Paper Title
A Study of Soil Moisture Estimation Using Logarithmic Agricultural Drought Indices

Abstract
In recent years, climate change and the impact of extreme weather conditions have led to a food crisis, with fluctuating temperatures and severe drought affecting agricultural practices. Dryland farming, in particular, heavily relies on natural rainfall patterns to determine the types of crops to cultivate, following the principle of "right crop at the right place (time)." However, abnormal rainfall disrupts cultivation management models established by farmers based on their experience. Additionally, variations in soil structure within specific local regions further complicate the implementation of appropriate supplementary irrigation to ensure adequate soil moisture for stable production and crop quality. This poses a significant challenge for agriculture in addressing climate change and sustainable development. Currently, related research has proposed the use of satellite imagery combined with developed agricultural drought indices to map soil moisture of farmlands. However, the consideration that the spectral reflectance of images and soil moisture should follow a logarithmic relationship has not been incorporated into the agricultural drought indices. Therefore, this study modifies the original agricultural drought indices to logarithmic ones and evaluates the precision of both models in mapping soil moisture in wheat fields in Kinmen, Taiwan. Furthermore, while satellite imagery is suitable for large-scale soil moisture mapping, it is not suitable for small regional planting areas in Kinmen. Hence, unmanned aerial vehicles equipped with multispectral sensors were used for soil moisture mapping, and Support Vector Regressor (SVR), Gradient Boosting Regressor (GBR), and Convolutional Neural Network Regressor (CNNR) were employed to estimate soil moisture contents (SMCs) and compare their estimation precisions. The results indicate that the logarithmic agricultural drought index could be helpful in improving estimation precision, and the SMC mapping precision is better with GBR. The highest R2 between estimated and measured soil moisture values is 0.93, with a Root Mean Square Error (RMSE) of 1.63%. Keywords - Logarithmic Agricultural Drought Index, Soil Moisture, Unmanned Aerial Remote Sensing System, Multi-Spectral Images.