Phosphorus transport in lateral subsurface flow at forested hillslopes

This study is concerned with the transport of Phosphorus (P) with lateral subsurface flow in the organic layer 10 and topsoil of three forested headwater sites in Germany. Sampling frequency was set proportional to the incident flow rate in high temporal resolution. With this approach we want to investigate intra-event dynamics of P transport in lateral subsurface flow to establish initial process understanding about this potentially relevant pathway of P loss in forested hillslopes. With the organic layer being an important transfer site in the P cycle of temperate forests, availability and transportability of P in short timescales may reveal details about the overall balance of P in theses ecosystems. Our 15 results demonstrate that P concentrations in lateral flow are highly variable within and in between distinct flow events as well as among our study sites. To determine possible controls of the P transport we constructed multiple linear models of the P concentration in lateral flow as a function of site specific environmental datasets. Site affiliation was responsible for more than half of total explained variability regarding P concentration in lateral flow, followed by flow rate, electric conductivity of subsurface lateral flow 20


Introduction
Phosphorus is essential for all life and its availability often limits primary productivity of ecosystems (Elser et al. 2007; 25 Elser et al. 2000).Original P input into unfertilized, terrestrial ecosystems is mainly caused by weathering of bedrock material and likely to a lesser degree by atmospheric deposition (Buendia et al. 2010;Mahowald et al. 2008;Newman 1995;Tipping et al. 2014) Both fluxes of P into the ecosystem are generally orders of magnitudes smaller than P fluxes within the ecosystem, like plant uptake and litter fall (Cole and Rapp 1981;Gosz et al. 1972;Yanai 1992).This necessitates the existence of efficient mechanisms to reduce losses of P from the ecosystem (Hättenschwiler and 30 Vitousek 2000; Lang et al. 2016;Wood et al. 1984), which are thought to occur mostly through solutes and suspended particles in soil water seepage, groundwater and eventually stream flow (Cole and Rapp 1981).
The organic layer plays a key role in the P cycle of temperate forest ecosystems: It receives large P inputs through dead biomass from above and below ground sources as well as from atmospheric deposition, which are then available for biotic turnover.P can be removed from the organic layer again by biological uptake or leaching with vertical or lateral 85 issues of P cycling at the same sites as our study with regards to microbes (Bergkemper et al. 2016a,b;Bünemann, Augstburger & Frossard 2016;Zederer, Talkner, Spohn & Joergensen 2017), soil P species (Missong et al. 2016;Prietzel, Klysubun & Werner 2016;Lang et al. 2017), hydrology (Julich et al. 2016;Julich et al. 2017), plants (Netzer, Schmid, Herschbach & Rennenberg 2017) and the effect of liming and nitrogen input (Holzmann et al. 2016).

--Table 1 --
The capacity of lateral subsurface flow in forest soils to transport large amounts of P in short timescales has been demonstrated by field studies, though the reasons for high spatio-temporal variations in lateral flow P concentrations are so far not explicitly clear.(Cox et al. 2000;Fuchs et al. 2009;Timmons et al. 1977).Lateral subsurface flow events 95 show a high temporal variability in terms of flow rate (Bachmair and Weiler 2012a;Bachmair et al. 2009) and chemical composition (Burns et al. 1998;Hill et al. 1999;Newman et al. 1998), and may thus not be accurately represented by grab sampling, bulk sampling or suction-based sampling of soil water.To overcome these limitations, we sample free draining, unfiltered lateral flow in the organic layer with a temporal resolution from minutes to hours, to match temporal dynamics of other parameters related to lateral flow such as precipitation input and changes in soil moisture.

