Deep Learning for Isotope Hydrology: The Application of Long Short-Term Memory to Estimate High Temporal Resolution of the Stable Isotope Concentrations in Stream and Groundwater

Recent advances in laser spectroscopy has made it feasible to measure stable isotopes of water in high temporal resolution (i.e., sub-daily). High-resolution data allow the identification of fine-scale, short-term transport and mixing processes that are not detectable at coarser resolutions. Despite such advantages, operational routine and long-term sampling of stream and groundwater sources in high temporal resolution is still far from being common. Methods that can be used to interpolate infrequently measured data at multiple sampling sites would be an important step forward. This study investigates the application of a Long Short-Term Memory (LSTM) deep learning model to predict complex and non-linear high-resolution (3 h) isotope concentrations of multiple stream and groundwater sources under different landuse and hillslope positions in the Schwingbach Environmental Observatory (SEO), Germany. The main objective of this study is to explore the prediction performance of an LSTM that is trained on multiple sites, with a set of explanatory data that are more straightforward and less expensive to measure compared to the stable isotopes of water. The explanatory data consist of meteorological data, catchment wetness conditions, and natural tracers (i.e., water temperature, pH and electrical conductivity). We analyse the model's sensitivity to different input data and sequence lengths. To ensure an efficient model performance, a Bayesian optimization approach is employed to optimize the hyperparameters of the LSTM. Our main finding is that the LSTM allows for predicting stable isotopes of stream and groundwater by using only short-term sequence (6 h) of measured water temperature, pH and electrical conductivity. The best performing LSTM achieved, on average of all sampling sites, an RMSE of 0.7‰, MAE of 0.4‰, R2 of 0.9 and NSE of 0.7. The LSTM can be utilized to predict and interpolate the continuous isotope concentration time series either for data gap filling or in case where no continuous data acquisition is feasible. This is very valuable in practice because measurements of these tracers are still much cheaper than stable isotopes of water and can be continuously conducted with relatively minor maintenance.

Application of Machine Learning Models to Predict Maximum Event Water Fractions in Streamflow

Estimating the maximum event water fraction, at which the event water contribution to streamflow reaches its peak value during a precipitation event, gives insight into runoff generation mechanisms and hydrological response characteristics of a catchment. Stable isotopes of water are ideal tracers for accurate estimation of maximum event water fractions using isotopic hydrograph separation techniques. However, sampling and measuring of stable isotopes of water is laborious, cost intensive, and often not conceivable under difficult spatiotemporal conditions. Therefore, there is a need for a proper predictive model to predict maximum event water fractions even at times when no direct sampling and measurements of stable isotopes of water are available. The behavior of maximum event water fraction at the event scale is highly dynamic and its relationships with the catchment drivers are complex and non-linear. In last two decades, machine learning algorithms have become increasingly popular in the various branches of hydrology due to their ability to represent complex and non-linear systems without any a priori assumption about the structure of the data and knowledge about the underlying physical processes. Despite advantages of machine learning, its potential in the field of isotope hydrology has rarely been investigated. Present study investigates the applicability of Artificial Neural Network (ANN) and Support Vector Machine (SVM) algorithms to predict maximum event water fractions in streamflow using precipitation, soil moisture, and air temperature as a set of explanatory input features that are more straightforward and less expensive to measure compared to stable isotopes of water, in the Schwingbach Environmental Observatory (SEO), Germany. The influence of hyperparameter configurations on the model performance and the comparison of prediction performance between optimized ANN and optimized SVM are further investigated in this study. The performances of the models are evaluated using mean absolute error (MAE), root mean squared error (RMSE), coefficient of determination (R2), and Nash-Sutcliffe Efficiency (NSE). For the ANN, the results showed that an appropriate number of hidden nodes and a proper activation function enhanced the model performance, whereas changes of the learning rate did not have a major impact on the model performance. For the SVM, Polynomial kernel achieved the best performance, whereas Linear yielded the weakest performance among the kernel functions. The result showed that maximum event water fraction could be successfully predicted using only precipitation, soil moisture, and air temperature. The optimized ANN showed a satisfactory prediction performance with MAE of 10.27%, RMSE of 12.91%, R2 of 0.70, and NSE of 0.63. The optimized SVM was superior to that of ANN with MAE of 7.89%, RMSE of 9.43%, R2 of 0.83, and NSE of 0.78. SVM could better capture the dynamics of maximum event water fractions across the events and the predictions were generally closer to the corresponding observed values. ANN tended to underestimate the events with high maximum event water fractions and to overestimate the events with low maximum event water fractions. Machine learning can prove to be a promising approach to predict variables that are not always possible to be estimated due to the lack of routine measurements.

