IoT-based data and analytic hierarchy process to map groundwater recharge with stormwater

Most countries in Southern Africa largely rely on surface water. With prolonged droughts in many countries that are expected to reoccur frequently in the future, there is a major shift to alternative water sources such as groundwater. The main challenge with the utilisation of these alternative water sources is the high monetary cost of quantity and quality monitoring coupled with limited human and financial resources. The novelty in this study is in the application of Internet of Things (IoT) for measurement and analytic hierarchy process (AHP) as a mapping approach for sustainable management of groundwater resources in developing countries. First, the application of IoT for measurement and monitoring is relatively new, while the novelty is the application in the Southern Africa context in the face of major challenges such as availability of energy and Internet; access to IoT components, i.e., sensors and gateways; safety of individuals installing the devices in the field; and vandalism.

Groundwater is the most extensive freshwater resource that is also resilient to climate change and less susceptible to pollution compared to sources such as surface water. To develop a sustainable and resilient water supply system, countries such as South Africa that are located in semi-arid areas and largely depend on surface water need to diversify water sources to include groundwater to meet future demands (DWA 2010). The need to diversify water sources is more urgent in urban areas to meet the rapidly increasing demand attributed to population growth and improving living standards due to rising incomes. In the city of Cape Town (CCT) in South Africa, identifying opportunities for managed aquifer recharge is crucial to sustain groundwater systems (Mauck 2017). Although groundwater resources are extensive and their resilience to climate change is well known as climate change projections for Southern Africa indicate an expectation of hotter and drier weather, particularly in the western part of the region, they need to be managed sustainably and preserved to meet the water demands of the future (DWA 2010). Unfortunately, efforts to sustainably manage groundwater resources are often hindered by limited data and inadequate monitoring capacity to support decision-making (Gaffoor et al. 2020). High-quality data are essential to determine the variability and viability of the groundwater resource (Cobbing & Hiller 2019). IoT technology has provided the opportunity to remotely collect data in real time with high spatial and temporal resolution that can support decision-making and enhance the management of the groundwater resources (Gaffoor et al. 2020). In this study, an investigation was undertaken to assess the application of IoT-based data and the use of AHPs to map groundwater recharge potential around two stormwater ponds in a small, urbanised area in Cape Town. The study involved the collection of groundwater-level data using an IoT monitoring system and then using the Geographic Information System (GIS) and AHP to generate groundwater recharge potential zones in the study area. Using a groundwater model in MODFLOW, the study also estimated the recharge rates and processes in the aquifer.

The topography is relatively flat (less than 3% average slope) with an unconfined aquifer that offers opportunity for groundwater recharge (Adelana 2010; Mauck 2017). The study area is 0.315 km² comprising the catchment area draining into the stormwater ponds and surrounding areas.

The maximum daily ET_o value was obtained as 3.28 mm/day (3.8 × 10⁻⁸m/s). This ET_o value was preferred to the pan evapotranspiration value due to limited data caused by vandalism of the weather station. An average ET_o value of 2.59 mm/day (3.0 × 10⁻⁸ m/s) was used in the model.

A combination of GIS and AHP was used to delineate and map the groundwater recharge zone. Seven thematic maps were considered the control factors of groundwater flow and storage in the area. These influencing factors were weighted according to their reaction to groundwater occurrence, expert opinion, and consideration from the reviews from previous studies. The factors and their importance are compiled from previous literature, and appropriate scores were assigned to each factor to produce a final groundwater recharge potential map of the area (Yeh et al. 2016). Various studies have been carried out globally using the integration of geospatial data, multi-criteria decision analysis techniques, remote sensing, and GIS technology to identify suitable sites for groundwater recharge potential through MAR suitability maps (Sallwey et al. 2019; Achu et al. 2020; Ahmed et al. 2021). The assigned values were chosen in comparison to studies like Sener & Davraz (2013) and Arulbalaji et al. (2019). Expert opinion is valuable but introduces subjectivity; therefore, the methodology was subjected to peer review for transparency and credibility.

The process required slope information that was generated from the Shuttle Radar Topography Mission (SRTM) data, drainage density, and lineament density maps that were generated from the DEM created using Lidar dataset. Land use/land cover (LULC) was determined through supervised image classification and the geology, lithology, and soil shapefiles of the Cape Flats Aquifer used to extract the shapefile of the study area using the extraction by mask tool in ArcGIS software. From these data, seven thematic layers/factors, including geology, lineament density, lithology, slope, soil, LULC, and drainage density, were analysed, weighted according to their reaction to groundwater occurrence, and assigned weights from Saaty's scale of relative importance (Table 3).

The thematic layers were compared with each other using a pair-wise comparison matrix as shown in Table 4.

A normalised pair-wise comparison matrix was created and the normalised weights for each component were calculated using the average of each row as shown in Table 5.

A pair-wise matrix is considered consistent when is equal to or more than the number of the layers being examined. Given that of 7 and n of 7 were obtained, a CI of 0 was calculated. The RCI was obtained from Table 6 and the CI was calculated as 0. Therefore, this analysis was considered since Saaty (1990) stated that a CR of equal to or less than 0.1 for larger matrices is acceptable to continue the analysis.

The final weight and ranking for the different sub-classes of the thematic layers are given in Table 7.

