Ensemble random forest filter: An alternative to the ensemble Kalman filter for inverse modeling

The ensemble random forest filter (ERFF) is presented as an alternative to the Kalman (EnKF) for inverse modeling. EnKF a data assimilation approach that forecasts and updates parameter estimates sequentially in time observations are collected. updating step based on experimental covariances computed from of realizations, given linear combinations differences between forecasted system state values. ERFF replaces combination update with non-linear function represented by forest. This way, relationships parameters be updated can captured, better produced. demonstrated log-conductivity identification piezometric head several scenarios varying degrees heterogeneity (log-conductivity variances going 1 up 6.25 (ln m/d)2), number realizations (50 or 100), (18 36). In all scenarios, works well, reconstructing spatial while matching observed heads at selected control points. For benchmarking purposes, compared restart find superior used (small typical applications). Only when grows 500 match performance ERFF, albeit more than double computational cost.