We consider the stochastic heat equation $\partial_{s}u =\frac{1}{2}Δu +(βV(s,y)-λ)u$, with a smooth space-time stationary Gaussian random field $V(s,y)$, in dimensions $d\geq 3$, with an initial condition $u(0,x)=u_0(\varepsilon x)$ and a suitably chosen $λ\in{\mathbb R}$. It is known that, for $β$ small enough, the diffusively rescaled solution $u^{\varepsilon}(t,x)=u(\varepsilon^{-2}t,\varepsilon^{-1}x)$ converges weakly to a scalar multiple of the solution $\bar u(t,x)$ of the heat equation with an effective diffusivity $a$, and that fluctuations converge, also in a weak sense, to the solution of the Edwards-Wilkinson equation with an effective noise strength $ν$ and the same effective diffusivity. In this paper, we derive a pointwise approximation $w^\varepsilon(t,x)=\bar u(t,x)Ψ^\varepsilon(t,x)+\varepsilon u_1^\varepsilon(t,x)$, where $Ψ^\varepsilon(t,x)=Ψ(t/\varepsilon^2,x/\varepsilon)$, $Ψ$ is a solution of the SHE with constant initial conditions, and $u^\varepsilon_1$ is an explicit corrector. We show that $Ψ(t,x)$ converges to a stationary process $\tilde Ψ(t,x)$ as $t\to\infty$, that $\mathbf{E}|u^\varepsilon(t,x)-w^\varepsilon(t, x)|^2$ converges pointwise to $0$ as $\varepsilon\to 0$, and that $\varepsilon^{-d/2+1}(u^\varepsilon-w^\varepsilon)$ converges weakly to $0$ for fixed $t$. As a consequence, we derive new representations of the diffusivity $a$ and effective noise strength $ν$. Our approach uses a Markov chain in the space of trajectories introduced in Gu, Ryzhik, and Zeitouni, "The Edwards-Wilkinson limit of the random heat equation in dimensions three and higher," as well as tools from homogenization theory. The corrector $u_1^\varepsilon(t,x)$ is constructed using a seemingly new approximation scheme on a mesoscopic time scale.