【Objective】 As an important indicator of soil quality, black soil layer thickness plays an irreplaceable role in sustainable soil development, food security and ecological functions. However, analyses based on soil profile survey data are often based on small sample sizes and small regional scales, and most of them are based on point data statistics only. However, the studies lacked spatial variability prediction analyses, hence, there is an urgent need for rapid surveys of the thickness of the black soil layer and high-performance spatial prediction methods.【Method】 In this paper, a series of sample data of black soil thickness at 357 sample points in Heilongjiang Province were obtained by the rapid acquisition method of "shallow excavation + deep soil drilling" for black soil thickness at multiple burrows in newly constructed sample points. The spatial variability of black soil thickness and its uncertainty were predicted through the optimisation of parameters of the Random Forest Prediction Model (RPFPM). The impacts of the different burrow observations and their mean samples on the optimization of the model's prediction accuracy and stability were analyzed, and the spatial prediction potentials of the model were evaluated.【Result】 The predicted average thickness of the black soil layer in the arable land in the study area was 53.42 cm, and the new method of rapid acquisition and prediction of black soil layer thickness was effective and can be used as an alternative to the profiling method. The spatial variation explanatory power R² of the optimized random forest model for predicting black soil thickness reached 60%, which could finely depict the spatial differentiation of black soil thickness. Also, the randomness of a single observation burrow at a sample point could change the importance value of the covariates predicted by the model, and affect the spatial prediction of the distribution of the black soil thickness. Compared with the spatial prediction on the mean value of several observations, the spatial prediction on a single observation had lower accuracy for uncertainty assessment of the spatial distribution and significantly reduced prediction performance. Interestingly, the cross-validation metrics and scatterplot analyses indicated that the optimized Random Forest model had a stable spatial prediction potential of the black soil thickness.【Conclusion】 This study provides a new perspective and new ways for high-precision and rapid investigation and prediction of black soil layer thickness.