100
At each hill slope site, a 10 m wide trench was dug along the local contour line.Lateral flow from the organic layer, which varied in thickness from 10 to 25 cm, is captured in this trench and channeled to a sampling point.This is archived by placing a drainage mat to the up-slope wall of the trench, which is connected to a pipe at the bottom.After the installation of the drainage equipment, the trench was filled in again.A more detailed description of the building SOIL Discuss., https://doi.org/10.5194/soil-2018-13Manuscript under review for journal SOIL Discussion started: 4 June 2018 c Author(s) 2018.CC BY 4.0 License.undisturbed water samples, as no significant filtering occurs.When lateral flow is captured in the trench, the water is directed to a tipping bucket with 100 ml resolution and a 2 L reservoir below.The content of the reservoir is pumped to a flow through cell with a volume of 300 ml when a pre-defined volume of lateral flow water has accumulated.In the flow through cell, electric conductivity (EC) is measured with a Decagon CTD-sensor, after which a water sample is drawn to a 100 ml glass bottle in an enclosed autosampler.40 samples can be drawn by one autosampler at minimum 110 intervals of 10 minutes.The process is fully automated and all data is transmitted in near real time via the cellular network.
At the field site CON, a meteorological station is available in a distance of less then 100 m, at the site MIT the distance to the nearest meteorological station is approximately 500 m.At the site VES there was no meteorological station in the direct vicinity and the nearest DWD station 10 km away was used.These meteorological stations are used as sources for 115 time series of air temperature and precipitation.Soil moisture in the organic layer is measured in 3 locations, 2 m up slope of the trench installations.
The water samples were collected in the field twice a week at the site CON and every three weeks at the sites VES and MIT.Collection intervals of this length must be assumed to be too long to allow for the subsequent analysis of different P species, as biological transformation in untreated samples will change their distribution in much shorter timescales.As 120 a consequence, only total phosphorus of the unfiltered samples (P tot ) was determined.
While we did attempt to provide full coverage over the sampling period, equipment breakdowns or sample contaminationmostly by insects which drowned in the sample bottleswere a frequent occurrence.We successfully collected samples for about 40 % of the totally occurring flow events.
Samples are digested according to the US EPA ESS digestion method 310.2 (US EPA 1992) with the addition of sulfuric 125 acid (H 2 SO 4 ), ammoniumperoxidisulfate ((NH 4 ) 2 S 2 O 8 ) and pressurized heating to 130 °C in an autoclave.To solubilize P from possible biological growth on the surfaces of sampling bottles, part of the sulfuric acid used for the digestion was added directly into the sampling bottle, acidifying the sample to pH 0.7.The acidified samples are then shelved for three weeks and placed in to a mechanical shaker once every week, 12 hours at a time, after which the remaining steps of the digestion process are carried out.

130
After the digestion, sample pH is adjusted to pH 7 with ultra pure NaOH solution.This is done, because the pH of samples varied between 0.7 and 1.2 after the digestion, which can influence the analysis method and also cause interferences with Si and As.P tot is measured with the molybdenum blue method by Murphy and Riley (1962) adapted by Drummond and Maher (1995) using a Autolab 4 device by Green Eyes LLC.Every sample is measured 3 times, standards are interspersed at a 135 rate of 3 standards every 14 samples.Deviations among repeated measurements on the same sample were usually below 5 % of the measured value.From the repeated measurements, the range of the measurement error is calculated as the standard deviation.The error range for P loads in flow events is calculated as the accumulated total measurement error.
To avoid outside contamination during sampling and analysis as much as possible, only dedicated equipment, sample containers and lab space was used, which were cleaned with deionized water and phosphate free detergent before each 140 use.When the possibility of outside contamination was apparent, the respective samples were discarded.Possible systematic contamination was monitored through the inclusion of standards in every aspect of sample handling.This includes standards being placed in randomly chosen, open sample bottles in the autosamplers at the field sites for the time spans between sample collection dates.The standards are also subjected to the same treatment as the actual samples with regard to acidification, storage and digestion.From the point of collection to the final measurement of P tot , 145 a mass balance is documented for each sample to correct for evaporation losses during the digestion and dilution For each hillslope site, time series of lateral flow rate, electric conductivity (EC), air temperature and soil moisture in the organic layer were compiled for the observation period.Intra-event data and event-wise aggregated data (called inter-event data from here on) is used to construct multiple linear models with the P tot concentration in lateral flow as 150 the dependent variable.Depending on the parameter, aggregated inter-event data consists of sums (flow volume), arithmetic averages (flow rate, soil moisture, air temperature) or flow weighted averages (EC, P tot ) of the original data.
Flow events are defined according to two criteria; If only a single peak in lateral flow rate occurs within a distinct precipitation event and until the cessation of lateral flow, the whole period is regarded as a single event.If multiple flow peaks occur within a period of uninterrupted flow, the period may be split into multiple events.If the flow rate falls 155 below 10 % of the preceding flow peak before increasing again and if this increase corresponds to a prior increase in rainfall intensity, this is used as a cutoff point between events.
To identify environmental variables that may explain the observed intra-and inter-event variability in P concentration in the lateral subsurface flow and to test, whether similar relations can be established for all three sampling sites, multiple linear models are utilized.By comparing intra-with inter even data, we want to test to which extend our high resolution 160 sampling is helpful to improve process understanding of lateral P transport.
The parameter selection for inter-and intra-event models respectively was carried out by stepwise forward selection/backwards elimination of variables with respect to the Akaike Information Criterion (AIC) through the step()function in the R base-environment (R-Core-Team 2014).The full set of parameters from which the algorithm choses a reduced set, contains both "real time" indicators, such as lateral flow rate, EC and soil moisture, as well as "between-165 event" indicators like time duration between events, such as average soil moisture and temperature since the last event (see Tab. 3).The IDs of the three experimental sites are included as a categorical variable, which can also serve as an error term.The "between-event" parameters are exactly the same for both model approaches, while the "real-time" parameters are represented by average values in the inter-event model and the respective raw data in intra-event models.
The relative parameter importance was calculated according to (Grömping 2006) in the R-environment.170

