HYDROLOGICAL PROCESSES AND MODEL REPRESENTATION: IMPACT OF SOFT DATA ON CALIBRATION J. G. Arnold, M. A. Youssef, H. Yen, M. J. White, A. Y. Sheshukov, A. M. Sadeghi, D. N. Moriasi, J. L. Steiner, D. M. Amatya, R. W. Skaggs, E. B. Haney, J. Jeong, M. Arabi, P. H. Gowda ABSTRACT. Hydrologic and water quality models are increasingly used to determine the environmental impacts of climate variability and land management. Due to differing model objectives and differences in monitored data, there are currently no universally accepted procedures for model calibration and validation in the literature. In an effort to develop accepted model calibration and validation procedures or guidelines, a special collection of 22 research articles that present and discuss calibration strategies for 25 hydrologic and water quality models was previously assembled. The models vary in scale temporally as well as spatially from point source to the watershed level. One suggestion for future work was to syn- thesize relevant information from this special collection and to identify significant calibration and validation topics. The objective of this article is to discuss the importance of accurate representation of model processes and its impact on cali- bration and scenario analysis using the information from these 22 research articles and other relevant literature. Models are divided into three categories: (1) flow, heat, and solute transport, (2) field scale, and (3) watershed scale. Processes simulated by models in each category are reviewed and discussed. In this article, model case studies are used to illustrate situations in which a model can show excellent statistical agreement with measured stream gauge data, while misrepre- sented processes (water balance, nutrient balance, sediment source/sinks) within a field or watershed can cause errors when running management scenarios. These errors may be amplified at the watershed scale where additional sources and transport processes are simulated. To account for processes in calibration, a diagnostic approach is recommended using both hard and soft data. The diagnostic approach looks at signature patterns of behavior of model outputs to deter- mine which processes, and thus parameters representing Submitted for review in April 2014 as manuscript number SW 10726; them, need further adjustment during calibration. This approved for publication by the Soil & Water Division of ASABE in overcomes the weaknesses of traditional regression-based October 2014. Mention of company or trade names is for description only and does calibration by discriminating between multiple processes not imply endorsement by the USDA. The USDA is an equal opportunity within a budget. Hard data are defined as long-term, provider and employer. measured time series, typically at a point within a water- The authors are Jeffrey G. Arnold, ASABE Fellow, Agricultural shed. Soft data are defined as information on individual Engineer, USDA-ARS Grassland Soil and Water Research Laboratory, Temple, Texas; Mohamed A. Youssef, ASABE Member, Associate processes within a budget that may not be directly meas- Professor, Department of Biological and Agricultural Engineering, North ured within the study area, may be just an average annual Carolina State University, Raleigh, North Carolina; Haw Yen, Post- estimate, and may entail considerable uncertainty. The Doctoral Research Associate, Texas AgriLife Research, Temple, Texas; advantage of developing soft data sets for calibration is Michael J. White, Agricultural Engineer, USDA-ARS Grassland Soil and that they require a basic understanding of processes (wa- Water Research Laboratory, Temple, Texas; Aleksey Y. Sheshukov, ASABE Member, Assistant Professor, Department of Biological and ter, sediment, nutrient, and carbon budgets) within the spa- Agricultural Engineering, Kansas State University, Manhattan, Kansas; tial area being modeled and constrain the calibration. Ali M. Sadeghi, Soil Scientist, USDA-ARS Hydrology and Remote Keywords: Calibration, Field-scale models, Point models, Sensing Laboratory, Beltsville, Maryland; Daniel N. Moriasi, ASABE Member, Research Hydrologist, and Jean L. Steiner, Supervisory Soil Validation, Watershed models. Scientist, USDA-ARS Grazinglands Research Laboratory, El Reno, Oklahoma; Devendra M. Amatya, ASABE Member, Research Hydrologist, USDA Forest Service, Cordesville, South Carolina; R. Wayne Skaggs, ASABE Fellow, W. N. Reynolds and Distinguished ater quality and hydrologic models are com- University Professor, Department of Biological and Agricultural monly used to assess the environmental im- Engineering, North Carolina State University, Raleigh, North Carolina; Elizabeth B. Haney, Graduate Research Assistant, and Jaehak Jeong, pacts of land