The Gnangara groundwater system is a highly productive water resource in southwestern Australia. However, it is considered one of the most vulnerable groundwater systems to climate change, due to consistent declines in precipitation and recharge, and regional climate models project further declines into the future. This study introduces a new framework underpinned by machine learning techniques to provide reliable estimates of precipitation-based recharge over the whole Perth Basin (including the Gnangara system). By combining estimates of baseflow, groundwater evaporation, and extraction, groundwater recharge was estimated over the Perth (testing site) and Gnangara (calibration site) systems using downscaled Groundwater Storage Anomalies (GWSA) from the Gravity Recovery and Climate Experiment (GRACE) mission. The random forest regression (RFR) model was used to downscale the spatial resolution of GRACE to 0.05° (approx. 5 km), providing estimable signals over the relatively small calibration site (∼2,200 km2) in order to discern any meaningful signals from the original GRACE resolution. Our study reveals that downscaled signals from GRACE can be used to provide precipitation-based recharge estimates for groundwater systems accurately. However, the growing impacts of climate change, which has led to sporadic precipitation patterns over Western Australia, can limit the efficiency of satellite remote sensing methods in estimating recharge, especially in deep and complex aquifers.