Statistical analysis and simulation of subsurface drainage flow and nitrate-N loads

Xiuying Wang, Purdue University

Abstract

Artificial drainage has been an important water management practice for farming some of the most productive soils of the Midwest; however there have been environmental costs. The objectives of this study were to statistically determine the accuracy and precision of nitrate-N loss estimates from subsurface drainage based on various sampling frequencies, to predict subsurface drain flow, crop yield, and nitrate-N losses using DRAINMOD 5.0, which is a commonly used hydrology/water quality model, and to incorporate uncertainty analysis into modeling. A methodology was developed to determine the accuracy and precision of nitrate-N loss estimates from subsurface drainage plots based on various sampling frequencies. Results showed that the probability of estimating the annual nitrate-N mass loss within ±15% of the “true” mass loss was 92% for the weekly sample frequency, 68% for the monthly frequency, and 51% for the 90-day frequency. An automatic calibration framework was developed to calibrate DRAINMOD for predicting drain flow and nitrate-N losses. Fifteen years of data were used to test the hydrology component and five years of data were used to test the nitrogen component. Nash-Sutcliffe efficiency for drain flow predictions ranged from −0.66 to 0.81, with average deviations from 0.01 to 0.07 cm/day and standard errors from 0.03 to 0.17 cm/day. DRAINMOD correctly predicted the pattern of yearly relative yield change. The relative corn and soybean yields were well predicted based on average measure, with percent errors from 1.3 to 9.7% for corn and from −0.8 to 10.3% for soybean. The percent errors for yearly nitrate-N mass losses ranged from −16.6 to 97.5%. The Generalized Likelihood Uncertainty Estimation procedure was used to evaluate the uncertainties in DRAINMOD drain flow predictions associated with uncertainties in input parameters. The width of 90% confidence interval band of predicted drain flow ranged from 7.6 to 12.4 cm/yr, with predicted means of 8.3 to 20.2 cm/yr and standard deviations of 2.3 to 3.8 cm/yr. The most influential parameters were the vertical saturated hydraulic conductivity of the restrictive layer and the lateral hydrologic conductivity of the deepest soil layer. The main contributions to the prediction uncertainty are due to parameter interactions.

Degree

Ph.D.

Advisors

Kladivko, Purdue University.

Subject Area

Agricultural engineering|Environmental science

COinS