In this article, we study stochastic homogenization of non-homogeneous Gaussian free fields $$\Xi ^{g,\textbf{a}} $$
Ξ
g
,
a
and bi-Laplacian fields $$\Xi ^{b,\textbf{a}}$$
Ξ
b
,
a
. They can be characterized as follows: for $$f=\delta $$
f
=
δ
the solution u of $$\nabla \cdot \textbf{a} \nabla u =f$$
∇
·
a
∇
u
=
f
, $$\textbf{a}$$
a
is a uniformly elliptic random environment, is the covariance of $$\Xi ^{g,\textbf{a}}$$
Ξ
g
,
a
. When f is the white noise, the field $$\Xi ^{b,\textbf{a}}$$
Ξ
b
,
a
can be viewed as the distributional solution of the same elliptic equation. Our results characterize the scaling limit of such fields on both, a sufficiently regular domain $$D\subset \mathbb {R}^d$$
D
⊂
R
d
, or on the discrete torus. Based on stochastic homogenization techniques applied to the eigenfunction basis of the Laplace operator $$\Delta $$
Δ
, we will show that such families of fields converge to an appropriate multiple of the GFF resp. bi-Laplacian. The limiting fields are determined by their respective homogenized operator $${{\,\mathrm{\bar{\textbf{a}}}\,}}\Delta $$
a
¯
Δ
, with constant $${{\,\mathrm{\bar{\textbf{a}}}\,}}$$
a
¯
depending on the law of the environment $$\textbf{a}$$
a
. The proofs are based on the results found in Armstrong et al. (in: Grundlehren der mathematischen Wissenschaften, Springer International Publishing, Cham, 2019) and Gloria et al. (ESAIM Math Model Numer Anal 48(2):325-346, 2014).