Diagnosis of Model Errors With a Sliding Time‐Window Bayesian Analysis

Deterministic hydrological models with uncertain, but inferred-to-be-time-invariant parameters typically show time-dependent model structural errors. Such errors can occur if a process is active in certain time periods nature, not resolved by the model. missing processes could become visible during calibration as best-fit values of parameters. We propose formal time-windowed Bayesian analysis to diagnose this type error, formalizing question \In which period time-series does statistically disqualify itself quasi-true?" Using evidence (BME) performance metric, we determine how much data windows support or refute Then, track BME over sliding obtain dynamic, (tBME) and search for sudden decreases that indicate an onset error. tBME also allows us perform formal, likelihood-ratio test against data. Our proposed approach designed detect error occurrence on various temporal scales, especially useful modelling. illustrate applying our method soil moisture modeling. indicator several synthetic real-world cases vary sources scales. Results prove usefulness framework detecting dynamic models. Moreover, sequence posterior parameter distributions helps investigate reasons provide guidance improvement.