Integrating remotely sensed data for sustainable groundwater resource management across the Steenkoppies Catchment

Abstract

Effective monitoring networks are central to achieving sustainable groundwater resource management. Management of water demand against supply, and the prevention of long-term mismanagement impacts can only be successful when the spatio-temporal resolution of monitoring data can inform viable management interventions. Irrigated agriculture is the primary global consumer of groundwater, so ensuring sustainable use is a major challenge in many regions worldwide. Given the high costs of development, equipment, and maintenance for groundwater monitoring networks, along with spatial and temporal resolution limitations, remotely sensed data offer an attractive and innovative complementary means for monitoring groundwater. This thesis investigated the application of remote sensing technology to monitor groundwater resources and irrigation use, specifically focusing on quaternary catchment A21F, known as the Steenkoppies Catchment, in South Africa (RSA). The Steenkoppies Catchment includes the Steenkoppies Dolomitic Compartment (SDC) and the intergranular and/or fractured aquifers to its north primarily drained by the Magalies River. Significant debate surrounds whether the increase in depth to groundwater levels across the SDC, along with the reduced discharge from Maloney's Eye, are due to excessive groundwater extraction versus decreasing precipitation (Holland 2009). Since 1995, downstream users of the Magalies River have frequently requested actions to limit or cease groundwater abstraction from the SDC when Maloney’s Eye flow decreased. The strategies formulated, however, often go unrealised following periods of average and above-average precipitation. Irrespective of management and mitigation strategies, however, effective implementation and compliance are unachievable without productive and integrated monitoring. In 2008 for example, flow at the Maloney’s Eye gauging station was measured at 5.49 Mm3 yr-1, significantly below the 100-year average of 14.4 Mm3 yr-1 (Meyer 2014). This study used the Steenkoppies Catchment, an area heavily reliant on groundwater for irrigation, to evaluate how remotely sensed data can enhance holistic monitoring of groundwater aquifer supply and demand. Key objectives focused on addressing the low spatial resolution limitations of remotely sensed groundwater storage anomaly (GWSA) data to enable higher spatio-temporal monitoring of aquifer supply status, groundwater abstraction, and recharge rates. For groundwater demand monitoring, the objectives included adapting remote sensing data to provide near real-time estimates of cultivated areas, crop water use, and irrigation demand, allowing for direct comparison with aquifer status data. Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) data to in situ automatic weather stations (AWS) comparisons were prioritised because the product was essential for the downscaling of remotely sensed groundwater storage (GWS) estimates, estimating irrigation water use and calculating high-resolution drought indices. Although inherent errors are expected due to the limited spatial coverage of individual stations relative to the larger area each pixel represents, the average monthly root mean square error (RMSE) and mean absolute error (MAE) values across 33 stations were 41 mm and 25 mm, respectively. The findings confirmed that CHIRPS data reliably represent local and regional precipitation patterns across the study area. Groundwater storage anomalies from the monthly, Global Land Data Assimilation System (GLDAS) Version 2.2 (0.25°) product, that assimilates terrestrial water storage (TWS) observations from the Gravity Recovery and Climate Experiment (GRACE) missions to enhance GWS simulations, were downscaled to the higher resolution of 0.05° using a random forest (RF) machine learning (ML) algorithm. A representative and accurate comparison between the downscaled GWSA estimates and the in situ observation derived GWSA were achieved using a novel approach by adjusting the specific yield (Sy) values in increments of 0.01 to determine that Sy values of 0.02 and 0.04 yield the lowest RMSE and MAE for the intergranular and/or fractured aquifer, and karst aquifer, respectively. Although the Sy values fall within the acceptable range of literature values, the values were considered relatively high. The downscaled GWSA analysis with in situ data resulted in a coefficient of correlation (r) of 0.3 to 0.6, RMSE of 40 to 50 mm, and MAE of 31 to 41 mm. The greater range of downscaled GWSA values suggest that the high-resolution output successfully integrated precipitation and ETa inputs, offering a more detailed and accurate reflection of groundwater dynamics. A streamlined approach was proposed to enhance the temporal resolution of land use land cover (LULC) datasets, enabling high-resolution monitoring of crop and irrigation water use in cultivation areas through remote sensing, without requiring ground-truth data. Leveraging the distinct spectral signature of cultivation, seasonal Sentinel-2 composites were effectively applied to train a RF ML model to accurately identify cultivation from monthly composites. Monthly rainfed and irrigated cultivated areas across the SDC were then distinguished using a published Department of Agriculture, Land Reform and Rural Development (DALRRD) annual product. The cultivation analysis concluded that the annual irrigated cropping area was 6 106 ha in 2019/20 (December 2019 to November 2020) and 6 065 ha in 2020/21. This indicated an increase of around 700 ha (14%) in irrigated area since 2012. It was further demonstrated how combining the irrigated area data with remotely sensed precipitation and crop water use estimates could inform whether increased irrigation water use (groundwater abstraction) is driven by expanded cultivation or weather patterns. The generated product can be applied to generate cropping and irrigation intensity products to support more strategic and effective allocation of water regulatory efforts. The near real-time Earth observation data could enable monthly cultivation monitoring to track compliance for measures like planting only outside dry seasons to reduce abstraction or reducing planting frequencies. The strong correlation between surface water flux and GWSA during low to no precipitation periods indicate that net GWS changes are a reliable indicator for estimating both groundwater recharge and abstraction. The seasonal pattern indicates that groundwater abstraction peaks in dry seasons due to low natural recharge, while wet seasons focus on aquifer recharge, reducing reliance on groundwater for irrigation. Additionally, it was noted that the downscaled GWS estimates consistently exceed surface water flux estimates. In situ derived GWSA are calculated by multiplying groundwater level anomalies (GWLA) with the typical or average Sy value considered representative of the aquifers. The feasibility of creating a high spatial-resolution, monthly GWLA product through algebraic rearrangement to solve for GWLA was investigated, and to what extent the MAE values, compared to those derived from the downscaled to in situ derived GWSA comparison, could be reduced if these monthly GWLA were multiplied by new Sy values. Using a new Sy value of 0.007 for the hard rock aquifer, decreased the MAE from 31 mm (when using a Sy of 0.02) to 19 mm when compared to the in situ derived GWSA. Using a Sy value of 0.02 across the karst aquifer, decreased the average MAE from 39 mm (when using a Sy of 0.04) to 20 mm when comparing the in situ derived and downscaled, Sy adjusted GWSA. Groundwater abstraction for irrigation in the SDC is generally estimated at 20 to 25 Mm³. Using a Sy of 0.03, the downscaled GWSA product produced an annual average abstraction estimate of 26.3 Mm³, closely matching this range. This Sy also yielded recharge estimates of 11 and 17% for the SDC, while a lower Sy of 0.007 for the MRA generated recharge rates of 6 and 7%, aligning with published aquifer recharge estimates. With this optimal Sy, the Sy adjusted GWSA product now better represents local hydrogeological characteristics, allowing for more accurate, monthly, and real-time net GWS increase and decrease monitoring. This thesis illustrated how remotely sensed data can be valuable for effective groundwater resource monitoring, without the need for extensive skills and expensive in situ data. Remotely sensed data can revolutionise groundwater resource status and demand monitoring, specifically for intensively irrigated areas reliant on groundwater. The near real-time, high spatio-temporal resolution monitoring data can inform and facilitate proactive resource mitigation strategies to the benefit of local farmers and regulatory authorities.