Characterizing Multi-Timescale Soil Moisture Memory across Australia's CosmOz Network

Memory effects are ubiquitous in geophysical systems, arising from internal dynamics and interactions with external forcings across multiple timescales. Within land surface systems, soil moisture memory is a key factor governing land–atmosphere feedbacks, influencing the intensity, persistence, and predictability of hydro-climatic extremes such as droughts and floods. This study quantifies soil moisture memory across the CosmOz-Australia network using long-term Cosmic Ray Neutron Sensing (CRNS) observations and characterizes memory across land surface and meteorological timescales.The CRNS technique offers a novel, field-scale measurement of soil moisture with high temporal resolution and a time-varying effective sensing depth, thereby overcoming the limitations of traditional point-scale observations and enabling the robust characterization of soil moisture memory across various timescales. Despite the widespread application of CRNS data for soil moisture monitoring and validation, their potential for systematic, multi-timescale soil moisture memory estimation has not yet been explored.This study estimates the short-term energy-limited (τs) and long-term water-limited (τL) memory components applying a hybrid stochastic-deterministic modeling framework that represents rapid surface-layer responses and slower root-zone and subsurface controls at the land surface scale. In addition, to capture memory at the meteorological scale, we estimate a non-parametric, model-free entropy-based effective memory timescale that quantifies information persistence beyond linear correlations, and compute the e-folding memory timescale as a standard measure of decorrelation. Results reveal pronounced spatial heterogeneity in soil moisture memory across Australia. Short-term memory is consistently low (median τs ≈ 0.3–1.0 days), reflecting rapid drying over the effective sensing depth and low memory in drylands. Long-term memory (median τL ≈ 4–11 days) is highest over the humid eastern and south-eastern regions, consistent with a water-limited evapotranspiration regime where higher precipitation frequency, lower aridity, finer soils, and denser vegetation enhance root-zone storage and slow anomaly decay. Entropy-based effective memory ranges from approximately 19 to 36 days, indicating substantial information retention at monthly timescales, while e-folding timescales extend up to ~70 days in temperate and monsoon-influenced regions. The strong spatial agreement between entropy-based and correlation-based metrics suggests robust and consistent soil moisture memory regimes across Australia, highlighting their dependence on hydro-climate, soil texture, and vegetation. The results provide observation-based characterization of multi-timescale soil moisture memory using CRNS data, with important implications for land surface model evaluation, drought diagnostics, and sub-seasonal to seasonal climate forecasting.