Isotopic tracers of sources of water for springs from the Edwards Aquifer, Central Cexas, USA

The Edwards Aquifer (EA) in Central Texas provides water supply for over two million people and contains springs that are hydrologically and ecologically important to the region. The residence time of groundwater in the EA ranges from a few days to many thousands of years, since water in the aquifer is contained and transported within both matrix porosity and large conduits. In this study, stable isotopes of water from five springs are investigated for tracing the origin of water and hydrological processes in the EA system during 2017–2019. There is a quick response of the isotopic signals measured at these springs to changes in the isotopic compositions of precipitation. By utilizing an isotope mixing model, we have identified sources of water for these springs with a bi-modal distribution of groundwater supply in the EA: water supplied from deep groundwater with a longer residence time (an average of 67%) and supplemental epikarst interflow with a shorter residence time (an average of 33%). The evolution of hydrochemical water types from HCO3–Ca to HCO3·Cl–Ca·Mg along the EA flowpaths indicates that inputs from epikarst interflow are greater in springs within the artesian zone than the springs within the contributing zone.

Assessing Surface water- alluvial aquifer water exchange using a multitracer approach and modelling

<p>Alluvial aquifers are generally highly productive in terms of groundwater and are therefore particularly exploited. The study site is a drinking water production facility located on the alluvial plain of the Rhône river, France. This site consists of several pumping wells and observation piezometers organized along the riverbank. The site is continuously supplying water to neighboring agglomerations with intermittent pumping. In this situation, the pumping produces a piezometric depression allowing leading to a water exchange from the river to the aquifer which is a common feature in the case of alluvial aquifer exploitation along a riverside.</p><p>The four pumping wells and five piezometers were equipped with continuous automatic temperature and water level measurement probes, the river stage is monitored as well. These data are used to determine the exchange (direction and magnitude) between the aquifer and the river. Although pumping is intermittent, it does not allow a sufficient recovering of the natural piezometric level, i.e. the aquifer is permanently below the river stage.</p><p>In addition to the automatic probes, additional data acquisition campaigns were carried out. During these campaigns different tracers were used such as conductivity, stable isotopes of water and radon activity. Together with the continuously measured temperature, these various tracers were used to identify hydrodynamic variables and parameters, such as Darcy’s velocity, dispersivity, transit times. A MODFLOW model was developed, integrating the site geometry and hydrodynamic context, with the Rhone River at the western boundary and the Ouveze river at the eastern boundary. Model calibration was performed using the study site piezometric records and the optimization package PEST. The flow was reproduced at the site for two situations, a natural situation without groundwater pumping, and the exploitation situation with the groundwater withdrawals. Finally, the tracer’s data were integrated into the model to reproduce the transport of different tracers, in order to quantify the exchanges and the water fractions coming from the different hydraulic boundaries.</p>

Study of water salinity in the Chtouka plain (Morocco) using TEM geophysical method, chemical and isotopes tracers

