Effect of mulching techniques on plot scale runoff: FDTF modeling and sensitivity analysis

International audience

English. Water balance of agricultural field are strongly modified by mulching techniques with crop residues as they act upon the production function (formation of the excess rainfall) and the transfer function (transport of the excess rainfall to the plot outlet). The aim of this paper is to investigate the effect of crop residues on the production and the transfer functions at the plot scale. Previous experiments conducted on runoff plots (20 m2) were analyzed using the first differenced transfer function (FDTF) model which is based on the unit hydrograph theory. Four treatments were analyzed: bare soil (T0), no till and no plant with 1.5 t haK1 of residue mulch (T1), direct drilling of corn with 1.5 and 4.5 t haK1 of residue mulch (T2 and T3, respectively). Hyetographs and runoff hydrographs were collected at a 20 s time step. Transfer functions and implicit production functions were estimated on each plot using an iterative algorithm for the calibration. Then, model precision was tested (i) by stochastic simulations to estimate the local uncertainty due to experimental errors, and (ii) by performing a bootstrap analysis to assess the global uncertainty due to the effect of rainfall event sampling. The calibrated functions were consistent with experimental data: for T0, the large amount of runoff and the small time to peak of the experimental hydrographs was represented by a sharp and narrow transfer function, whereas for T3 the small and delayed amount of runoff was represented by a broader and more spread transfer function. Also, model precision was slightly affected by measurement errors on experimental data, whereas rainfall event sampling generated substantial uncertainties on the transfer functions. The FDTF model proved powerful in analyzing the effect of mulch covers when runoff data is available. The model was able to separate the effect of predicting rainfall excess and the transport of the excess rainfall over the plot. Also, the bootstrap methodology appeared useful to assess uncertainties on model parameters due to rainfall event sampling.