al., 2008), Soil Moisture and Ocean Salinity (SMOS) (Kerr et al., 2001) of European Space Agency (ESA), and Advanced Scatterometer (ASCAT) (Bartalis et al., 2007) have widely been used in hydrology. Immerzeel and Droogers (2008) calibrated the Soil Water Assessment Tool (SWAT) model with MODIS ET using Gauss–Marquardt–Levenberg (GML) algorithm which increased correlation between simulated and observed ET resulting in improved streamflow predictions. Using a two-parameter surface conductance (Gs) model, optimized by catchment water balance estimates of ET (precipitation minus runoff), Zhang et al. (2008) generated 8-day composite ET (ERS) from MODIS LAI. The ERS was useful in estimating long-term runoff using the concept of water balance. But the study did not address the capability of ERS in calibrating rainfall–runoff models. Zhang et al. (2009) concluded that multi-objective calibration of SimHyd model with streamflow and ERS produced better daily and monthly runoff compared to calibration with streamflow alone. Droogers et al. (2010) showed that optimization of the Soil-Water-Atmosphere-Plant (SWAP) model using satellite-derived actual ET can predict irrigation demand with acceptable accuracy.
Hydrological application of RS SM has been primarily on its assimilation into land surface models to improve profile SM and other outputs linked to the soil (Crow and Van den Berg, 2010, Han et al., 2012, Pauwels et al., 2001, Reichle and Koster, 2005, Renzullo et al., 2014). Recent studies have shown that assimilation of satellite SM can improve streamflow predictions of rainfall–runoff models (Alvarez-Garreton et al., 2014, Alvarez-Garreton et al., 2015, Crow and Ryu, 2009, Parajka et al., 2006, Pauwels et al., 2001). Similarly, the assimilation of a Soil Wetness Index (SWI) derived from ASCAT into a continuous distributed hydrologic model resulted in improved discharge predictions (Brocca et al., 2010, Brocca et al., 2012).
Very few studies have incorporated satellite SM in model calibration. Campo et al. (2006) used SM retrievals from European Remote Sensing (ERS) scatterometer signals to optimize the parameters of soil dynamics in distributed hydrologic model. Parajka et al. (2009) used ERS scatterometer signals to calibrate semi-distributed hydrologic model along with runoff. Although both studies demonstrated improved streamflow predictions, Campo et al. (2006) study was limited to the areas with no or sparse vegetation while the study by Parajka et al. (2009) showed deterioration in SM predictions. Zhang et al. (2011) calibrated the Australian Water Resource Assessment landscape model (AWRA-L) with streamflow, NOAA-AVHRR LAI and TRMM-MI (Tropical Rainfall Measuring Mission – Microwave Imager) SM using multi-objective criteria. Even though the study resulted in marked improvement of LAI and SM, the improvement of streamflow was marginal. Sutanudjaja et al. (2014) calibrated a physically based large-scale coupled groundwater-land surface model, called PCR-GLOBWB-MOD, using SWI derived from ERS scatterometer and discharge data. The resulting optimum parameter set predicted discharge, SM and groundwater dynamics with acceptable accuracy.
Challenges in using remote sensing data for model calibration and data assimilation are discussed comprehensively in Van Dijk and Renzullo (2011). One of the major challenges is that the information content of RS observations varies based on transient vegetation (Barrett and Renzullo, 2009, Campo et al., 2006) and topography (Parajka et al., 2009). Appropriate specification of errors, as part of the retrieval process, is critical for effective calibration and data assimilation (Alvarez-Garreton et al., 2014). Suitability of coarse-resolution remote sensing data for small-scale studies is also a concern.