<p>The Chtouka plain in Morocco suffers from groundwater overexploitation and a significant increase in water salinity. In this study, a multidisciplinary approach combining water chemistry, stable isotopes of water (18O, 2H) and Transient Electromagnetic (TEM) method was used. The main objective was to identify the water salinity sources and the extension of the marine intrusion. Water samples were collected from wells and boreholes, springs, the Massa river and the main source of freshwater in the region, the Youssef Ibn Tachfine Dam. Geophysical (TEM) measurements (12 profiles comprising 83 measurement points) were carried out along the coastal zone and around the northern bank of the Massa river. The results show a spatial variability of water salinity, indicating rock-water interaction, seawater intrusion and anthropogenic influence. The interpretation of the TEM soundings allow to draw the front line of the marine intrusion in the aquifer. The results, compared to previous numerical simulations, show a significant progress of the marine intrusion into the coastal aquifer. The intrusion indeed reaches a distance of 2.5 km from the coast, far beyond models’ predictions. The local water authorities can use these results to improve their monitoring network and better assess the progress of the seawater intrusion.<br>Keywords: Water salinity, TEM geophysical method, chemical and isotopes tracers, marine intrusion</p>

Assessing the residence time of water in volcanic lakes of north Cameroon using bomb-36Cl and stable isotopes

<p>The Cameroon Volcanic Line (CVL) in Central Africa hosts numerous volcanic lakes. While Nyos and Monoun lakes in western Cameroon were well studied following the catastrophic release of CO<sub>2</sub> that occurred in 1980s, other volcanic lakes such as those of the Adamaoua Plateau remain less documented. Although some of these (Mbalang and Tizon) have been investigated through their sedimentary archives in order to reconstruct past-environments, the functioning of these hydro-systems located in the northern part of the CVL is not well constrained. Here, we characterize the hydrological functioning of five volcanic lakes by coupling classical hydrology methods and isotope tracers. Specifically, we assess water residence time in these lakes using radioactive (<sup>36</sup>Cl) and stable isotopes of water.</p><p><sup>36</sup>Cl is a cosmogenic isotope of chlorine produced naturally in the stratosphere by spallation of <sup>40</sup>Ar induced by cosmic-rays and has been massively injected into the atmosphere by nuclear tests during the 1950s. This pulse of bomb-<sup>36</sup>Cl can thus be used as a tracer to estimate recharge rates in the unsaturated zone and to constrain water transit times at a regional scale. While water stable isotopes have been widely used to establish lakes hydrological balance in Sahelian regions, only a few studies have been reported to date using <sup>36</sup>Cl for the same purpose in tropical areas.</p><p>In this study, together with major elements and stable isotopes, we analyzed <sup>36</sup>Cl contents in water from lakes Mbalang, Tabere, Tizon, Gegouba and Baledjam around Ngaoundere, to assess residence time in these lacustrine systems. <sup>36</sup>Cl/Cl ratios range from 1400.10<sup>-15</sup> to 2800.10<sup>-15</sup> at/at and are significantly higher than the natural baseline as assessed by data obtained in local groundwater or at a larger scale in the Lake Chad Basin (<sup>36</sup>Cl/Cl ~200.10<sup>-15</sup> at/at, see Bouchez et al., Scientific Reports, 2019). These <sup>36</sup>Cl/Cl ratios above the natural baseline are clearly tagged with the bomb-<sup>36</sup>Cl footprint. We will illustrate at the meeting how a simple transient-state one-box model can be used to explain why these lakes have different <sup>36</sup>Cl/Cl ratios, and how these results can help to constrain the E/I ratios of the lakes, and be compared with their hydrological characteristics and stable isotopes signatures.</p>

Assessment of flow paths and groundwater storage processes in an island fractured rock aquifer system using the stable isotopes of water (H2O) and strontium (Sr), and environmental tracers’ tritium (3H) and sulfur hexafluoride (SF6).

