On the improvement of catchment scale simulations of nitrate transport through tile drains

Abstract

Nitrate transport from cultivated areas poses a significant risk to water quality of inland and marine water bodies. During subsurface transport nitrate may undergo reduction under anaerobic conditions. However, in temperate climates tile drainage are widely used in agriculture, which provides short-circuits between the root zone and surface water systems, with limited or no nitrate reduction. To identify areas most prone to loss of nitrate to the aquatic environment, it is thus vital to assess and quantify not only the transport out of the root zone, but also the fraction of nitrate being transport by tile drains vs. groundwater transport. The spatio-temporal pattern of tile drainage can be estimated by use of physically-based distributed hydrological models, but their setup and evaluation are generally challenged by limited data on drains with respect to both the tile drain network and in particular with respect to the efficiency of the drains, i.e. the amount of recharging water that is transported via drains. To support water management, the models must cover relevant scales (100 – 1000 km2) posing an additional upscaling modelling challenge. As part of nitrogen usage regulation in Denmark, a national nitrogen model has been developed, which is currently under revision. An important task is to improve the description of drain transport. This is achieved through detailed hydrological modelling of fields with drain flow observations from which drain fractions, i.e. the fraction of precipitation being drained, are calculated for each model grid. A machine learning algorithm (gradient boosted decision tree) is then used to regionalise the drain fraction to the national scale. Results are used in model calibration to improve the spatial and temporal description of drain flow. While the drainage estimates are needed at a fine scale, preferably at grid scale, data to evaluate model accuracy in terms of nitrate transport is not available at this scale. At catchment scale, the seasonal dynamics of observed nitrogen transport in streams provide valuable information on the amount and temporal variation of the contributions from drains. Analyses of the observed time series are used to further constrain nitrate drainage transport at catchments scale. Uncertainty in model results is assed using a stochastic approach calling for numerous model runs. To limit computational time, model simulations are carried using different spatial resolutions (100 and 500 m grids), where the coarser model is run for an entire 30-year period while the finer resolution is only run for a decade. Results from the overlapping simulation period is used for tuning a downscaling of the drain flow in 500 to a 100 m resolution, thereby providing model results of nitrate transport via drainage at a 100 m resolution for the entire 30 years.