Abstract: The topographical and geomorphological conditions in complex Karst regions of southern China are special and highly intricate, which makes high-precision mapping pose significant challenges. This study explores the spatial distribution prediction methods of soil pH in Karst areas by comparing the performance differences of various spatial prediction models, based on 5 441 surface (0 - 20 cm) soil samples collected using zigzag sampling method during the Soil Testing and Fertilizer Recommendations (STFR, 2007-2009) in Long'an County. The study area was divided into Karst and non-Karst regions, four digital soil mapping models were constructed, including Ordinary Kriging (OK), Geographically Weighted Regression Kriging (GWRK), Linear Regression Kriging (LRK), and Random Forest Kriging (RFK). Results show that: ① The soil pH in the Karst regions is generally high, predominantly alkaline, while the non-Karst regions are mostly acidic. Acidic soils are concentrated in the southeastern non-Karst regions, whereas neutral to alkaline soils are more common in the Karst regions and their central transitional belts. ② There is a strong spatial autocorrelation, moderate variation, and obvious clustering characteristics in the soil pH of cultivated land in both Karst and non-Karst regions. The spatial variation of soil pH is mainly caused by structural factors. ③ The spatial distribution patterns of the prediction results from the four models are similar at both global and zonal scales, showing that acidic soils are concentrated in the southeastern part of the non-Karst regions, while neutral and alkaline soils are concentrated in the Karst regions and at the boundaries between the Karst and non-Karst regions within their central regions. ④ Structural factors such as long-term precipitation, soil type, annual temperature, and erosion modulus, as well as random factors such as drainage and irrigation, are the dominant factors affecting the soil pH of cultivated land in the Karst regions. ⑤ In terms of model accuracy, the order of global prediction accuracy from high to low is RFK>OK>RF>LRK>GWRK, with RFK performing the best (R²=0.628), followed by the OK model (R²=0.611). The OK model(R²=0.593) performs best in the Karst regions. The RF model (R²=0.621) has the highest accuracy in the non-Karst regions, followed by the OK model (R²=0.616). The OK model's global prediction does not show significant performance in high-value areas, while the LRK and GWRK models expand the high-value and low value areas after partition prediction. The RFK model for global prediction and the OK and RF models for partition prediction have the best accuracy and mapping details. When predicting soil pH in Karst areas, partition prediction at the county scale may not necessarily improve model accuracy. If environmental variables are readily available, the RFK model is recommended for prediction at the county scale. Otherwise, the OK model is recommended for prediction in both Karst and non-Karst areas. This study provides methodological support for high-precision mapping of soil pH, soil acidification regulation, and precision fertilization in the complex Karst areas of southern China.