<p><strong> </strong></p><p>Access to fresh water is a critical humanitarian issue in many regions of the world and on the most important sustainable development goals. This crisis is exacerbated by the effects of climate change, pollutants, increases in demand and overuse. Fractured rock aquifers have been providing potable groundwater for many regions of the world, but there still many unknowns about the storage capacity, transit times and flow paths under changing climate scenarios. In this on-going study we aim to understand the magnitude of groundwater storage and Inter-basin flow for water supply development and sustainable use on the island of Tobago, WI. Samples of springs, surface water and production wells were analyzed for the stable isotopes of water (H<sub>2</sub>O) and strontium (Sr), tritium(<sup>3</sup>H) and sulfur hexafluoride (SF<sub>6</sub>). The stable isotopes of water δD<sub>vsmow</sub> and δ<sup>18</sup>O<sub>vsmow</sub> indicates that the groundwater in the northern region of the island is not hydrologically connected to the groundwater to the south because of their distinct D-excess signature.</p><p><sup>87</sup>Sr/<sup>86</sup>Sr and Na-normalized strontium concentrations produces five possible mixing lines which radially increase from the lowest based value of <sup>87</sup>Sr/<sup>86</sup>Sr ~ 0.70396 and Na/Sr (mg/L) ~0.00652. In detail, the maximum values of each line represent: extremes in Na concentration <sup>87</sup>Sr/<sup>86</sup>Sr 0.70576 and Sr/Na 0.0008 mg/L, the groundwater to seawater mixing line<sup>87</sup>Sr/<sup>86</sup>Sr 0.70506 and Sr/Na 0.0023 mg/L, the precipitation to rock equilibration mixing line <sup>87</sup>Sr/<sup>86</sup>Sr 0.70506 and Sr/Na 0.0023 mg/L, water located in silicate rocks to carbonate rocks mixing <sup>87</sup>Sr/<sup>86</sup>Sr 0.70871 and Sr/Na 0.0085 mg/L, and wells that were once affected by seawater intrusion <sup>87</sup>Sr/<sup>86</sup>Sr 0.70563 and Sr/Na 0.0692 mg/L.</p><p>Tritium results range from 0.02 to 0.60 TU and calculations suggest that wells contain a range of 2 to 44% modern groundwater. When compared to other islands at similar latitudinal locations, Tobago’s groundwater presents the lowest mean and median tritium values even though it is the closest to the equator. Basin flux and effective porosity were calculated for the 10 wells using the apparent age obtained from these results under the assumption of piston flow. It was found that all watershed volumes were magnitudes of orders larger than the sub-basins where the wells were located ranging between 0.09 km<sup>3 </sup>to 8.23 km<sup>3</sup>. Basin Flux and effective porosity also contain large range differences 1.40 *10<sup>5</sup> m<sup>3</sup>/yr to 9.93 *10<sup>6 </sup>m<sup>3</sup>/yr, and 0.014 to 0.094, respectively.</p><p>This results also suggest that the groundwater in the southern regions of the island contains the oldest water with one well sample >60 years. SF<sub>6 </sub>results reflect similar ages except for 3 wells samples which are suspected to be contaminated by excess air. This novel discovery illustrates that small, fractured rock island aquifers can possess structural complexities that lead to older groundwater ages and variances in basin characteristics.</p><p> </p>

Water accessed differently by Larix decidua according to an alpine gradient

<p>Vegetation is the primary connection between land and atmosphere, thus the main player mediating the consequences of a changing climate on land cover and hydrology. A protected alpine catchment, with a larch grove (<em>Larix decidua</em>) at the upper limit of the forested area, the Vallon de Nant (Vaud alps, 1200 – 3050 m. a.s.l.), was chosen as a study site in parallel with ongoing hydrological observation. We analyzed the stable isotopes of water, δ<sup>18</sup>O and δ<sup>2</sup>H, in the xylem extracted from samples of 10 trees in 2 transects just above and below 1500 m. a.s.l. over the course of the 2017 and 2018 growing seasons.  We compared isotopic ratios withconcurrent observations of isotopes in precipitation, stream, soil water, and groundwater.  Isotopic content of xylem water was found to be fairly consistent, independent of the date within the season, and closely resembled shallow soil water, suggesting that recent precipitation dominated the water source.   Our results support that vegetation could experience a drought due to low levels of rainfall before the streamflow is impacted. Furthermore, they affirm current discussions that water source is elevation dependent for trees in mountain ecosystems, with summer precipitation being favored by higher elevation trees, such as these. This preference has significant implications when we project current changes of quantities of rain falling as snow versus rain in the future. And more importantly, changes the view of forest from that of a water “user” to that of a store and player in complex feedback mechanisms.</p>