Gradual Variation Analysis for Groundwater Flow

Groundwater flow in Washington DC greatly influences the surface water quality in urban areas. The current methods of flow estimation, based on Darcy’s Law and the groundwater flow equation, can be described by the diffusion equation (the transient flow) and the Laplace equation (the steady-state flow). The Laplace equation is a simplification of the diffusion equation under the condition that the aquifer has a recharging boundary. The practical way of calculation is to use numerical methods to solve these equations. The most popular system is called MODFLOW, which was developed by USGS. MODFLOW is based on the finite-difference method in rectangular Cartesian coordinates. MODFLOW can be viewed as a “quasi 3D” simulation since it only deals with the vertical average (no z-direction derivative). Flow calculations between the 2D horizontal layers use the concept of leakage. In this project, we have established a mathematical model based on gradually varied functions for groundwater data volume reconstruction. These functions do not rely on the rectangular Cartesian coordinate system. A gradually varied function can be defined in a general graph or network. Gradually varied functions are suitable for arbitrarily shaped aquifers. Two types of models are designed and implemented for real data processing: (1) the gradually varied model for individual (time) groundwater flow data, (2) the gradually varied model for sequential (time) groundwater flow data. In application, we also established a MySQL database to support the related research. The advantage of the gradually varied fitting and its related method does not need the strictly defined boundary condition as it is required in MODFLOW.

💡 Research Summary

The paper addresses a fundamental limitation of the widely used MODFLOW groundwater modeling system: its reliance on a regular Cartesian grid, quasi‑3D vertical averaging, and strictly prescribed boundary conditions. While MODFLOW solves the diffusion (transient) or Laplace (steady‑state) equations using finite‑difference discretization, these assumptions become problematic in urban settings where aquifers have irregular shapes, heterogeneous hydraulic properties, and complex recharge/discharge boundaries. To overcome these constraints, the authors introduce a novel modeling framework based on Gradually Varied Functions (GVFs). A GVF is an integer‑valued function defined on the nodes of a general graph or network, constrained such that the absolute difference between adjacent nodes never exceeds one. This “gradual variation” condition enforces smoothness of the hydraulic head (or potential) while freeing the model from any predefined coordinate system. Consequently, GVFs can be applied directly to arbitrarily shaped aquifers represented as irregular networks of observation wells, stream‑aquifer interactions, or any spatial discretization that captures connectivity rather than geometry.

Two distinct GVF‑based models are developed. The first, a single‑time‑step model, takes a snapshot of groundwater observations (head, potential) and computes the optimal GVF surface that best fits the data. The fitting procedure combines a least‑squares misfit term with a Laplacian smoothing regularizer, yielding a sparse linear system that can be solved efficiently without explicit boundary specifications; the observed values themselves act as de‑facto boundary constraints. The second model extends this concept to sequential (time‑series) data. By augmenting the spatial graph with a temporal dimension, the authors construct a three‑dimensional (space‑time) network and solve a discretized transient flow equation that respects both spatial smoothness and temporal continuity. This approach naturally interpolates missing observations, preserves the evolution of hydraulic gradients, and reduces noise through temporal regularization.

The methodology is validated using real groundwater data from the Washington, D.C. metropolitan area. A MySQL relational database was built to store well locations, depths, timestamps, measured heads, and model parameters, ensuring reproducibility and facilitating future GIS integration. When compared with conventional MODFLOW simulations on the same dataset, the GVF models produced head distributions that closely matched MODFLOW in data‑dense regions, yet displayed more realistic behavior near boundaries where MODFLOW’s fixed Dirichlet or Neumann conditions often impose artificial constraints. Importantly, the GVF framework required far less memory because it does not generate a full Cartesian mesh; only adjacency lists are needed. The resulting linear systems are highly sparse and amenable to parallel solution on modern multi‑core CPUs or GPUs, offering scalability to large, irregular networks.

Beyond accuracy and computational efficiency, the authors highlight several practical advantages. Because GVFs do not demand explicit boundary condition specification, model setup is dramatically simplified, reducing the expertise and time required for urban groundwater studies. The graph‑based representation aligns naturally with existing monitoring networks, allowing new wells or sensors to be added without re‑meshing. Moreover, the integration with a relational database streamlines data management, version control, and post‑processing workflows such as uncertainty analysis or scenario testing.

The paper also acknowledges current limitations. The present GVF formulation assumes linear flow behavior and a single averaged aquifer layer; it does not yet incorporate nonlinear hydraulic conductivity, multi‑layer leakage, or surface‑water–groundwater exchange processes that are often critical in urban hydrology. Future work is proposed to extend GVFs to nonlinear regimes (e.g., using iterative linearization), to embed multi‑layer graph structures that capture vertical leakage, and to couple the framework with data assimilation techniques for real‑time forecasting. The authors envision a web‑based decision‑support platform where incoming sensor data automatically update the GVF model, providing city planners with near‑real‑time insight into groundwater level trends, contaminant transport risk, and the impacts of recharge management strategies.

In summary, this study demonstrates that Gradually Varied Functions constitute a flexible, computationally lightweight alternative to traditional finite‑difference groundwater models. By leveraging the inherent connectivity of monitoring networks rather than imposing a rigid grid, the approach delivers accurate head reconstructions, reduces the burden of boundary condition specification, and scales efficiently to complex urban aquifers. The work opens a promising pathway toward more adaptable, data‑driven groundwater modeling tools that can better support sustainable water resource management in densely populated regions.