Value and limitations of machine learning in high-frequency nutrient data for gap-filling, forecasting, and transport process interpretation.

High-frequency monitoring of water quality in catchments brings along the challenge of post-processing large amounts of data. Moreover, monitoring stations are often remote and technical issues resulting in data gaps are common. Machine learning algorithms can be applied to fill these gaps, and to a certain extent, for predictions and interpretation. The objectives of this study were (1) to evaluate six different machine learning models for gap-filling in a high-frequency nitrate and total phosphorus concentration time series, (2) to showcase the potential added value (and limitations) of machine learning to interpret underlying processes, and (3) to study the limits of machine learning algorithms for predictions outside the training period. We used a 4-year high-frequency dataset from a ditch draining one intensive dairy farm in the east of The Netherlands. Continuous time series of precipitation, evapotranspiration, groundwater levels, discharge, turbidity, and nitrate or total phosphorus were used as predictors for total phosphorus and nitrate concentrations respectively. Our results showed that the random forest algorithm had the best performance to fill in data-gaps, with R2 higher than 0.92 and short computation times. The feature importance helped understanding the changes in transport processes linked to water conservation measures and rain variability. Applying the machine learning model outside the training period resulted in a low performance, largely due to system changes (manure surplus and water conservation) which were not included as predictors. This study offers a valuable and novel example of how to use and interpret machine learning models for post-processing high-frequency water quality data.

Transport characteristics of DNA-tagged silica colloids as a colloidal tracer in saturated sand columns; role of solution chemistry, flow velocity, and sand grain size.

In recent years, DNA-tagged silica colloids have been used as an environmental tracer. A major advantage of this technique is that the DNA-coding provides an unlimited number of unique tracers without a background concentration. However, little is known about the effects of physio-chemical subsurface properties on the transport behavior of DNA-tagged silica tracers. We are the first to explore the deposition kinetics of this new DNA-tagged silica tracer for different pore water chemistries, flow rates, and sand grain size distributions in a series of saturated sand column experiments in order to predict environmental conditions for which the DNA-tagged silica tracer can best be employed. Our results indicated that the transport of DNA-tagged silica tracer can be well described by first order kinetic attachment and detachment. Because of massive re-entrainment under transient chemistry conditions, we inferred that attachment was primarily in the secondary energy minimum. Based on calculated sticking efficiencies of the DNA-tagged silica tracer to the sand grains, we concluded that a large fraction of the DNA-tagged silica tracer colliding with the sand grain surface did also stick to that surface, when the ionic strength of the system was higher. The experimental results revealed the sensitivity of DNA-tagged silica tracer to both physical and chemical factors. This reduces its applicability as a conservative hydrological tracer for studying subsurface flow paths. Based on our experiments, the DNA-tagged silica tracer is best applicable for studying flow routes and travel times in coarse grained aquifers, with a relatively high flow rate. DNA-tagged silica tracers may also be applied for simulating the transport of engineered or biological colloidal pollution, such as microplastics and pathogens.

Climate variability effects on eutrophication of groundwater, lakes, rivers, and coastal waters in the Netherlands.

Many aquatic ecosystems in densely populated delta areas worldwide are under stress from overexploitation and pollution. Global population growth will lead to further increasing pressures in the coming decades, while climate change may amplify the consequences for chemical and ecological water quality. In this study, we explored the effects of climatic variability on eutrophication of groundwater, streams, rivers, lakes, estuaries, and marine waters in the Netherlands. We exploited the relatively dense monitoring information from the Dutch part of the Rhine-Meuse delta to evaluate the water quality response on climatic variability, in combination with anthropogenic pressures. Our results show that water quality of all water systems in the Netherlands is affected by climate variability in several ways: 1) through the process of global climate change (mainly temperature and sea level rise), 2) through changes Atlantic ocean circulation patterns (more southwestern winds), 3) through changes in continental precipitation and river discharge fluctuations, and 4) through local climatic fluctuations. The impact of climate variability propagates through the hydrological system 'from catchment to coast'. The fluctuations in water quality induced by climatic variability shown in this study give a preview for the potential effects of climate change.

