Autocalibration in hydrologic modeling: Using SWAT2005 in small-scale watersheds

https://doi.org/10.1016/j.envsoft.2007.06.002Get rights and content

Abstract

SWAT is a physically based model that can simulate water quality and quantity at the watershed scale. Due to many of the processes involved in the manual- or autocalibration of model parameters and the knowledge of realistic input values, calibration can become difficult. An autocalibration-sensitivity analysis procedure was embedded in SWAT version 2005 (SWAT2005) to optimize parameter processing. This embedded procedure is applied to six small-scale watersheds (subwatersheds) in the central Texas Blackland Prairie. The objective of this study is to evaluate the effectiveness of the autocalibration-sensitivity analysis procedures at small-scale watersheds (4.0–8.4 ha). Model simulations are completed using two data scenarios: (1) 1 year used for parameter calibration; (2) 5 years used for parameter calibration. The impact of manual parameter calibration versus autocalibration with manual adjustment on model simulation results is tested. The combination of autocalibration tool parameter values and manually adjusted parameters for the 2000–2004 simulation period resulted in the highest ENS and R2 values for discharge; however, the same 5-year period yielded better overall ENS, R2 and P-values for the simulation values that were manually adjusted. The disparity is most likely due to the limited number of parameters that are included in this version of the autocalibration tool (i.e. Nperco, Pperco, and nitrate). Overall, SWAT2005 simulated the hydrology and the water quality constituents at the subwatershed-scale more adequately when all of the available observed data were used for model simulation as evidenced by statistical measure when both the autocalibration and manually adjusted parameters were used in the simulation.

Introduction

Public concern regarding the degradation of water quality due to nonpoint sources and point sources has driven policy regulators to scrutinize land management practices and examine how water quality conditions can be improved. Agricultural practices are commonly regarded as being sources of water and soil contamination (Sharpley, 1995, Abbozzo et al., 1996, Burkholder et al., 1997). Land application of manure provides nutrients and organic matter that enhance crop growth and can improve soil physical properties; however, when applied in excess, runoff from manured lands can result in the impairment of nearby water resources. Phosphorus (P) is a recognized contaminant that can cause adverse conditions in surface waters (Sharpley et al., 1994, Grobbelaar and House, 1995, Sims et al., 1998, Daniel et al., 1998).

Environmental regulation has expedited the necessity of agricultural producers to design and implement more environmentally suitable practices. There is a need to identify critical nutrient and their loss/transport potentials. Computer models can simulate multiple watershed management scenarios that can help environmental policy managers make decisions that could ultimately reduce P and N loss from agricultural lands. Models are inexpensive tools that can identify optimum watershed management practice scenarios for pollutant transport reduction.

Limited monitoring data exist at the watershed-scale for poultry litter application sites due to naturally inherent complexities such as rainfall variation, the requirement for a large amount of land, and the equipment and personnel required for data collection (Harmel et al., 2003a, Harmel et al., 2003b, Gilley and Risse, 2000). Long-term watershed monitoring data are rare due to the expense involved (Santhi et al., 2006); however, long-term simulations are needed to account for the inherent environmental variability (Rao et al., 2007). The ability of water quality models to accurately estimate environmental impacts from manure application needs to be determined.

Grayson et al. (1992) provided guidelines for analyzing any model which included testing measured data against simulated data and for a model's hydrologic processes to be tested over a wide range of watersheds and conditions, with both positive and negative results reported (Arnold et al., 1999, Chu and Shirmohammadi, 2004; and Rosenthal et al., 1995). Small-scale watershed studies have been conducted by Fohrer et al. (2001) and Srinivasan et al. (2005) at 26 and 39.5 ha, respectively. Fohrer et al. (2001) successfully analyzed the SWAT model (Arnold et al., 1998, Arnold and Fohrer, 2005) model for sensitivity to crop parameters and land use change. These studies are considered “small-scale” due to the relative size of watersheds that have been simulated with SWAT.

Barlund et al. (2007) used the SWAT model in a Finnish catchment to assess its usefulness to evaluate management impacts, such as nutrient load reductions. While the model proved its worthiness, it also demonstrated the necessity to adequately parameterize, calibrate and validate the model. These authors identify the need to include a parameter sensitivity analysis to concentrate on the more influential parameters that impact calibration. Krysanova et al. (2007) and Rao et al. (2007) agree with the previous authors that there is a demonstrated need for powerful calibration and validation techniques for hydrological models. In addition, there is a need to identify the criteria to achieve an adequate validation, which is based on sensitivity and uncertainty analyses to determine the most influential parameters and evaluate the model's uncertainty in relation to input data. Miller et al. (2007) emphasize the importance of the process used for parameter estimation; the higher the degree of spatial variability, the greater the complexity of correctly estimating parameter values.

This study evaluates the SWAT model's autocalibration-sensitivity analysis embedded procedures to simulate the stream discharge, sediment, organic nitrogen (N) and P, soluble P, and nitrate-N (NO3-N) loss after poultry litter application to small-scale agricultural land at a research site in central Texas. The periods of calibration and validation are also tested to emphasize the impact that the calibration time period has on model autocalibration results. The purpose of applying the SWAT model to these subwatersheds is to test if the autocalibration-sensitivity procedures embedded in SWAT2005 can be applied to small-scale watersheds (4.0–8.4 ha) resulting in realistic output.

