Abstract: Spatial distribution mapping of topsoil available phosphorus content of cultivated land is essential for precise agricultural management and soil environmental assessment. Most research has focused on sufficient soil samples to map topsoil phosphorus content of cultivated land in flat areas. However, there are few studies on soil available phosphorus mapping based on sparse samples in the hilly areas of southern China. Jian’ou City was selected as the study area, which is a hilly area in southern China and has the largest cultivated land area among all county-level cities in Fujian Province. A total of 96 soil measurements, Sentinel-2 remote sensing data with a spatial resolution of 10 m, and climate and topographical variables were used to predict topsoil (0–20 cm) available phosphorus content. Random forest models with five combinations of environmental variables were constructed and their performance was compared for model prediction. Three assessment criteria, namely the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) were used to evaluate the performance of random forest models for five combinations of environmental variables. The results showed that the prediction accuracy of the random forest model using climate variables, topographic variables, and soil pH values significantly improved after adding remote sensing variables, with an R² increase from 0.36 to 0.59 and an RMSE decrease of 20.34%. In addition, the random forest model using all variables (remote sensing, topography, climate, and soil pH) obtained the optimal performance (R² = 0.59, MAE = 19.04 mg∙kg⁻¹, RMSE = 25.26 mg∙kg⁻¹) among five combinations of environmental variables. Therefore, remote sensing variables are of great value for the mapping of soil available phosphorus based on sparse samples, and we suggested that the use of remote sensing variables should be increased in future studies to improve prediction accuracy. Remote sensing variables, climate variables, topographic variables, and soil pH could explain 22.87%, 30.64%, 30.38%, and 16.11% of the topsoil available phosphorus content, respectively. Furthermore, the spatial distribution of soil available phosphorus content in the study area was found to be mainly affected by the mean annual temperature, soil pH, soil moisture index, and elevation. The spatial distribution maps of soil available phosphorus content by the five random forest models were similar. High values of topsoil available phosphorus content were distributed in the central and western regions, whereas low values were distributed in the eastern and southern regions. The spatial variation of the soil available phosphorus in the distribution map produced by the optimal random forest model with total environmental variables was the most precise. Therefore, a random forest model that uses all variables (i.e., soil pH, topographic, remote sensing, and climate) can be used as a robust method to resolve soil available phosphorus content mapping with sparse soil samples in the hilly regions of southern China. Thus, this research can provide some guidance for other researchers interested in mapping the soil available phosphorus content in the hilly regions of China.