Urban hydrogeology: Transport routes and mixing of water and solutes in a groundwater influenced urban lowland catchment.

Urban areas in coastal lowlands host a significant part of the world's population. In these areas, cities have often expanded to unfavorable locations that have to be drained or where excess rain water and groundwater need to be pumped away in order to maintain dry feet for its citizens. As a result, groundwater seepage influences surface water quality in many of such urban lowland catchments. This study aims at identifying the flow routes and mixing processes that control surface water quality in the groundwater-influenced urban catchment Polder Geuzenveld, which is part of the city of Amsterdam. Geuzenveld is a highly paved urban area with a subsurface rain water collection system, a groundwater drainage system, and a main surface water system that receive runoff from pavement and roofs, shallow groundwater and direct groundwater seepage, respectively. We conducted a field survey and systematic monitoring to identify the spatial and temporal variations in water quality in runoff, ditch water, drain water, and shallow and deep groundwater. We found that Geuzenveld receives a substantial inflow of deep, O2-depleted groundwater, which is enriched in ammonium and phosphorus due to the subsurface mineralization of organic matter under sulfate-reducing conditions. This groundwater is mixed in the ditches during wet periods with O2-rich runoff, and iron- and phosphate-rich drain water. Unlike natural catchments, the newly created, separated urban flow routes lead to mixing of water in the main surface water itself, shortcutting much of the soil and shallow subsurface. This leads to low O2 and high ammonia concentrations in dry periods, which might be mitigated by water level management or artificially increasing O2 levels by water inlet or artificially aeration of the main water canals. Further research is necessary how to optimize artificial urban systems to deliver a better ecological and chemical status of the surface water.

Passive Dosing of Organic Substrates for Nitrate-Removing Bioreactors Applied in Field Margins.

Denitrifying bioreactors are dependent on organic matter supply as a substrate for effective NO removal. In this study, the difference in removal efficiency and side effects when using different organic matter sources and dosing strategies was tested in two field experiments. The organic matter sources tested were woodchips and ethanol. The effect of woodchips was tested using woodchip-enveloped drains. Ethanol was supplied to a flow-through reactor by passive dosing by diffusion through silicone tubing. The woodchip-enveloped drains showed a removal efficiency of 80% during the first year of application, but this rate decreased during the second and third years of application, coinciding with a decrease in dissolved organic C and an increase in redox potential. The removal efficiency was higher and remained higher over a longer period of time when the drains were installed more deeply. The flow-through reactor with ethanol could lead to a higher removal efficiency (up to 95%) at higher hydraulic retention time (HRT, 0.1 d) than the woodchip-enveloped drains (HRT = 5 d). Passive dosing of organic substrates is simple, needs little maintenance and no energy, and can be performed independent of electricity. A denitrifying bioreactor with a controlled drainage inlet and outlet is a promising setup for optimizing N removal and minimizing side effects.

Current Insights into the Effectiveness of Riparian Management, Attainment of Multiple Benefits, and Potential Technical Enhancements.

Buffer strips between land and waters are widely applied measures in diffuse pollution management, with desired outcomes across other factors. There remains a need for evidence of pollution mitigation and wider habitat and societal benefits across scales. This paper synthesizes a collection of 16 new primary studies and review papers to provide the latest insights into riparian management. We focus on the following areas: (i) diffuse pollution removal efficiency of conventional and saturated buffer strips, (ii) enhancing biodiversity of buffers, (iii) edge-of-field technologies for improving nutrient retention, and (iv) potential reuse of nutrients and biomass from buffers. Although some topics represent emerging areas, for other well-studied topics (e.g., diffuse pollution), it remains that effectiveness of conventional vegetated buffer strips for water quality improvement varies. The collective findings highlight the merits of targeted, designed buffers that support multiple benefits, more efficiently interrupting surface and subsurface contaminant flows while enhancing diversity in surface topography, soil moisture and C, vegetation, and habitat. This synthesis also highlights that despite the significant number of studies on the functioning of riparian buffers, research gaps remain, particularly in relation to (i) the capture and retention of soluble P and N in subsurface flows through buffers, (ii) the utilization of captured nutrients, (iii) the impact of buffer design and management on terrestrial and aquatic habitats and species, and (iv) the effect of buffers (saturated) on greenhouse gas emissions and the potential for pollution swapping.

