Historic record of pasture soil water and the in ﬂ uence of the North Atlantic Oscillation in south-west England

The North Wyke Farm Platform for sustainable grassland research in south-west England contains infrastructure measuring soil moisture and ﬁ eld runoff. Its time series of sensor data is used to validate the parsimonious SH 2 O-NW model for soil water at ﬁ eld-scale. Thirty-four years of daily soil moisture and runoff is simulated, and used to detect long-term trends and produce a risk analysis. The model accounts for wetter periods of soil moisture and the main summer soil de ﬁ cit and autumn re-wetting; limitations involve short-term, rapid changes in drying and re-wetting. The soil moisture sensor observations however do not re ﬂ ect ﬁ eld variability. Analysis of more than one ﬁ eld allows an assessment of unexpected sensor anomalies. The paper recommends that soil moisture sensor con ﬁ dence levels be provided, for comparison against modelled data. The simulations show a historic reduction in the occurrence of summer soil moisture de ﬁ cits above a third of water capacity, while the winter precipitation and runoff simulation shows a stable long-term trend, matching the direction and magnitude of the North Atlantic Oscillation Index. A large runoff of 400 m 3 /day from a 1.75 ha pasture has a 0.07% probability, having a return period of once in 4 years during the 34-year period.


INTRODUCTION
The year 2010 tied with 2005 was the warmest year on record globally (NOAA ). Rising frequency of heavy downpours is an expected consequence of a warming climate. Some areas will see more droughts as overall rainfall decreases and other areas will experience heavy precipitation more frequently, or see rain come in rarer, more intense bursts (Huber & Gulledge ).

Field investigations between 2001 and 2011 identified
widespread structural degradation of 38% of intensively managed agricultural soil surveyed in south-west England (Palmer & Smith ). Findings showed surface water runoff was enhanced, increasing the risk of flooding. The loamy stagnogley soils were one of the most frequently damaged soils. Soil moisture is a medium for studying the overall balance of changes in precipitation with changes in temperature.
Runoff data, on the other hand, can allow us to analyse how frequently to expect overland flow constituting a risk.
Soil water models are often categorized in terms of their degree of complexity based on the treatment of the soil profile, in addition to the number of processes employed (Ranatunga et al. ). Relatively simple models may have a fixed number of soil layers and a tipping bucket approach to water inflows and outflows, while relatively more complex models seek to incorporate a continuous soil profile. Within the simple (or fixed soil layer) modelling category, models are divided into single layer or multiple layer approaches.
The simplest types of soil water flow models act as tipping buckets. They ignore the vertical moisture gradient within the root zone (Feddes & Raats ), to discharge water from one layer to another when the water carrying capacity of the soil layer is exceeded. It is generally accepted that the Richards' equation (Richards ) is used to improve upon tipping bucket models incorporating Darcy's law for solute transport and capillary action (Feddes & Raats ). At a field scale with sufficient observation data for calibration and validation, a tipping bucket model with minimal requirements of parameterization can be useful (Walker & Zhang  It is proposed to use automated instrumentation to provide good quality, continuous observations, to allow a robust model calibration. Without any change in land use or field management, this field-scale study assesses the extent to which a relatively simple water model can be used requiring minimal parameterization. Applying the model to create long-term soil moisture and runoff datasets, the historic trends of soil moisture deficit and runoff are determined and a risk assessment is produced for the probability of runoff occurrence.
The main study of observed and simulated data is carried out on a field, using Longlands South as a case study; however, Wyke Moor is used as a secondary check of the simulation accuracy.

Site description and data sources
The North Wyke Farm Platform (NWFP) (Orr et   Moor is an upland with no surrounding higher ground, and the single source of runoff was from rainfall before these fields were hydrologically isolated. The fact that they have been iso- field studies, operates on a daily timestep and determines soil moisture, and also drainage and runoff from soil.
Soil moisture in the root zone is determined by a water balance: • The model assumes that rainfall is the only source of water input to the soil.
• The effective rainfall is calculated by subtracting surface runoff from rainfall, surface runoff is calculated according to SCS runoff curves (USDA-SCS ) created using the observed precipitation and observed field runoff.
• Water loss from evapotranspiration (ET) is subtracted from the effective rainfall calculated using a modified • A tipping bucket mechanism is employed, i.e., if the effective rainfall is higher than potential ET it replenishes the soil moisture. Soil moisture above field capacity becomes drainage and is lost from the system, and the soil remains at full water holding capacity.  two-tailed, 5% confidence limit (CL)) to see whether there is any significant bias in the simulated values compared to the observed values.