Section snippets

SWAT model background

The SWAT model is a continuation of modeling efforts by the U.S. Department of Agriculture Agricultural Research Service (USDA ARS; Arnold et al., 1998, Arnold and Fohrer, 2005) and has become an effective means for evaluating nonpoint source water resource issues (flow, sediment, and nutrients) for a large variety of national and international water quality applications. The model is part of the U.S. Environmental Protection Agency (USEPA) Better Assessment Science Integrating Point & Nonpoint

Autocalibration and sensitivity analysis in SWAT2005

SWAT is a complex model with many parameters that can complicate manual model calibration. A parameter sensitivity analysis tool is embedded in SWAT to determine the relative ranking of which parameters most affect the output variance due to input variability (van Griensven et al., 2002). The SWAT model, version 2005 (SWAT2005) has an embedded autocalibration procedure that is used to obtain an optimal fit of process parameters. This procedure is based on a multi-objective calibration and

The Riesel research site as a case study

Data used in this study were obtained from an experimental research site located at the USDA ARS Grassland, Soil, and Water Research Laboratory near Riesel, Texas (31.1°N, 97.32°W) (Harmel et al., 2004, Wang et al., 2006). The simulated areas at Riesel are designated as “subwatersheds” due to their small size. The numerical distinction between labeling an area as a subwatershed versus a watershed in the literature is unclear. However, since the Riesel areas being simulated are some of the

Model evaluation methods

The performance of SWAT was evaluated using statistical analyses to determine the quality and reliability of the predictions when compared to observed values. Summary statistics included the mean and standard deviation (SD), which were used to assess SWAT's ability to reproduce the distribution of the observed data and to assess the variability between the observed and simulated data. The goodness-of-fit measures used were the coefficient of determination (R2; Eq. (5)) and the Nash–Sutcliffe

Model simulation approach

The initial N and P values were extrapolated from the percent organic carbon values and model defaults were utilized for the nutrient pools. The model defaults were used when initial values were not obtainable. The parameter values were allowed to vary within reasonable uncertainty ranges (Table 4) to calibrate for monthly and daily discharge, and monthly sediment and nutrient loss values.

Two data scenarios were used to demonstrate the impact of using all of the data available. Santhi et al.

Sensitivity analysis

Using SWAT's parameter sensitivity analysis procedure resulted in a slight variability amongst the six subwatersheds with CN2 and ESCO alternating as the most responsive parameter. The CN2 and ESCO parameters were found to be more sensitive to input variability than the SURLAG and FFCB parameters. The autocalibration tool embedded in SWAT allows the option of including sediment, organic N and P, and soluble P; however, the nitrate parameter is not yet included in the tool's options. As the

Conclusions

The ability of the SWAT model, version 2005, to simulate runoff, sediment, and nutrient loss data from small-scale subwatersheds in Texas was assessed in this study. Six subwatersheds were evaluated for sediment and nutrient water quality effects from poultry litter randomly applied in rates of 0–13.4 Mg ha−1 using both manual and autocalibrated adjusted parameters. Two data scenarios were employed, 2000–2004 and 2002. The first used data from 2000 to 2004 to demonstrate the carryover effect of

References (56)

  • J.E. Nash et al.

    River flow forecasting through conceptual models; part I – a discussion of principles

    J. Hydrol.

    (1970)
  • M. Rao et al.

    A web-based GIS Decision Support System for managing and planning USDA's Conservation Reserve Program (CRP)

    Environ. Model. Softw.

    (2007)
  • C. Santhi et al.

    A modeling approach to evaluate the impacts of water quality management plans implemented in a watershed in Texas

    Environ. Model. Softw.

    (2006)
  • P. Abbozzo et al.

    Environmental quality and hog production

    Environ. Monit. Assess.

    (1996)
  • J.G. Arnold et al.

    Automated base flow separation and recession analysis techniques

    Groundwater

    (1995)
  • J.G. Arnold et al.

    SWAT2000: current capabilities and research opportunities in applied watershed modeling

    Hydrol. Process

    (2005)
  • J.G. Arnold et al.

    Large-scale hydrologic modeling and assessment

  • J.G. Arnold et al.

    Large-area hydrologic modeling and assessment: Part I. Model development

    J. Am. Water Res. Assoc.

    (1998)
  • J.M. Burkholder et al.

    Impacts to a coastal river and estuary from rupture of a large swine waste holding lagoon

    J. Environ. Qual.

    (1997)
  • T.W. Chu et al.

    Evaluation of the SWAT model's hyrodology component in the piedmont physiographic region of Maryland

    Trans. ASAE

    (2004)
  • S.W. Chung et al.

    Validation of EPIC for two watersheds in southwest Iowa

    J. Environ. Qual.

    (1999)
  • S.W. Chung et al.

    Evaluation of EPIC for assessing tile flow and nitrogen losses for alternative agricultural management systems

    Trans. ASAE

    (2002)
  • T.C. Daniel et al.

    Agricultural phosphorus and eutrophication: a symposium overview

    J. Environ. Qual.

    (1998)
  • M. Di Luzio et al.

    Integration of watershed tools and SWAT model into BASINS

    J. Am. Water Res. Assoc.

    (2002)
  • Q.D. Duan

    Global optimization for watershed model calibration

  • Q. Duan et al.

    Effective and efficient global minimalization for conceptual rainfall–runoff models

    Water Resour. Res.

    (1992)
  • W.B. Gass

    Plant, Soil, and Water Testing Laboratory Recommendations

    (1987)
  • J.E. Gilley et al.

    Runoff and soil loss as affected by the application of manure

    Trans. ASAE

    (2000)
  • Cited by (165)

    View all citing articles on Scopus
    View full text