A Fourier Neural Operator for Coupled Hygrothermal Ranking of Low-Cost Wall Materials under Measured Climate Forcing

Abstract

Selecting low-cost indigenous wall materials that limit heat penetration is critical for indoor thermal comfort in hot-dry rural regions, where summer temperatures routinely exceed 45 °C. The thermal performance of the porous earthen materials used in such construction is inseparable from their moisture state, yet material rankings are typically established under static-moisture or heat-only assumptions. We present a two-stage framework that ranks five indigenous wall materials (mud brick, clay–straw adobe, lime-stabilised bamboo panel, fired clay brick, and lime–mud composite) under fully coupled heat and moisture transport. First, a Crank–Nicolson finite difference method (FDM) solves the one-dimensional coupled Künzel heat-and-moisture (HAM) system over a 30-day summer window, driven by measured NASA POWER climate reanalysis rather than idealised periodic forcing, and generates 2000 solutions across a 12-dimensional material and boundary parameter space by Latin Hypercube sampling. Second, a Fourier Neural Operator learns the parameter-to-solution operator μ → (T(x,t), w(x,t)) and acts as a fast surrogate for large-scale material ranking and global sensitivity analysis. The trained operator attains relative L² field errors of 0.11% on temperature and 1.0% on moisture, reproduces the FDM material ranking exactly, and evaluates a new configuration 293× faster than direct simulation; a data-efficiency study establishes the number of solver runs required to resolve the closest inter-material contrast. Dynamic moisture redistribution shifts the inner-surface temperature by up to 0.4 K relative to the static-moisture case while leaving the recommended material unchanged, and a variance-based sensitivity analysis identifies the indoor temperature and, among material properties, the dry thermal conductivity and wall thickness as the strongest drivers of inner-surface temperature, with the hygric parameters contributing negligibly. The framework supports evidence-based, cost-effective wall material selection for post-flood reconstruction in similar hot-dry rural contexts.