For stable soybean cultivation, techniques are necessary for evaluating the distribution of soil moisture in crop fields. UAV (Unmanned Aerial Vehicles) sensing has been attracting attention as a tool to evaluate the distribution of soil characteristics in the field. Recently, the use of AI (Artificial Intelligence) technology with UAV observation images has also shown promise. On the other hand, there are still few examples of UAV observations of soil physical properties and moisture content, and the estimation accuracy is not high. One of the factors is the limited number of soil samples that can be obtained. Thus, in order to enable the highly accurate spatial prediction of the moisture environment in a soybean field from a small amount of ground truth data, this study conducted the following analysis: (1) Moisture content was measured at 64 points in the field using a soil penetration sensor. Multispectral, thermal infrared, and DSM (Digital Surface Model) images of the target field were acquired by UAV observations. (2) Assuming that the soil moisture was homogeneous within a certain range, we augmented the ground-truth data by mapping the measured moisture content to the pixel value of each image contained within the range. (3) Using the dataset prepared by data augmentation, we tried to construct a highly accurate AI model to estimate the moisture content distribution in the soybean field by random forest regression. As a result, the distribution of moisture content was successfully estimated with high accuracy. This indicated that a sufficient sample size for machine learning was prepared by our data augmentation methods based on an assumption that the soil properties are uniform within a certain range. Particularly, in the present case, buffers in the 1.0 m to 1.5 m diameter range (about 15 to 40 samples in a buffer) were effective in improving the accuracy of the model. In addition, the model showed that the pixel values of thermal infrared images and the elevation of the DSM as the most important features for estimating the moisture content in the field.