Multi‐frequency oscillatory pumping, in which the groundwater is extracted during a half period, and then reinjected, has recently been used to characterize aquifer heterogeneity. After the initial transition time, a steady periodic head can be observed at the observation well with constant amplitude and phase shift. However, the efficacy of utilizing multiple frequencies to enhance parameter estimation in hydraulic tomography is debatable. Most studies suggest that using multiple frequencies in joint inversion can gradually improve aquifer imaging. However, some researchers argue that additional frequencies add little to the resolution of hydraulic properties. This raises the question: how much non‐redundant information do multiple frequencies provide? In this study, we derive closed‐form analytical sensitivity maps (i.e., Fréchet kernels) of oscillatory pumping in 2D, unbounded, confined weakly heterogeneous aquifers by the sensitivity equation. It is found that the sensitivity maps of amplitude or phase shift are very similar to the sensitivity of drawdown to hydraulic parameters in constant‐rate pumping. Fréchet kernels indicate that multi‐frequency information, much like multi‐time information, can be instrumental for hydrogeological parameter inversion. According to the Fréchet kernels and their derivatives, we propose a method to select the optimal observation frequency, that is, the distance between the observation well and the pumping well is converted to frequency based on the hydraulic background values. Finally, we compare the effectiveness of single‐frequency and multi‐frequency observations in inverse transmissivity and storativity modeling using the iterative ensemble smoother. The results show that multi‐frequency oscillatory pumping can better characterize aquifer heterogeneity than a single frequency.
Accurately estimating water flux in variably saturated soils is crucial, yet traditional methods, such as the Richardson‐Richards equation, often suffer from some limitations. As an advancement of Physics‐Informed Neural Networks (PINNs), we introduce the Adaptive Constrained Neural Network (ACNN), designed to estimate variably saturated soil water flux using sparse volumetric water content (VWC) data alone, without requiring initial conditions (ICs), boundary conditions (BCs), or soil hydraulic constitutive relationships. The framework integrates domain decomposition methods and adaptive constraints within a sequential training framework. Testing on synthetic cases and soil column experiments demonstrates its capability to achieve unified modeling of variably saturated flow, seamlessly linking the unsaturated and saturated zones. It effectively addresses issues of incomplete knowledge and noisy data, not only mitigating ill‐conditioning compared to traditional PINNs, but also reducing flux estimation errors caused by uncertainties in ICs and BCs compared to numerical methods. Notably, under conditions of rapid moisture content change, ACNN successfully captures instantaneous flux changes at the boundary with an accuracy of approximately R2 = 0.9, highlighting its potential in applications such as estimating surface infiltration, evapotranspiration, and groundwater recharge.
Solutions of the linearized Richardson–Richards Equation (RRE) for one‐dimensional soil water flow in layered soils with sinusoidal flux in the frequency domain are derived. We evaluate the accuracy of our analytical and other analytical solutions by comparing them with results from a standard numerical model. Our analytical solution agrees with the numerical solution under multi‐layered heterogeneous soil, while others disagree. We also demonstrate the capability of the proposed solution to simulate soil moisture dynamics under a realistic, multi‐frequency flux case. The procedure described in the paper is valid for any series of arbitrary periodic flux superpositions for layered heterogeneous . Moreover, our solution is efficient in the calculation compared with numerical solutions, especially when dealing with long‐time series soil moisture, which can provide a validation of numerical models.
Time series of groundwater head in a semi‐infinite, one‐dimensional confined aquifer due to variable boundary forcing can be decomposed into two signals: head rise from a sudden boundary increase, and head fluctuation from boundary sinusoidal changes. Analyzed from time and frequency domains, these signals reconstruct the original series via convolution or superposition, offering insights into transmissivity ( T ). In this study, the time‐domain head data and frequency‐domain fluctuation data are used to estimate T . For a two‐zone aquifer with a pair of T values, the information content on T from the head at a given time is similar to that of the amplitude of the fluctuation at a certain frequency. In addition, the phase shift of the fluctuation also contains non‐redundant information on T . We introduce two concepts, that is, equivalent and interpreted T , to describe representative T , with which homogeneous aquifer can produce the same head/amplitude/phase shift signal and its temporal/frequency derivative as those observed in a heterogeneous aquifer. By applying Fréchet kernels as a spatial weight, we provide a connection between equivalent/interpreted T and the heterogeneous spatial distribution of T . Based on Monte Carlo simulations, we compare the equivalent/interpreted T against the local‐scale geometric mean T . The equivalent T from high‐frequency phase shift produces the better estimation. Finally, we investigate the effectiveness of observations at different frequencies and times for estimating heterogeneous T during a hydraulic tomography survey. The results show the amplitude and phase shift of multi‐frequency fluctuations can better characterize aquifer heterogeneity than head from the time‐domain perspective alone.