Smoothing techniques improve trend detection in rate-transient analysis (RTA) of tight gas reservoirs

Production data used in rate-transient analysis (RTA) are often contaminated by noise and operational artifacts, obscuring diagnostic signatures and introducing uncertainty in flow-regime identification and reservoir characterization. While outlier detection (OD) methods are commonly applied, they cannot fully eliminate anomalous behavior, leaving residual noise that distorts rate-normalized pressure (RNP) trends. This study demonstrates that smoothing is not merely a visualization tool but a critical preprocessing step for preserving flow-regime signatures in noisy RNP data and improving the reliability of RTA interpretation. Smoothing methods enhance the visualization of scattered data and improve the accuracy of extracted flow-regime times and reservoir characteristics by predicting representative data points. This helps recover the underlying RNP trend, particularly when OD methods are limited. We evaluated five smoothing techniques using noisy synthetic RNP datasets generated from a homogeneous, single-layer, low-permeability gas reservoir produced from a single-fractured vertical well. Noise and outliers were introduced using additive white Gaussian noise (AWGN) at multiple noise-contamination levels and noise-presence scenarios to represent realistic production conditions. We assessed the performance of each method based on goodness-of-fit and the accuracy of the extracted flow-regime characteristics, including linear-flow times and permeability—squared fracture half-length (\(\:k{{x}_{f}}^{2}\)) estimation. The locally weighted scatterplot smoothing (LoWeSS) method demonstrates superior performance, achieving the lowest average deviation while effectively reducing noise and preserving underlying trends in the log-log RNP plot. An optimal window size of 19 provides the most reliable results, particularly in the boundary-dominated flow (BDF) regime. The application of LoWeSS results in significantly improved reservoir parameter estimation, with approximately 30% higher accuracy in \(\:k{{x}_{f}}^{2}\) derived from square-root time analysis. This is supported by the improved regression quality of the fitted line during linear flow. These findings establish smoothing as an essential component of RTA workflows, reducing subjectivity in the interpretation of diagnostic plots (associated with the start and end of flow regime times) and improving the robustness of reservoir characterization (associated with the \(\:k{{x}_{f}}^{2}\)) and production forecasting (associated with the Arps \(\:b\)-factor), even when applied after OD methods such as the general mixture model (GMM). The findings are further supported by application to field data, demonstrating improved clarity in flow regime identification under real production conditions and enabling more reliable engineering decision-making.