Previous studies show that RS data can be used to optimize the model parameters under certain conditions (Kunnath Poovakka et al., 2013, Mohanty, 2013, Zhou et al., 2013). Zhang et al. (2008) suggest that RS ET and SM can be used in the calibration of rainfall–runoff models to improve runoff estimations in ungauged catchments. A calibration scheme which relies soley on remote sensing data will be greatly beneficial in modelling at ungauged catchments, especially if it can be demonstrated to result in improved estimation compared with uncalibrated model. In this study, Microwave SM retrievals from the AMSR-E and daily estimates of ET from CSIRO MODIS ReScaled potential ET (CMRSET) model are used to calibrate a hydrologic model using 15 different objective functions considering various combinations of RMSD and correlation of ET and SM. The Shuffled Complex Evolution (SCE) calibration algorithm is used to calibrate a grid-based hydrologic model modified from the Australian Water Resource Assessment – Landscape (AWRA-L) model. Main research objectives of this study are to investigate (1) how effective RS ET and SM are in calibrating the model and (2) how biases in RS ET affects prediction of calibrated models.
Section snippets
Study catchments
The efficacy of calibrating AWRA-L with RS ET and SM is assessed for 11 catchments in eastern Australia. First, the calibration method is developed and tested in the Loddon River catchment at Newstead, Victoria, Australia and then the method is evaluated in 10 catchments along eastern Australia (Fig. 1). The Loddon River catchment contains the Wombat flux tower station upstream and is one of the hydrologic reference stations maintained by the Australian Bureau of Meteorology (BoM). The observed
Sensitivity analysis
The Sobol’ parameter sensitivity to streamflow, ET and SM without any objective function and in terms of RMSD and the linear correlation is shown in Fig. 3 for two grid cells in Loddon Catchment. A subset of nine parameters showed higher sensitivity to at least one of those three outputs. The nine parameters (see Table 2) include four land cover specific parameters (Tgrow, Tsenc, hveg and Ud0), two parameters defining soil water drainage (beta and FdrainFC), one parameter representing soil
Discussion
Calibrating hydrologic models with ET and SM has important implications for streamflow prediction in ungauged catchments. A widely used method to predict streamflow at ungauged or sparsely gauged catchments is the regionalization of parameters or flow indices (Abdulla and Lettenmaier, 1997, Jakeman et al., 1992, Parajka et al., 2007, Post et al., 1998, Sefton and Howarth, 1998, Viney et al., 2009, Wagener and Wheater, 2006, Wagener et al., 2004, Zhang et al., 2011, Yadav et al., 2007). Many
Conclusion
The current study analyzed the ability of RS ET (CMRSET) and SM (AMSR-E) in calibrating a grid based hydrologic model, AWRA-L. Efficacy of the calibration was evaluated on the accuracy of streamflow predictions. A total of 15 objective functions considering RMSD and correlation of ET and SM were used to calibrate the model.
Most objective functions performed well in the catchments with medium to high average runoff. In particular, calibration based on NRMSDET provided the best streamflow
Acknowledgements
This research is financially supported by International Postgraduate Research Scholarships (IPRS), the Australian Postgraduate Award (APA) and the Carlton Connect Initiative Fund Project 22514, representing collaboration between the Melbourne School of Engineering, The University of Melbourne, and the South Australian Government Department of Environment, Water and Natural Resources. We would like to acknowledge Commonwealth Scientific and Industrial Research Organization (CSIRO) and Bureau of
References (81)
- et al.
Development of regional parameter estimation equations for a macroscale hydrologic model
J. Hydrol.
(1997) - et al.
The impacts of assimilating satellite soil moisture into a rainfall–runoff model in a semi-arid catchment
J. Hydrol.
(2014) - et al.
Use of remotely sensed precipitation and leaf area index in a distributed hydrological model
J. Hydrol.
(2002) - et al.
Retrieving leaf area index of boreal conifer forests using Landsat TM images
Remote Sens. Environ.
(1996) - et al.
Regional evaporation estimates from flux tower and MODIS satellite data
Remote Sens. Environ.
(2007) A review of assessing the accuracy of classifications of remotely sensed data
Remote Sens. Environ.
(1991)- et al.
An evaluation of AMSR-E derived soil moisture over Australia
Remote Sens. Environ.
(2009) - et al.
Estimating actual irrigation application by remotely sensed evapotranspiration observations
Agric. Water Manage.
