Abstract. Sedimentary charcoal records are widely used to reconstruct regional changes in fire regimes through time in the geological past. Existing global compilations are not geographically comprehensive and do not provide consistent metadata for all sites. Furthermore, the age models provided for these records are not harmonised and many are based on older calibrations of the radiocarbon ages. These issues limit the use of existing compilations for research into past fire regimes. Here, we present an expanded database of charcoal records, accompanied by new age models based on recalibration of radiocarbon ages using IntCal20 and Bayesian age-modelling software. We document the structure and contents of the database, the construction of the age models, and the quality control measures applied. We also record the expansion of geographical coverage relative to previous charcoal compilations and the expansion of metadata that can be used to inform analyses. This first version of the Reading Palaeofire Database contains 1676 records (entities) from 1480 sites worldwide. The database (RPDv1b – Harrison et al., 2021) is available at https://doi.org/10.17864/1947.000345.

Aim Biomisation has been the most widely used technique to reconstruct past regional vegetation patterns because it does not require an extensive modern pollen dataset. However, it has well‐known limitations including its dependence on expert judgement for the assignment of pollen taxa to plant functional types (PFTs) and PFTs to biomes. Here we present a new method that combines the strengths of biomisation with those of the alternative dissimilarity‐based techniques. Location The Eastern Mediterranean‐Black Sea Caspian Corridor (EMBSeCBIO). Taxon Plants Methods Modern pollen samples, assigned to biomes based on potential natural vegetation data, are used to characterize the within‐biome means and standard deviations of the abundances of each taxon. These values are used to calculate a dissimilarity index between any pollen sample and every biome, and thus assign the sample to the most likely biome. We calculate a threshold value for each modern biome; fossil samples with scores below the threshold for all modern biomes are thus identified as non‐analogue vegetation. We applied the new method to the EMBSeCBIO region to compare its performance with existing reconstructions. Results The method captured changes in the importance of individual taxa along environmental gradients. The balanced accuracy obtained for the EMBSeCBIO region using the new method was better than obtained using biomisation (77% vs. 65%). When the method was applied to high‐resolution fossil records, 70% of the entities showed more temporally stable biome assignments than obtained using biomisation. The technique also identified likely non‐analogue assemblages in a synthetic modern dataset and in fossil records. Main conclusions The new method yields more accurate and stable reconstructions of vegetation than biomisation. It requires an extensive modern pollen dataset, but is conceptually simple, and avoids subjective choices about taxon allocations to PFTs and PFTs to biomes.

No abstract

This supplementary contains: 4 SI Table 1. Information of the charcoal records (sites and entities) in the Reading Palaeofire Database version 1. Latitude and longitude are in decimal degrees, and elevation in metres above/below sea level. Fields where information could be available but was never recorded or has subsequently been lost are represented by -999999, fields where we were unable to obtain this information but it could be included in subsequent updates of the database are represented by -777777, fields where specific information is not applicable are represented by -888888. SI Table 2. List of pre-defined valid choices for restricted fields in the Reading Palaeofire Database version 1. SI Table 3: List of charcoal measurement units currently used in the Reading Palaeofire Database version 1 SI Figure 1: Supplementary Figure 1. Summary of the stages used to select the optimum RBacon age models for from ageR. Plots A.-C. show the modelling output from ageR for an example entity from the RPD (Geral core), where the optimum age model selected by ageR A. is a table ranking the age model scenarios by the lowest area between the prior and posterior accumulation rate curves. Note that only the top 5 model scenarios of a total of 25 run for this entity are listed B. Shows the plots for the prior and posterior accumulation rates and the area between curves for the top 5 model scenarios.C. Is the top ranked RBacon age model (Accumulation rate = 15, thickness = 10) which was visually checked to verify that the interpolation through the dates was valid and consistent with the dates. In this example, the top ranked model scenario selected by ageR (Accumulation rate = 15, thickness= 10) was accepted as the chosen model scenario as the interpolation through the dates is valid. SI Figure 2. An example of alternative model scenario selection where the top ranked ageR model scenario is deemed to be inaccurate. In this example, the top ranked model scenario from King Tableland Swamp (accumulation rate = 45, thickness = 5)(A.) with the lowest area between the prior and posterior accumulation rate curves (B.) does not accurately represent the date at 157.5cm. This age was included by the original authors and lies in stratigraphic order with the other dates. Therefore, this model is rejected in favour of the model with the next lowest abc score which accurately reflects the dates included (ageR model ranking 3 in A.). The RBacon plot for this age model scenario is shown in D. (accumulation rate = 90, thickness = 5) and is more accurately and precisely modelled through the dates than the model selected by ageR. Site name Entities (#) Elevation (m) Latitude (°) Longitude (°) Site Type Water depth (m) Basin size (km 2 ) Citation(s)

<p>The circum-Mediterranean region is characterized by high climatic diversity derived from its orographic heterogeneity and the influence of global marine and atmospheric circulation patterns. The region also has a long and dynamic history of human occupation dating back to ~ 8000 years BP.  The complexity of this area is a challenge for reconstructing the dynamics of the vegetation through the Holocene. Rule-based approaches to reconstructing changing vegetation patterns through time are insufficient as they require the imposition of subjective boundaries between biomes and can be affected by known biases in pollen representation.  We have developed and tested a new method that characterises biomes as a function of observed pollen assemblages based on a similarity index, conceptually related to the likelihood function, which takes account of within-biome variability in taxon abundances. We use 1181 modern pollen samples from the EMBSeCBIO database and assign these samples to biomes as represented in a map of potential natural vegetation that was developed using machine learning. The method was applied down-core to reconstruct past vegetation changes. Preliminary results show that this new methodology produces more accurate biome assignments under modern conditions (<80% accuracy) and more stable down-core reconstructions, apparently reducing the "flickering switch" problem found when using the traditional biomisation method for this purpose. Climate-induced vegetation changes are observable on a sub-regional scale in the Eastern Mediterranean through the Holocene. Most of the records show a change from humid to more arid biomes between 4000 and 3000 years BP. However, they are distinct subregional patterns in the expression and timing of wetter conditions during the Holocene. Mountain regions appear to show more muted changes during the Holocene, although there are biome shifts everywhere across the Pleistocene-Holocene transition.</p>