AbstractThis study presents a sequential algorithm for the identification of the position s of the moving boundary of a variable one‐dimensional or two‐dimensional domain from discrete measurements, temperatures and fluxes, collected at the fixed boundary. The inverse problem is solved by minimization, with respect to s, of a penalized output least square criterion defined on a sliding time horizon of length τ. At every time step, several iterations are performed to estimate the unknown s. Each iteration consists in a guess of s, a computation of the corresponding value of the output y (direct model) and the criterion J, and a step towards a new estimation of s. The impact of the different parameters, space and time discretization intervals, regularization coefficient, dimension of the unknown parameter s, length of the observation horizon, choice of the observed output (temperature or flux) and choice of the direct model is thoroughly analysed for an analytical test case.