Machine learning improves groundwater recharge estimates in WA

A new study led by Griffith University has unveiled a machine learning-based framework to accurately estimate groundwater recharge in the Perth Basin, with a particular focus on the Gnangara groundwater system.

Located in southern Western Australia, the Gnangara aquifer system is one of the region’s most critical water resources but also among the most vulnerable to climate change.

The Perth Basin served as the training site for the downscaling operation, while the Gnangara system served as the calibration site to estimate recharge.

The Gnangara system lies within the Perth Basin, which provides 35%-50% of Perth’s drinking water and supports key industries such as agriculture and mining.

Groundwater recharge is the process of water from the surface (e.g. from rain, rivers or lakes) replenishing groundwater reserves.

Declining precipitation and recharge, coupled with projections of continued declines, have posed significant challenges to managing this essential resource.

The study, led by Griffith’s Australian Rivers Institute PhD Candidate, Ikechukwu Kalu using the Gravity Recovery and Climate Experiment (GRACE) satellite data, combined advanced random forest regression models with groundwater storage anomalies to overcome the current limitations in spatial resolution.

By downscaling GRACE data to a fine resolution of 0.05° (~5 km), researchers achieved reliable recharge estimates over the Gnangara system, which is a relatively small calibration site of approximately 2,200 km².

“This improved GRACE data allows us to track groundwater changes in the Gnangara system and estimate recharge more accurately,” said Dr Christopher Ndehedehe, an ARC DECRA Fellow at Griffith University and co-author of the study.

Understanding groundwater dynamics within the Perth Basin’s complex aquifer system – which comprises the Superficial, Leederville, and Yarragadee aquifers – had been a long-time challenge.

The study revealed the response of these aquifers to rainfall varied significantly. The Superficial aquifer, which directly recharged from rainfall, showed a response of 44%, while the confined Leederville and Yarragadee aquifers, which recharged in specific areas, responded at 23% and 14%, respectively.

Lead author Mr Kalu emphasised the importance of integrating remote sensing with ground-based bore monitoring for a holistic understanding of groundwater dynamics.

“These findings underscore the potential of leveraging emerging remote sensing technologies for groundwater monitoring, management, and policymaking, offering tools to safeguard the Perth Basin’s groundwater systems in the face of escalating climate pressures.”
PhD Candidate Ikechukwu Kalu

“Understanding groundwater recharge at local scales is crucial, as many productive aquifers are relatively small and lack sufficient in-situ monitoring.

“We must invest in more efficient, low-cost, and accurate technologies to manage our groundwater systems, ensuring the sustainability of this vital resource for the benefit of our ecosystems and communities.”

The study ‘Remote Sensing Estimation of Shallow and Deep Aquifer Response to Precipitation-Based Recharge Through Downscaling’ has been published in Water Resources Research.