Fossil charcoals as a means for reconstruction of past hydroclimate variability through the Last Glacial Maximum: a case study from two archaeological sites in the Winter and Year-round Rainfall Zones, South Africa

The hydrology of the Last Glacial Maximum (LGM) in South Africa has been an active area of research over the last six decades due to the richness of many proxy records and the high preservation potential of deposits of this age at sites recording human occupation. Here we review the merits of using a Weighted Average – Partial Least Squares regression on the archaeological charcoal records from Elands Bay Cave (EBC) and Boomplaas Cave (BPC) to quantitatively reconstruct palaeoclimate. These sites are both spatially and temporally ideal to track the changes in the Southern Hemisphere Westerlies – the proposed driver of the change in hydroclimate and have been used to infer LGM climate conditions. A database of the modern-day distribution of the taxa identified in the stratigraphy at EBC and BPC was created using the Global Biodiversity Information Facility, this was then paired with the modern climate data from WorldClim to perform a WA-PLS regression to reconstruct Mean Annual Temperature and Total Annual Precipitation. Many of the WA-PLS regressions reconstructed temperature differences from the LGM to present close to 5°C (consistent with regional records). The TAP reconstructions suggest decreased precipitation during the LGM at BPC and despite unrealistic values for the EBC charcoal record, they suggest higher TAP values during the LGM. The TAP reconstructions posed more problems and highlighted some major flaws in the reconstructions. When compared to the pollen record from EBC, MAT reconstructions reflect the same trend however the TAP reconstructions from pollen indicate slightly lower TAP values during the LGM. Due to the nature of these records, there are some important differences that influence the outcomes of the reconstruction. These being the biases which exist within the records themselves and propagating through to the reference databases. With the use of a Canonical Correspondence Analysis, we can compare the modern distribution of the taxa and better understand what the reconstructions are reflecting and in some cases failing to reflect. While a WA-PLS regression is widely used for reconstructions, other regressions should be compared alongside it.