| Abstract (english) | One of the most important consequences of climate change, which threatens both surface and groundwater resources in coastal areas, is sea level rise. Although global sea levels have been rising throughout the 20th century, these processes have intensified towards the end of the century. According to data from the 2013 IPCC report, the mean value of global sea level rise was 1.7 mm per year in the period from 1901 to 2010, while it rose to 3.2 mm per year in the period from 1993 to 2010. Despite the global nature of the problem of sea level rise, the Mediterranean region, including the Adriatic, is one of the hotspots and most vulnerable areas. Sea level rise can intensify the natural processes of seawater intrusion (SWI) into coastal aquifers and surface waters, which can be exacerbated by anthropogenic activities such as the regulation of watercourses, excessive groundwater abstraction, etc. These processes particularly endanger lowland coastal areas such as estuaries and deltaic plains, which play a crucial socio-economic role in addition to their wealth of natural resources and biodiversity. SWI and changes in salinity within river deltas are the result of interactions between morphology and topography, tidal regimes and the inflow of freshwater from the catchment area. In the long term, SWI can have serious consequences in terms of degrading the quality of surface and groundwater resources. Consequently, the salinization of water resources can lead to soil salinization. Worldwide, more than 900 Mha of soil are classified as salt affected. In the short term, soil salinization can lead to reduced crop yields, while the long-term consequences are more severe, including a reduction in soil fertility and productivity, which could ultimately lead to permanent loss through desertification. In order to control and evaluate changes in vulnerable areas such as river deltas, environmental monitoring systems should be introduced, particularly to monitor soil and water quality. Many countries have developed and implemented water quality monitoring systems, most of which are regulated by law. Although technological advances have led to the development of means for automated and continuous water quality monitoring, most water quality monitoring systems are still carried out in the traditional way, usually with monthly sampling and laboratory analyses. While sampling-based monitoring is useful for the general characterization of water quality and the detection of long-term trends and seasonal variations, high-frequency in-situ sensor monitoring enables quantification of extreme events, short-term trends and sub-day variations in water quality parameters. Although weekly or monthly sampling may be sufficient for some parameters, such as pH, continuous data for parameters such as nutrients and ECw provide better opportunities for understanding hydrochemical processes in various water bodies. Continuous high-frequency sensor monitoring enables the collection of large amounts of data (big data) that can be used for advanced statistical modeling and the development of time series and machine learning models for long and short-term predictions of dynamic parameters such as ECw. Based on the above, two research hypotheses were tested: (I) surface and groundwater salinity indicators can be measured as accurately with in-situ sensors as with traditional monitoring methods; (II) surface and groundwater salinity will change over time due to natural and anthropogenic influences. The objectives of the research were (I) to compare salinity indicators measured with in-situ sensors and traditional monitoring methods; (II) to assess the impact of anthropogenic and natural factors on salinization of surface and groundwater over time using data obtained from in-situ measurements.
The research was carried out in the Neretva River delta on the east coast of the Adriatic Sea in Croatia. Within the delta, locations Vidrice (42°59'13'' S, 17°31'39'' I) and Luke (43°1'37'' S, 17°33'39'' I) were selected, which are characterized by a different spatial position and distance from the main watercourses, different soil properties and different land use and agricultural management practices. At both locations multiparameter probes measuring water temperature, depth, pH, ECw and ORP were installed in drainage canals for monitoring surface water and shallow piezometers (4 m deep) for monitoring groundwater quality. In the immediate proximity of the piezometers, FDR soil sensors were installed at both locations to measure temperature, moisture and ECb. In addition, sensors for soil water potential were installed at depths of 25 cm and 50 cm. Meteorological data was recorded using the automatic station installed at the Vidrice. Data loggers and modems were used to record and transmit real-time data in high temporal resolution. The collected in-situ data on water salinity were evaluated against the data from traditional monthly monitoring at the same locations. Descriptive statistics and ANOVA were performed for both long-term monthly monitoring (2010-2022) and continuous in-situ monitoring (2021-2022). High temporal frequency data collected through the established in-situ monitoring, in addition to water level data at the main watercourses, were used for the development of time series (ARIMA) and two machine learning models (MLR and XGB). The developed models were used to predict ECw seven, 14 and 30 days ahead and the predictions were evaluated using MAE and RMSE.