Groundwater-surface water relations in regulated lowland catchments; hydrological and hydrochemical effects of a major change in surface water level management.

In lowland deltas with intensive land use, surface water levels are human controlled letting in river water during dry periods and discharge by pumping during wet periods. The water levels are usually maintained at a fixed level year-round or at fixed winter and (higher) summer levels. Several water authorities in The Netherlands consider implementing a more natural and flexible water level regime in nature reserves, with low levels in summer and high levels in winter. The objective of this study was to assess the catchment-scale hydrological and hydrochemical effects of such a change using water and solute balance modeling. We focus on ten study nature reserves where a conversion to flexible water management was planned or recently implemented. Monitoring data from the catchments were used for validating the water balance and as boundary condition input for the solute balance calculations. For all catchments, the results show relevant changes after implementing flexible water level management. For example, the surface water residence times increased (avg. +25%), the inlet and outlet fluxes reduced (avg. -38% and -72%), the chloride concentrations reduced (avg. -14%), and the N-tot concentrations increased (avg. +13%). Both the quantification of water flux changes and the detection of water quality risks were highly relevant for the water authorities. Customizing our approach to the specific circumstances in other low-lying artificial catchments worldwide may help local water managers in optimizing their water level management.

Improving nitrate load estimates in an agricultural catchment using Event Response Reconstruction.

Low-frequency grab sampling cannot capture fine dynamics of stream solute concentrations, which results in large uncertainties in load estimates. The recent development of high-frequency sensors has enabled monitoring solute concentrations at sub-hourly time scales. This study aimed to improve nitrate (NO3) load estimates using high-resolution records (15-min time interval) from optical sensors to capture the typical concentration response to storm events. An empirical model was developed to reconstruct NO3 concentrations during storm events in a 100-km2 agricultural catchment in Germany. Two years (Jan 2002 to Dec 2002 and Oct 2005 to Sep 2006) of high-frequency measurements of NO3 concentrations, discharge and precipitation were used. An Event Response Reconstruction (ERR) model was developed using NO3 concentration descriptor variables and predictor variables calculated from discharge and precipitation records. Fourteen events were used for calibration, and 27 events from four periods of continuous records of high-frequency measurement were used for validation. During all selected storm events, NO3 concentration decreased during flow rise and increased during the recession phase of the hydrograph. Three storm descriptor variables were used to describe these dynamics: relative change in concentration between initial and minimum NO3 concentrations (rdN), time to maximum change in NO3 concentration (TdN) and time to 50% recovery of NO3 concentration (TN rec ). The ERR consisted of building linear models of discharge and precipitation to predict these three descriptors. The ERR approach greatly improved NO3 load estimates compared to linear interpolation of grab sampling data (error decreased from 10 to 1%) or flow-weighted estimation of load (error is 7%). This study demonstrated that ERR based on a few months of high-resolution data enables accurate load estimates from low-frequency NO3 data.

Sensors in the Stream: The High-Frequency Wave of the Present.

New scientific understanding is catalyzed by novel technologies that enhance measurement precision, resolution or type, and that provide new tools to test and develop theory. Over the last 50 years, technology has transformed the hydrologic sciences by enabling direct measurements of watershed fluxes (evapotranspiration, streamflow) at time scales and spatial extents aligned with variation in physical drivers. High frequency water quality measurements, increasingly obtained by in situ water quality sensors, are extending that transformation. Widely available sensors for some physical (temperature) and chemical (conductivity, dissolved oxygen) attributes have become integral to aquatic science, and emerging sensors for nutrients, dissolved CO2, turbidity, algal pigments, and dissolved organic matter are now enabling observations of watersheds and streams at time scales commensurate with their fundamental hydrological, energetic, elemental, and biological drivers. Here we synthesize insights from emerging technologies across a suite of applications, and envision future advances, enabled by sensors, in our ability to understand, predict, and restore watershed and stream systems.

