Calibration and validation of DRAINMOD to design subsurface drainage systems for Iowa's tile landscapes

Abstract

The study presents the potential use of existing common data sources such as a soil database, and techniques such as a pedotransfer function and a non-linear parameter estimator to calibrate and validate a hydrologic model to design subsurface drainage systems for Iowa's tile landscapes. A field scale deterministic hydrological model, DRAINMOD has been calibrated and validated for two soils: Webster soil cultivated with continuous corn (WEBS_CC) and Canisteo soil cultivated with corn–soybean rotation (CANI_CS). The cumulative subsurface drainage over the calibration and validation years from 1990 to 2003 was predicted 2% higher than the observed subsurface drainage for WEBS_CC (coefficient of mass residual CRM = 0.02) and 10% lower for CANI_CS (CRM = −0.10). The overall values of index of agreement IoA and model efficiency EF were higher than 0.85 for both WEBS_CC and CANI_CS, and showed a close agreement between the predicted and observed subsurface drainage. The calibrated and validated DRAINMOD was further used to simulate impacts of varying designs of subsurface drainage system for WEBS_CC over the 14 (1990–2003) years of weather record in Iowa's tile landscapes. Simulation results suggest that a drainage system designed for a drainage intensity of 0.46 cm day⁻¹ with a drain depth of 1.05 m and drain spacing of 25 m is sufficient enough to maximize crop production (average relative yield ≈98%) while minimizing subsurface drainage and its associated nitrate-nitrogen (NO₃-N) loss. Though the results are based on the simulations with simple and approximate methods to represent the real system, they indicate that installing the drains at shallower depth (<1.05 m) might help to reduce subsurface drainage, but there might be negative impacts in terms of increased excess water stress on crop production and increased surface runoff. Field experiments are recommended to study the impact of shallow drainage on the complete water balance and crop production in Iowa's tile landscapes.

Introduction

Large parts of the Iowa State (USA) are relatively flat and consist of poorly drained soils, which are artificially drained through subsurface drainage systems for improved crop production. Subsurface drainage removes excess water and improves crop production, but increases leaching of nitrate-nitrogen (NO₃-N) from agricultural lands. This is mainly due to the increased subsurface drainage, which transports NO₃-N from the soil profile. The increased leaching of NO₃-N, particularly from Iowa's agricultural areas, is identified as a major nonpoint source of pollution of the Mississippi River, and suspected of creating hypoxia in the northern Gulf of Mexico (Goolsby et al., 1997, Goolsby et al., 1999, Iowa Water Summit, 2003). The current drainage research, therefore, is challenged to develop methods to design and manage subsurface drainage systems to maximize crop production while minimizing NO₃-N leaching or in other words “their detrimental water quality impacts”.

In recent decades, researchers have devoted much effort developing hydrologic models to analyze such environmental and water use problems. Though hydrologic models are based on simplified mathematical representations of complex processes in the natural system, they are now reasonably capable of integrating any change to the system, and offer a wide range of applications in the field of drainage water management and environmental protection. They are useful for understanding crop water requirements, crop growth, drainage design, and leaching of solutes and pesticides. Example of these models are DRAINMOD (Skaggs, 1978, Skaggs, 1982) and ADAPT (Agricultural Drainage and Pesticide Transport Model) (Alexander, 1988, Schalk, 1990). DRAINMOD is one of the most widely used hydrologic models to simulate subsurface (tile) drainage systems. It simulates surface runoff, infiltration, evapotranspiration, subsurface drainage and seepage from the soil profile. The U.S. Natural Resources Conservation Service has adopted DRAINMOD for design and evaluation of water management systems.

Though DRAINMOD has been successfully used for different soil, water and crop conditions (Chang et al., 1983, Fouss et al., 1989, Breve et al., 1998, Helweig et al., 2002, Wang et al., 2006), its prediction capability depends upon the proper identification of the input parameters. In representing natural conditions, DRAINMOD conceptualizes and aggregates relatively complex processes through simplified and approximate methods, and hence might contain a certain extent of modeling error in the predictions. Prediction error might increase with erroneous and uncertain input parameters. Most of the input parameters of DRAINMOD are measured directly in the field experiments with high accuracy, some remain uncertain such as soil hydraulic parameters. Point scale field measurements of soil hydraulic parameters are difficult, time-consuming, expensive, and have limitations to represent their spatial heterogeneity on field scale, the scale of model application. Studies have suggested that soil hydraulic parameters should be estimated and used in model predictions at the same scale (Abbasi et al., 2004). This can be accommodated through the process of model calibration and validation using field measurements. Model calibration is the process of determining indirectly the uncertain input parameters, where the predicted state variables are compared with the observed state variables. Comparing predicted and observed state variables for an independent dataset validate the indirectly determined input parameters.

Except one study (Sanoja et al., 1990), there has been little work done to calibrate and validate DRAINMOD for Iowa's tile landscapes. The present study, therefore, has been designed to calibrate and validate DRAINMOD to predict subsurface drainage for two predominant soils in Iowa's tile landscapes: Webster soil cultivated with continuous corn (WEBS_CC) and Canisteo soil cultivated with corn-soybean rotation (CANI_CS). More specifically the study explored the potential use of existing common data sources and techniques such as a soil database (ISPAID 7.0, 2004), pedotransfer function (ROSETTA model; Schaap et al., 2001) and a non-linear Parameter ESTimation (PEST) (Doherty et al., 1995) program in combination with DRAINMOD to arrive at an adequate fit of the predicted and observed subsurface drainage. The necessary field measurements were collected from experimental plots located in Pocahontas County, Iowa. Performance of DRAINMOD to predict subsurface drainage was evaluated through different statistical measures such as root mean square error, coefficient of mass residual, index of agreement (Willmott, 1982) and model efficiency (Nash and Sutchliffe, 1970), which quantify the differences in the predicted and observed subsurface drainage.

Calibrated and validated DRAINMOD can then be used to extrapolate field investigations to prepare guidelines for future drainage research and policy decisions. In Iowa's tile landscapes, traditionally single lines of tile were installed at different spacing and angles to drain “wet spots” in a field. In recent years, farmer's interest in the use of pattern drainage, installing tiles in a grid pattern through out the field, has increased in order to improve drainage conditions and maximize crop production. With this comes a question of what should be the optimum drain depth and spacing, which will depend upon topography, soil characteristics and climatic conditions in a certain region. In this study, as a potential use of the calibrated and validated DRAINMOD, we simulated different drain depth and spacings for WEBS_CC over a range of climatic variations from 1990 to 2003. Simulation results are analyzed to investigate the impact of different drainage designs on surface runoff, subsurface drainage and crop production in Iowa's tile landscapes.