Discretization error in time-domain wave simulations often appears as intermittent, spike-like fluctuations synchronized with wave arrivals, reflections, and interference. Such nonstationary behavior hinders cross-time comparison and spatial localization of dominant error growth. This work proposes an error-scale estimation framework for solution verification that enables interpretable spatiotemporal characterization of discretization error in time-domain room acoustic simulations. Given a reference solution, we compute the pointwise error e(t,r) between a finite-difference time-domain simulation and a modal expansion solution under rigid-wall boundary conditions. From a nonnegative error series x(t,r) = phi(e(t,r)) (here phi(e) = |e|), we estimate a representative error scale mu_hat(t,r) by generalized nearly isotonic regression. The estimator suppresses spike-like local variations while preserving the overall increasing tendency of accumulated error, allowing small local decreases induced by periodicity and interference. The resulting mu_hat(t,r) serves as an interpretable envelope of error growth. We demonstrate the method on a 2D rigid-wall room. While raw error snapshots |e(t,r)| contain localized peaks that vary drastically over time and hinder cross-time comparison, the estimated error-scale maps mu_hat(t,r) provide stable spatial patterns that reveal how discretization error spreads and accumulates. In particular, mu_hat(t,r) highlights regions of amplified error growth near boundaries and modal antinodes, enabling localization of dominant error-growth mechanisms in time-domain wave simulations.