Results
Our observations cover the full vegetation period at the the sites MIT and CON and start shortly after leafs shoot at the site VES (Tab.2).Most precipitation events > 1 mm produced measurable lateral flow, totaling 151 discrete subsurface flow events during the observation period on all 3 sites.Mostly due to field equipment breakdowns, only 62 of these 175 events with a total of 417 individual samples were successfully analyzed for P tot .
While the amount of total precipitation during the observation periods was relatively similar among the three sites, the amounts of lateral flow were very different, with 7 times higher flow volume at the site MIT in comparison to site CON and the site VES almost 3 times higher in comparison to CON (Tab.2).Since our design does not allow for the determination of the source area of lateral flow, it can only be assessed as a total amount, and not as a fraction of the 180 respective rainfall event.
--Table 2 --Soil moisture responds to precipitation events in most instances, but at the sites MIT and VES there is a regular 185 occurrence of precipitation events that do not seem to cause changes volumetric water content in the organic layer and topsoil.Overall, there is a marked decline in soil moisture in the organic layer from mid-June to the end of the SOIL Discuss., https://doi.org/10.5194/soil-2018-13Manuscript under review for journal SOIL Discussion started: 4 June 2018 c Author(s) 2018.CC BY 4.0 License.vegetation period at the sites MIT and CON (Fig. 1).The P tot concentration in the lateral flow samples ranges from 5 mg l-1 to below the detection limit of our method at 10 µg l-1.Highest P tot concentrations were predominantly found at the site CON, the lowest at the site VES.The range of P tot concentrations in samples from a single event can be as large 190 as the range of average event concentrations throughout the measurement period at a given site.The existing range of occurring P tot concentrations allowed for the estimation of P loads during events with only a few percent of uncertainty in most cases (Fig. 2).
The average event concentration of P tot in lateral flow does not show a pronounced seasonal pattern and can differ considerably among successive events (Fig. 1).

195
The variation of intra-event P tot concentrations appears more ordered, with recognizable periods of increase or decrease (Fig. 2).Still, the underlying processes are not immediately recognizable, as neither flow rate, soil moisture or EC appear to be in a consistent relation with the Ptot concentration.For some events, the P tot concentration decreases consistently over time, irregardless of changes in the flow rate, while being both positively and negatively correlated to --Figure 1 Results of the multiple linear models, fitted to both intra-and inter-event Ptot concentrations of lateral flow are displayed in Fig. 3 and Tab. 3. In both cases, differences between the study sites are responsible for most of the variability in P tot concentrations of lateral flow explained by the model.The final models, were parameters were selected according to the respective model AIC, contain different parameters for the intra-and inter-event data models.The total 210 explained variability is notably higher for the model fitted to inter-event data.