Frequency analysis
Since the NWFP was created in 2011, its high quality continuously measured data are excellent for validation of a model, but the time period covered will, for a long time, be too short to use the data directly in a daily frequency analysis. Long-term records, or simulations from applying long-term climate records, are essential for risk assessment.
A risk assessment provides a likelihood of occurrence to the modelled impacts, and puts 34 years of soil moisture and runoff data into context. The two issues are that there is an increasing risk of a soil moisture deficit (most commonly occurring on a short-term basis during summer) and conversely that there is an increasing risk of runoff during winter.
A cumulative frequency analysis is used (Oosterbaan ) to determine the risk of exceedance of the data thresholds: 1. Twenty-two data threshold intervals between 30 and 690 m 3 runoff per day chosen for the amount of volumetric soil moisture deficit below field capacity, and for daily runoff. As the frequency of events is 1 or 0 near the upper limit of runoff, the intervals are wider. To detect for a progressive change in the climate record, the full dataset of 34 years was divided up into three 11-year periods (1982-1992, 1993-2003 and 2004-2015) and com- Seven of the ten highest rainfall years occur after 1998. There is a likely 94% confidence of increasing minimum temperature in summer. There were no trends detected in winter or spring temperatures or with precipitation.
In terms of agricultural management, the progressive    Depending on soil moisture data from one centrally located sensor is over-simplifying the system and making assumptions because the soil moisture data will vary across the field, and neither the process model, nor the moisture sensors, account for spatial variability. A denser network of point location sensors may be desirable, but in reality on a farm, one sensor per field requiring protection from trampling by cattle is a practical option. Soil moisture and runoff data for this project are taken from the sensor data downloaded from the data portal of the NWFP (https://nwfp.rothamsted.ac.uk/), not from fieldwork.
Therefore it would be advisable to have a statistical measure of confidence for the sensor data, but in this study parameters were not yet available to calculate this.   Table 3 for the simulated-observed runoff comparisons for 2012, since runoff data were scarce in the drier year of 2013.
Runoff is more variable and less easy to simulate than moisture. The simulation under-predicts, which may be due to under-prediction of the runoff curve in the model, or the assumption of a tipping bucket mechanism for a field layer, i.e., when a soil layer is at full water carrying capacity, it allows excess water to drain, in these cases laterally to the drainage channels. The model has limitations in that it assumes a homogeneous vertical soil layer, which covers up heterogeneity of the soil. There is also an unknown element to how leak-proof the field system is, especially after a prolonged dry period on clays in which cracks have developed.
Rainfall-observed runoff ( Figure 8) shows a non-uniform relationship below 6 mm of rainfall which makes the linear nature of the SH 2 O-NW model's processes more applicable to runoff from precipitation over 6 mm.
The runoff output from a DayCent simulation was also compared against observed data (Table 3). The high correlation coefficient but also relatively high errors reflect the high association between simulation and observation but   In the south-west of England, risk assessment in terms of field runoff can be linked to flooding. A risk assessment for   A frequency analysis adds probability to the simulated runoff ( Figure 11(a)). From this the return period ( Figure 11(b)) for different thresholds has been calculated based on the data of Longlands South from 1982 to 2015.
The runoff frequency is based on data which include periods of intense short-term flooding, so the higher range refers to reasonably severe runoff thresholds with long return   give unexpected anomalies, and cannot reflect the variability in moisture around the field that has been obtained by soil sampling measurements. Using more than one field in a study allows an assessment of whether anomalies are due to sensor or model. It is recommended that CL are provided for the soil moisture sensor data, to compare modelled data against.
The total field drainage was measured at the flume, so it does not encounter the problems of spatial variability seen with sensing soil moisture, thus deviation in simulation compared to observed runoff would most likely be due to the model. Runoff validation is satisfactory above 6 mm rainfall, but below that there is a non-uniform relationship, so overall the model is more appropriate to wetter conditions or years. On Wyke Moor, while model performance for soil moisture slightly improved compared to Longlands South, model performance for runoff slightly reduced.
In comparing our model against DayCent for its soil moisture simulation, we have used a globally known model many agricultural scientists will be familiar with employing a more sophisticated water balance involving multiple soil layers, sub-daily timesteps and using a version of Darcian unsaturated water flow. This model compared favourably with our model. On the whole, the results support the hypothesis that you can use a simple model to obtain a satisfactory water balance at field-scale to assess annual and seasonal patterns and trends, and also that the climate and winter runoff are influenced by the NAO. A useful addition would be to implement a Darcian unsaturated water flow by either the Richards equation or the Green-Ampt equation to account for upward water flow.
The model was applied to the 34-year historic time series of climate to produce a simulated soil moisture and field runoff history of Longlands South pasture. The historic climate influencing the soils of North Wyke and the soil runoff has been shown to track the NAO. The pattern for the whole 34-year period shows longer consecutive years of temporary summer deficits and no or little summer runoff during the 1980s and 1990s, and mixed wetter and drier summers over consecutive years since. This is supported by the literature (Marsh ); there are reports that southern England had increased soil moisture deficits from 1988 to 1992 and from 1995 to 1997, but that above average rainfall since mid-1997 has counterbalanced any higher evaporative demands.
A Mann-Kendall trend analysis shows that the occurrence in the number of soil deficits per year below a third of water carrying capacity has decreased over 34 years but shows a stable trend for runoff consistent with the stable trend in precipitation. There are indications of a progressive historical rise in minimum temperature.
The model was used in a risk assessment to assess the likelihood of varying degrees of soil water runoff. A very heavy runoff which we would expect to cause localized flooding of 400 m 3 /day from one field has a 0.07% probability, which makes its return period 1 in 1,339 days or just less than 4 years during the 34-year period.