New empirical equation to estimate the soil moisture content based on thermal properties using machine learning techniques

Information about soil moisture content is crucial for the sustenance of agricultural system because it helps to make decision on irrigation scheduling and water management. However, the conventional procedures for determining the soil moisture content need much effort, and time-consuming with large dataset. It is known that soil thermal properties have significant influence on the moisture content of soil. Therefore, the soil moisture content can be determined based on the soil thermal properties, which can easily be measured with portable equipment known as KD2 Pro. This study presents an alternative technique for estimating the soil moisture content from thermal properties using machine learning (ML). Actual measurements of moisture contents and thermal properties at seventy-five points were used. Three ML techniques including artificial neural network (ANN), fuzzy logic (FL), and support vector machine (SVM) were used to predict the moisture content of soil from its thermal properties (thermal conductivity, thermal diffusivity, and specific heat). The results show that all the three techniques (ANN, FL, and SVM) were able to predict moisture content with acceptable errors where the average absolute error is around 5.65%. Moreover, a new empirical equation is presented to allow quick estimation of the moisture content. Ultimately, the developed models can be employed to predict the soil moisture content in any farmland with known thermal properties, which will lead to cost reduction and less time and effort to determine soil moisture content.