Discussion
Across our sites, most measurable precipitation events produced lateral flow in turn, even when no changes in soil moisture were recorded.This is a strong indication that lateral flow is an important characteristic of the hydrological system in the organic layer and topsoil of the experimental sites.Still, there are notable differences among the sites 215 regarding the generation of lateral flow following rainfall events.At the site MIT there is a regular occurrence of precipitation events in the summer months that do not coincident with an increase in soil moisture, but which do cause lateral flow.At the sites VES and CON, soil moisture is increased following rainfall events nearly without exception, while lateral flow is not always generated, especially in the drier summer months at the site CON.This indicates, that the fractions of rainfall that contribute to lateral and vertical flow respectively may differ between the study sites 220 depending on the hydrological setting.The observation that generation of lateral flow is more pronounced at the site MIT than at the other sites is supported further by the observation, that the total amount of lateral flow relative to rainfall is by far highest at this site.The relatively high yields of lateral flow at the site MIT despite the fact that this sites has by far the lowest slope angle (see Tab. 1) could be an indication of hydrophobicity in the organic layer and topsoil, which might prevent the soil material from retaining water, which in turn would increase potential for lateral 225 subsurface flow.The possible occurrence of hydrophobicity in the dry period at the site MIT might also explain that average P concentrations in lateral flow at this site are significantly lower in the dry period compared to the wet period before: Lower contact times of infiltrating water with soil material as well as a smaller contact area and less exchange The very high temporal variability of the flow rate, with peaks often lasting for some minutes only, confirms the need for high temporal resolution of measurements to adequately observe this parameter.Depending on the research in question, precipitation should also be measured in a comparative resolution and directly on site.Our data shows, that 30 minute resolution in precipitation measurements is not enough to match the temporal variability of lateral flow generation.At the site MIT, a distance of less than two kilometers between our field site and the meteorological station 235 apparently was high enough to cause situations where lateral flow was measured up to an hour before precipitation onset at the rain gauge.At he field site CON, where the distance between our field site and the meteorological station was only about 100 m and the temporal resolution of the precipitation measurements was 5 minutes, deviations between the onset of lateral flow and precipitation were very low.These differences in the quality of precipitation data among out sites prohibited us from using precipitation as a model parameter.

240
Our data clearly demonstrates the existence of high temporal variability in P tot concentrations over the course of single events as well as in average concentrations among distinct flow events.This shows that bulk sampling of single lateral flow events does not allow for generalization about average event conditions.Intra event changes in subsurface flow P tot concentration do seem to represent orderly patterns with distinct phases of P tot concentration in-or decreases.While outliers, where the P tot concentration of single samples highly deviates from the rest of the event do exist, a sampling 245 frequency from 10 to 20 minutes appears to be high enough to capture the overall event P tot dynamic.
Overall, lateral flow sampling in high temporal resolution actually decreases the ability of our model to predict P tot concentrations in lateral flow when compared to the model fitted to aggregated data, which is meant to represent the results of a fictional bulk sampling approach.The high significance of both models and the minor difference between R² 250 and adj.R² in the respective models (see Tab. 2) make it appear unlikely, that the additional explained variability in the inter-event data model is simply due to a reduction in sample size, or over-parametrization. Including terms for interactions between the model co-variables soil moisture, temperature and time span in between events did not significantly improve either model.
The AIC-dependent parameter selection algorithm chose somewhat different parameters for intra-and inter-event 255 models.Both model structures are alike in such a way, that differences between sites (included in the model as a categorical variable) are responsible for more than half of the total explained variability.This implies that under given conditions, consistent differences in lateral subsurface flow P concentrations exist among our experimental sites.
For both models, antecedent soil moisture correlates significantly with P tot concentration in lateral flow.Since the estimate of this parameter is positive in both cases, the models suggests that higher antecedent soil moisture is 260 associated with higher P tot concentration in lateral flow under given conditions.While the cumulative within-event flow volume is an important parameter in the intra-event model, total event volume is not significantly correlated to average event concentration of P tot in the inter-event model.This indicates, that while high-volume events are not significantly associated with overall lower Ptot concentrations, there is a significant decrease of the P tot concentration during the progression of individual events, leading on average to higher concentrations at the event start in comparison to the end.