Continuous 'Passive' flow-proportional monitoring of drainage using a new modified Sutro weir (MSW) unit.

In view of their crucial role in water and solute transport, enhanced monitoring of agricultural subsurface drain tile systems is important for adequate water quality management. However, existing monitoring techniques for flow and contaminant loads from tile drains are expensive and labour intensive. The aim of this study was to develop a cost-effective and simple method for monitoring loads from tile drains. The Flowcap is a modified Sutro weir (MSW) unit that can be attached to the outlet of tile drains. It is capable of registering total flow, contaminant loads and flow-averaged concentrations. The MSW builds on a modern passive sampling technique that responds to hydraulic pressure and measures average concentrations over time (days to months) for various substances. Mounting the samplers in the MSW allowed a flow-proportional part of the drainage to be sampled. Laboratory testing yielded high linear correlation between the accumulated sampler flow, q total, and accumulated drainage flow, Q total (r (2) > 0.96). The slope of these correlations was used to calculate the total drainage discharge from the sampled volume, and therefore contaminant load. A calibration of the MSW under controlled laboratory condition was needed before interpretation of the monitoring results was possible. The MSW does not require a shed, electricity, or maintenance. This enables large-scale monitoring of contaminant loads via tile drains, which can improve contaminant transport models and yield valuable information for the selection and evaluation of mitigation options to improve water quality. Results from this type of monitoring can provide data for the evaluation and optimisation of best management practices in agriculture in order to produce the highest yield without water quality and recipient surface waters being compromised.

Improving load estimates for NO3 and P in surface waters by characterizing the concentration response to rainfall events.

For the evaluation of action programs to reduce surface water pollution, water authorities invest heavily in water quality monitoring. However, sampling frequencies are generally insufficient to capture the dynamical behavior of solute concentrations. For this study, we used on-site equipment that performed semicontinuous (15 min interval) NO(3) and P concentration measurements from June 2007 to July 2008. We recorded the concentration responses to rainfall events with a wide range in antecedent conditions and rainfall durations and intensities. Through sequential linear multiple regression analysis, we successfully related the NO(3) and P event responses to high-frequency records of precipitation, discharge, and groundwater levels. We applied the regression models to reconstruct concentration patterns between low-frequency water quality measurements. This new approach significantly improved load estimates from a 20% to a 1% bias for NO(3) and from a 63% to a 5% bias for P. These results demonstrate the value of commonly available precipitation, discharge, and groundwater level data for the interpretation of water quality measurements. Improving load estimates from low-frequency concentration data just requires a period of high-frequency concentration measurements and a conceptual, statistical, or physical model for relating the rainfall event response of solute concentrations to quantitative hydrological changes.

Application and evaluation of a new passive sampler for measuring average solute concentrations in a catchment scale water quality monitoring study.

We present a field based testing, optimization, and evaluation study of the SorbiCell sampler (SC-sampler); a new passive sampling technique that measures average concentrations over longer periods of time (days to months) for various substances. We tested the SC-sampler within a catchment-scale monitoring study of NO(3) and P concentrations in surface water and tile drain effluent. Based on our field experiences, we optimized the flow velocity control and the sample volume capacity of the SC-samplers. The SC-samplers were capable of reproducing the NO(3) concentration levels and the seasonal patterns that were observed with weekly conventional grab sampling and continuous water quality measurements. Furthermore, we demonstrated that average measurements produce more consistent load estimates than "snapshot" concentrations from grab sampling. Therefore, when the purpose of a monitoring program is to estimate reliable (trends in) average concentrations or loads, the SC-samplers are a cost-effective alternative for grab sampling.