Integrating historical sources with climate simulations: Temperature and precipitation anomalies in southern Italy (1000–1800 CE)

This paper presents the first systematic palaeoclimatic reconstruction of southern Italy, primarily focusing on the Puglia region, from 1000 to 1800 CE. Although documentary sources provide a wealth of climatic information for this area, they primarily record anomalies rather than normal climatic conditions, offering no baseline for comparison. To address this issue, we compiled a relational dataset comprising 526 documentary climate events, 285 of which were selected for quantitative analysis using the Pfister seven-point scale. We developed a Bayesian state-space model integrating these observations with CHELSA-TraCE21k palaeoclimatic simulations, which provide informative priors linked by a first-order autoregressive process. Observations enter the model through an ordinal logistic likelihood. The resulting posterior should be interpreted as a simulation-anchored latent climate state, informed primarily by documented extremes, rather than as a direct reconstruction of centennial annual-mean conditions. The temperature posterior indicates predominantly negative anomalies from the 15th century onward, broadly consistent with a Little Ice Age cooling phase, although the magnitude is influenced by the strong representation of extreme cold events and the 16th-century estimate is largely determined by the prior and temporal persistence. The precipitation reconstruction reveals a more complex picture: the CHELSA-TraCE21k prior exhibits a progressive drying trend not reproduced by an alternative forced simulation (ModE-Sim), while the documentary posterior remains closely aligned with the forced simulation. This agreement suggests that the prior–posterior divergence may reflect a regional feature of CHELSA-TraCE21k, although alternative explanations, including documentary reporting bias and model assumptions, cannot be excluded. Posterior uncertainty reflects data availability, with narrower credible intervals in well-documented periods. We compare both posterior reconstructions with two products from the Modern Era Reanalysis framework (ModE-RA and ModE-Sim), using anomalies re-baselined to a common reference period, and find broad agreement for temperature but a more informative divergence for precipitation. This comparison assesses consistency across climate products and highlights the sensitivity of the inferred precipitation trajectory to the simulation product used as prior, rather than providing an independent validation of either reconstruction. This framework can be applied to other regions where discontinuous historical evidence must be integrated with simulation-derived climate baselines.