The study area had an existing long-range wide area network (LoRaWAN) coverage; therefore, a SenseCAP LoRaWAN wireless sensor network, with a water-level sensor, node, gateway, and smartphone were used for monitoring and analysing groundwater levels. LoRa technology is distinguished by its extensive coverage, capable of transmitting data over long distances, provided there is a clear line of sight between the gateway and the sensor. LoRaWAN, as a network protocol, leverages LoRa technology for low-power wide area network communications and device management. In rural settings, research indicates that LoRa typically provides coverage of approximately 20 km, whereas in urban areas, LoRa can achieve ranges of up to 5 km (Sanchez-Iborra et al. 2018; Benites et al. 2019; Sendra et al. 2022).

The water-level data on The Things Network (TTN) were analysed and visualised using MATLAB and the ThingSpeak application. A channel was created on the ThingSpeak website to generate a right application programme interface (API) to connect the TTN and ThingSpeak and perform an in-depth analysis of the data, view/graph the necessary variables, send alerts, and export the data to Microsoft Excel. The IoT-based data were validated using the manually recorded water-level data at the MON71.

The dipped groundwater-level data recorded monthly at the MON71 borehole in August, September, and October 2022 were also obtained and compared with the IoT-based groundwater levels. The data showed an acceptable difference of less than 0.1 m. It was determined that IoT is reliable in collecting large amount of data while maintaining the desired accuracy levels, which solves the problem of insufficient hydrogeological data. It is possible to employ IoT-based systems to collect high-resolution data for use in developing and calibrating models for groundwater studies.

In the first 28 days, data were collected at 15-min intervals, which caused the battery voltage to drop below the required threshold of 3.5 mV, leading to a communication failure on 16 August 2022. A new battery was installed on 22 August 2022, and data collection continued at a 24-h resolution as the groundwater levels did not exhibit significant changes within a sub-daily time step. The battery ran out on 26 October 2022 and a replacement battery was installed on 15 November 2022, resulting in a noticeable increase in battery voltage and a corresponding jump in the groundwater data. It was determined that reducing the temporal resolution and frequency of data collection prolongs the battery life.

The drain package (DRN) was used to simulate the amount of stormwater runoff draining into the ponds to facilitate aquifer recharge. The drainage lines shapefile was assigned to the Model_Top with a conductance per unit length of 0.001. The steady state was run with a single time step, while the transient state was run for 20–140 days.

The recharge rates were adjusted according to the study by Xu & Beekman (2019), which classified the recharge rates in Cape Town in the range of 11–54 mm/year. The K_x values of the layers were adjusted within the allowable range of 10–50 m/day as specified by the CFA hydraulic conductivity profile. The evapotranspiration rates were calculated using the FAO Penman–Monteith equation, and the aquifer thickness was adjusted with reference to the CFA thickness. The steady and transient state models were calibrated by visually comparing simulated heads at MON71 and the IoT-based observed data. The root mean square error (RMSE) method was used to determine the accuracy of the model calibration.

The sensitivity analysis was carried out to determine the effect of varying different parameters and stresses on the model calibration process. The model showed that varying recharge rates, hydraulic conductivity, and aquifer thickness influences the fluctuations in the water table and the outflows presented in the water budget tables. An increase in the recharge rate registered an increase in the water table and outflow values; thus, groundwater levels are directly proportional to recharge. The results from the abstraction rate showed that groundwater levels are inversely proportional to the abstraction rate; as an increase in the abstraction rate decreased the water table, more water was taken from the aquifer. A reduction in the hydraulic conductivity in Layer 2 caused a decrease in the water table, while an increase in the hydraulic conductivity in Layer 1 led to an increase in the water table.

According to Dlodlo (2012), linking IoT to sustainable water resources management is very attractive and utilised in the region of water resources management. In this study, IoT-based groundwater-level data were collected using an IoT monitoring system that was easy to use and had the potential to accurately analyse and process large amounts of data used to calibrate the MODFLOW model. Therefore, IoT turned out to be a standout as a groundwater data collection system (Oke et al. 2019).

This study showed that IoT-based data can be leveraged to address challenges and limitations prevalent in groundwater resource management in Southern Africa. IoT data can be used to effectively address the lack of high-quality groundwater data, enabling accurate collection, analysis, and processing of groundwater levels. Battery drain was mitigated by reducing data collection frequency, ensuring prolonged sensor battery life. The study also utilised GIS-based analysis, AHP techniques, and MODFLOW software to identify and map the recharge potential of two stormwater ponds. The difference between the observed and simulated heads at the MON71 borehole was 0.86 m. The results from the modelling exercise confirmed the occurrence of recharge at the stormwater pond, validating the effectiveness of stormwater ponds for groundwater augmentation. It was also determined that comprehensive hydrogeological assessments including GIS and AHP techniques were essential for mapping recharge zones and creating groundwater augmentation suitability maps. Stormwater ponds were classified in the excellent groundwater recharge zone, demonstrating their potential for groundwater replenishment through infiltration. Fluctuations in the water table and outflows were influenced by recharge rates, hydraulic conductivity, and aquifer thickness. Higher recharge rates resulted in increased water table and outflow values, establishing a direct relationship between groundwater levels, recharge rates, and abstraction for effective groundwater resource management.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.