management and policy deci- ASABE Member, Assistant Professor, Texas AgriLife Research, Temple, Wsions. Models are increasingly being applied Texas; Mazdak Arabi, Assistant Professor, Department of Civil and to large varied agricultural landscapes to address contem- Environmental Engineering, Colorado State University, Ft. Collins, porary water resource issues in the context of climate Colorado; Prasanna H. Gowda, ASABE Member, Research Agricultural Engineer, USDA-ARS Conservation and Production Research Laboratory, change and sea level rise (Jayakrishnan et al., 2005). Mori- Bushland, Texas. Corresponding author: Jeffrey Arnold, USDA-ARS asi et al. (2012) summarized the calibration approaches of GSWRL, 808 East Blackland Road, Temple, TX; phone: 254-770-6508; e- 25 models in a special collection of 22 articles, each focus- mail: [email protected]. Transactions of the ASABE Vol. 58(6): 1637-1660 2015 American Society of Agricultural and Biological Engineers ISSN 2151-0032 DOI 10.13031/trans.58.10726 1637 ing on the individual model calibration and validation strat- algae and aquatic plants, and cycling in floodplains, estuar- egies. These models vary in scope from field-scale models ies, wetlands, and large hydraulic structures. Processes con- that focus on flow, heat, and solute transport to large-scale sidered and calibration techniques used by each of the 25 watershed models that incorporate complex processes over models are described in the special collection (Moriasi et spatially diverse subwatersheds. Regardless of scale, model al., 2012). accuracy is improved through calibration, and model uncer- This article describes the importance of realistically tainty (thus utility) is evaluated via validation. Calibration simulating all critical processes in the hydrological balance and validation are important factors in the development of for calibration and validation of small- and large-scale meaningful model predictions of potential future land use models. For example, if surface runoff is overestimated, it or climate effects. There are no universal standards for the is likely that evapotranspiration (ET) and/or subsurface and calibration and validation of models in the current litera- tile flow are underestimated, resulting in overestimation of ture, as the procedure is generally dependent upon the pro- sediment yields and underestimation of subsurface nitrate cesses in play at each model scale. Basic model processes and other soluble contaminant yields. This will cause fur- include hydrology (water budget), erosion and sedimenta- ther error when parameterizing variables related to sedi- tion, plant growth, nutrient and carbon cycling, and con- ment and nutrient transport and result in unrealistic policy taminant fate and transport. Model processes, in varying recommendations when running scenarios that target ero- degrees, are interconnected and impacted by land manage- sion and fertilizer management. Problems are even more ment, topography, climate, and scale. Point-scale models compounded at the watershed scale when multiple fields or are used primarily to simulate physical and biological pro- subbasins are simulated and output is routed through chan- cesses such as water and heat flow and reactive solute nels, flood plains, and reservoirs. Thus, it is important to transport through a soil column and may be in finer time reasonably simulate nutrient and sediment sources and scales of less than an hour. At the field scale, the water sinks within a watershed in addition to their loads at a balance processes include many variables that are interde- gauging station (outlet). If upland erosion is overpredicted, pendent with other processes, such as plant growth, soil channel erosion must be underpredicted to match measured properties, and weather. Nutrient cycling processes are gauge loads. The management practices designed to reduce complex and may vary greatly depending upon soil condi- erosion from the landscape may then show significant im- tions even at the field scale. The carbon cycle appears sim- pact on total sediment yields, while in reality the practices ple at first glance; however, individual components, such as would have little impact at the basin outlet. It is also im- photosynthesis and soil carbon dynamics, are complex to portant at the watershed scale to accurately simulate proper simulate using a model. For example, photosynthesis is source load allocations. For example, excellent calibration vegetation dependent and is influenced by resource (light, statistics can be obtained at a stream gauge outlet even water, and nutrients) availability and environmental condi- though point sources are underestimated and the loads from tions. Soil carbon dynamics are influenced by the amount agricultural