The results of the long-term traditional monthly monitoring showed that the average ECw of surface water at location Vidrice was 2.4 dS m-1, while a higher average ECw value (8.9 dS m-1) was measured in the groundwater, both with high variability. At location Luke, a higher salinity was found in the surface water (8.3 dS m-1) than in the groundwater (2.9 dS m-1). Analysis of the major ions showed that in both water bodies and at both locations the dominant cation was Na+ and the dominant anion was Cl-, with the exception of the groundwater at location Luke location the dominant anion was SO42-. The correlation analysis showed a positive, strongly significant correlation between ECw and Na+ and Cl- at both locations and for both water bodies. The ANOVA and the Tukey HSD post-hoc test showed that ECw differed significantly between the same water bodies at different sites as well as between different water bodies at each site. When analyzing the results of continuous in-situ water monitoring, no differences were determined between hourly and daily temporal frequency for any of the analyzed parameters at both locations in surface and groundwater. As with the traditional methods, higher average values of ECw were found in groundwater (12 dS m-1) compared to surface water (2.1 dS m-1) in the continuous in-situ monitoring at location Vidrice. In-situ continuous data showed clear differences in dynamics of ECw between surface water and groundwater. Intense precipitation events during non-growing periods resulted in highly dynamic changes in ECw in surface water, with hourly values ranging from 0.17 to 11 dS m-1. At the same time, each precipitation event in the groundwater led to a rapid decrease in ECw values. The less dynamic and slower changes in groundwater salinity compared to surface water are the result of the interaction of several parameters, such as pedological characteristics, aquifer recharge, distance to major watercourses and possible anthropogenic influences, such as the operation of pumping stations. At location Luke, a higher average ECw value was determined in the surface water (7.0 dS m-1) compared to the groundwater (3.7 dS m-1). The hourly time series of the ECw value in the surface water, where more dynamic changes were observed, show that precipitation had no direct influence on the ECw dynamics. The sudden and pronounced changes in the hourly data indicate that the salinity in the surface waters is predominantly subject to anthropogenic influences, namely the pumping regime. Similar to location Vidrice, the changes in ECw in the groundwater were less dynamic and slower. The ANOVA
revealed significant differences in ECw between surface water and groundwater at both locations and between the two study locations. The results indicate that the salinization of surface and groundwater at the selected locations are influenced by various natural and anthropogenic factors. The salinization of surface and groundwater at location Vidrice is mainly the result of the direct SWI into the coastal aquifer through the karstic coast in the southwest in combination with the permanent regulation of the water level (drainage canals, sluices, pumping stations, etc.). The changes in salinity of surface and groundwater at location Luke are primarily influenced by the stratified flow of the Neretva River and the SWI through the main riverbed. Evaluation of the in-situ measurements against the results of traditional monitoring showed a high correlation (0.97) and R2 value (0.94) for the surface water at location Vidrice with very low MAE (0.13 dS m-1) and RMSE (0.18 dS m-1) values, indicating a high degree of accuracy. At the same location, more significant deviations were observed for groundwater, with an R2 value of 0.28, together with errors of 2 dS m-1 for MAE and 3.16 dS m-1 for RMSE. The low accuracy and more significant differences in groundwater could be due to the different sampling depths in non-growing period when heavy rainfall occurred, leading to rapid changes in the water table. At location Luke, high correlation and R2 values were found between the two monitoring approaches for both surface water and groundwater. Lower error values were observed in the groundwater (MAE=0.32 dS m-1 and RMSE=0.39 dS m-1) compared to surface water (MAE=0.48 dS m-1 and RMSE=0.71 dS m-1). The changes in soil moisture and ECb up to 75 cm were influenced by precipitation and the associated rise in groundwater water table during the non-growing period as well as by precipitation and irrigation during the growing season at location Vidrice. The almost constant moisture in the deepest layer was mainly influenced by groundwater, which was within 1.5 m below the soil surface during most of the study period. A similar pattern was observed for ECb, where precipitation and the rise in groundwater water table during the non-growing period influenced the changes in ECb throughout the profile. The highest values were found in the deepest layer, which was also influenced by saline groundwater. During the growing season, when the water table was below 1.5 m and the orchard was irrigated, changes in ECb were observed in the top 75 cm of the soil profile. At location Luke, a higher average soil moisture was found in the upper half of the soil profile (0-50 cm), with the highest average value found in the 25-50 cm layer. The analysis of the changes in the dynamics of soil moisture and ECb showed that during the growing season in 2021 and 2022, irrigation measures had the greatest influence on the changes up to a depth of 75 cm. During the non-growing period, the changes in soil moisture and ECb were influenced by precipitation and the associated rise in the groundwater table. The data collected at hourly and daily temporal resolution was used to develop machine learning and time series for predicting ECw. The XGB model performed better than the MLR model on hourly input data. Satisfactory results with low MAE and RMSE were obtained for the prediction of seven days ahead in surface water at location Vidrice and up to 30 days ahead in surface and groundwater at location Luke as well as in groundwater at location Vidrice. Using daily input data, the ARIMA model showed the best performance, with the lowest errors observed for the seven-day-ahead prediction. The developed and tested models can be used for reliable short-term prediction of ECw in surface and groundwater at selected locations in the hydro-meliorated river delta. The results of this research can guide the planning of future water and soil management practices in vulnerable agro-ecosystems such as river deltas that are under pressure from climate change, especially sea level rise, and increased SWI. |
