Soil bulk density (BD) is an important indicator of soil quality, productivity, compaction, and porosity. Nevertheless, soil BD is a dynamic feature that changes with time and space. Therefore, the direct measurement of soil BD is time-consuming, tedious, and expensive, particularly for large-scale investigations. Therefore, several pedotransfer functions (PTFs) using global soil databases are employed to estimate soil BD from readily available soil properties. However, using global PTFs on local or regional scales not covered during the PTFs development might require a further attempt to recalibrate them prior to any use. Up our knowledge, there is no general conclusion whether recalibration of the existing PTFs or development of the new PTFs is the best choice when working in local to regional scales. Because the development of local or regional PTFs is often constrained by data sparsity and therefore proper selection of determinant readily available characteristics might not be guaranteed. Therefore, in this research we tested the effect of dataset size (with N = 15, 30, 60, and 360) on performance of recalibrated global PTFs as well as newly developed regional PTFs focusing on improved estimation of soil BD at the local to regional scales. We also tested the reliability of soil BD estimations provided in the global dataset of SoilGrids comparing them to predictions from recalibrated and developed PTFs. The results show that overall performance of the PTFs was very similar independent of the calibration dataset size (Nash-Sutcliffe efficiency (NSE) = 0.20 to 0.41; RMSE = 0.10 to 0.13 Mg m-3). The original uncalibrated existing PTFs generally provided inaccurate estimates of soil BD (NSE = -22.07 to 0.30; RMSE = 0.12 to 0.60 Mg m-3), while their recalibration with small regional datasets (N = 15 to 60) resulted in a better performance even compatible to or much better than the locally developed PTFs when applied over the same datasets. Meantime, our investigation