(2010) - et al.
Incorporating canopy physiology into a hydrological model: photosynthesis, dynamic respiration, and stomatal sensitivity
Ecol. Modell.
(2005) - et al.
Scaling of potential evapotranspiration with MODIS data reproduces flux observations and catchment water balance observations across Australia
J. Hydrol.
(2009)
Implementation of surface soil moisture data assimilation with watershed scale distributed hydrological model
J. Hydrol.
Calibration of a distributed hydrological model based on satellite evapotranspiration
J. Hydrol.
A systematic approach to modelling the dynamic linkage of climate, physical catchment descriptors and hydrologic response components
Math. Comput. Simul.
Using satellite remote sensing to extrapolate evapotranspiration estimates in time and space over a semiarid rangeland basin
Remote Sens. Environ.
An integrated error parameter estimation and lag-aware data assimilation scheme for real-time flood forecasting
J. Hydrol.
Global review and synthesis of trends in observed terrestrial near-surface wind speeds: implications for evaporation
J. Hydrol.
Improvements to a MODIS global terrestrial evapotranspiration algorithm
Remote Sens. Environ.
Remote sensing applications to hydrological modeling
J. Hydrol.
The importance of the spatial patterns of remotely sensed soil moisture in the improvement of discharge predictions for small-scale basins through data assimilation
J. Hydrol.
An improved methodology for predicting the daily hydrologic response of ungauged catchments
Environ. Modell. Softw.
Multi-sensor model-data fusion for estimation of hydrologic and energy flux parameters
Remote Sens. Environ.
Continental satellite soil moisture data assimilation improves root-zone moisture analysis for water resources assessment
J. Hydrol.
Relationships between dynamic response characteristics and physical descriptors of catchments in England and Wales
J. Hydrol.
Inter-comparison of microwave satellite soil moisture retrievals over the Murrumbidgee Basin, southeast Australia
Remote Sens. Environ.
Application of satellite remote sensing to estimate areal evapotranspiration over a watershed
J. Hydrol.
Parameter estimation and regionalization for continuous rainfall–runoff models including uncertainty
J. Hydrol.
Regionalization of constraints on expected watershed response behavior for improved predictions in ungauged basins
Adv. Water Resour.
Integration of MODIS data into a simple model for the spatial distributed simulation of soil water content and evapotranspiration
Remote Sens. Environ.
Improving runoff estimates using remote sensing vegetation data for bushfire impacted catchments
Agric. For. Meteorol.
Improving operational flood ensemble prediction by the assimilation of satellite soil moisture: comparison between lumped and semi-distributed schemes
Hydrol. Earth Syst. Sci.
Global synthesis of leaf area index observations: implications for ecological and remote sensing studies
Global Ecol. Biogeogr.
Predicting space–time variability of hourly streamflow and the role of climate seasonality: Mahurangi Catchment, New Zealand
Hydrol. Process.
On the efficacy of combining thermal and microwave satellite data as observational constraints for root-zone soil moisture estimation
J. Hydrometeorol.
Initial soil moisture retrievals from the METOP-A Advanced Scatterometer (ASCAT)
Geophys. Res. Lett.
Uniqueness of place and process representations in hydrological modelling
Hydrol. Earth Syst. Sci. Discuss.
Improving runoff prediction through the assimilation of the ASCAT soil moisture product
Hydrol. Earth Syst. Sci.
Assimilation of surface-and root-zone ASCAT soil moisture products into rainfall–runoff modeling
IEEE Trans. Geosci. Remote Sens.
Use of multi-platform, multi-temporal remote-sensing data for calibration of a distributed hydrological model: an application in the Arno basin, Italy
Hydrol. Process.
Calibration and validation of a distributed energy-water balance model using satellite data of land surface temperature and ground discharge measurements
J. Hydrometeorol.
An improved approach for estimating observation and model error parameters in soil moisture data assimilation
Water Resour. Res.
Cited by (82)
Ecohydrologic model with satellite-based data for predicting streamflow in ungauged basins
2023, Science of the Total Environment