265
Another notable discrepancy between the two models is the effect of EC, with accounts for 18 % explained variability in the inter-event model, but only for 4 % in the intra-event model.This would imply, that the average values of EC during events are correlated to a higher degree to average flow and P tot concentration, than their respective variation on intra-event timescales.as the low significance and explained variability of the actual flow rate in the intra-event model suggest, that immediate mobilization of transportable P forms in the organic layer may not be a critical process in lateral flow P tot transport and that higher flow rates may rather cause dilution of P in lateral flow than lead to increased mobilization.
The fact that neither the timespan, nor the average air temperature between lateral flow events contribute significantly to either of the two models has interesting implications for the question as to where the P in our lateral flow samples 275 originally came from.Precipitation is unlikely to be the main source, as typical concentrations of P tot in throughfall are much lower than in our lateral flow samples (Sohrt et al. 2017).Also, since higher flow rates are not associated with higher P tot concentrations, transport of particulate matter in the organic layer does not appear to be a critical process for P tot transport.What remains as a possible source for the P tot in our lateral flow samples, is leaching of highly transportable P forms from organic layer material.This leaching is thought to occur through degradation of organic 280 layer material by plants and microbes which convert P bound to organic matter to soluble forms.Since microbial activity is heavily influenced by water availability and temperature, we expected periods with higher temperatures to be associated with higher P tot concentrations in lateral flow.Our results do not show such a relation with mean air temperature, though, higher soil moisture does lead to higher P concentrations in lateral flow.The fact that the time spans between events are not significantly correlated to P tot concentrations in lateral flow indicates, that little or no 285 accumulation of water transportable P in the organic layer occurs in between flow events.

Conclusions
The generation of lateral flow in the organic layer in response to rainfall is very heterogeneous among our study sites, and so is the P tot transport associated to it.Average P tot concentration in lateral flow was not proportional to P 290 concentration or -content of the organic layer material at the respective site.This strongly suggests, that generalizations about lateral flow and the associated P transport from case studies into a broader context have to be treated with caution.This is empathized by our model results in which the site ID, which may also serve as a site specific error term, was by far the most significant parameter regarding P transport in the organic layer.
The fact that the set of environmental data we used to explain the P transport only accounted for 20-30 % explained 295 variability demonstrates, that our understanding of its underlying causes is still limited.What our results clearly show is, that simple scenarios for P transport in this context are not feasible.Neither water flow as the mobilizing force, nor accumulation of transportable P in the organic layer in between events can be supported as primary regulators for P transport in lateral flow by our study.Our advice to future researchers on this topic would be to focus on detailed case studies and controlled experiments in the field, rather than on large spatial and temporal coverage.This would allow for 300 the use of artificial tracers as well as the distinction between different P species in lateral flow, which may help in determining the drivers of organic layer P transport with later flow.
EC (Fig 2 B,C,D).In other events, multiple peaks in the Ptot concentration are evident without a clear relation to flow 200 peaks in the flow rate (Fig 2 A,E).
The negative correlation of the maximum flow rate during a respective event with P tot in the inter-event model, as well 270 Page 7 SOIL Discuss., https://doi.org/10.5194/soil-2018-13Manuscript under review for journal SOIL Discussion started: 4 June 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 1 Figure 2
Figure 1 Overview of precipitation, soil moisture, rate of lateral flow, mean (red dots) and range (error bars) of P tot concentration in lateral flow at the study sites Mitterfels (a), Vessertal (b) and Conventwald (c) during the observation period.

Figure 3
Figure 3 Intra-event data (a) and inter-event data data (b) of the P tot concentration in lateral flow, and results of the best-fit multiple linear model constructed from our data set.For model parameterization, p-values and explained variability per parameter see Tab. 3.

Table 1
Stand parameters of our field sites.

Table 2
Length of observation period, total observed precipitation and lateral flow at the three study sites during the period.

Table 3
Parameterization of multiple linear models to predict the P tot concentration in lateral flow.The full parameter set as well as the optimized parameter selection according to minimized AIC and their corresponding estimate, p-value and individual explained variability are shown.Parameters written in bold are included in the final model structure.SOIL Discuss., https://doi.org/10.5194/soil-2018-13Manuscript under review for journal SOIL Discussion started: 4 June 2018 c Author(s) 2018.CC BY 4.0 License.