Spatiotemporal modelling to predict soil moisture for sustainable smart irrigation

In this paper, we have proposed spatiotemporal soil moisture modeling to improve understanding of soil moisture and variability for sustainable irrigation practices. Increasing population growth and climatic changes have intensified the need to implement effective measures to conserve water in irrigation practices. Effective irrigation in parkland influences the overall plant growth processes including the final appearance of the plants. Furthermore, over-irrigation and poor water management in parks and gardens lead to wastage of water which may result in seepage, runoff, and leaching of nutrients into nearby streams. Conversely, under-irrigation results in reduced plant growth and unappealing appearance. Thus, appropriate irrigation management is required to maintain a lush green landscape. Due to large variability in soil properties, environmental conditions and landscape features, the soil moisture level might not be uniformly distributed within a given landscape. Hence, it is essential to understand the soil moisture distribution in the field. There are few existing soil moisture modeling exist but these modelings have not considered multiple vertical depth and time domain. Using Internet of Things(IoT) enabled Cyber–physical system and machine learning techniques, this paper has presented a soil moisture modeling for multiple vertical soil depth for a robust understanding of soil’s and plants need for sustainable sprinkling of the water. The proposed model has used both statistical and deep learning techniques to understand moisture variability in both vertical depth and time. The proposed model has used Correlation Index (CI) to understand multi depth moisture variability on the influence of soil types and local weather parameters, such as precipitation and temperature. To predict moisture data in multi depth, the model has evaluated the use of both statistical machine learning models, such as Support vector regression(SVR) and linear regression (LR), and deep learning models, such as Long Short-Term Memory (LSTM) using seasonal and non-seasonal dataset. The experiment has revealed closely related variability pattern between wind speed and soil moisture in multi depth whereas soil type shows a loosely related variability pattern with moisture in higher depth. With these diverse datasets, the proposed machine learning model has achieved almost 90% success rate to predict moisture content in higher depth using data of lower depth to reduce cumbersome sensors array deployment in higher depth for live moisture visualization. This proposed multiple depth and time domain-based model can enable a smart water dispersing system to conserve water and make a positive contribution to a sustainable future.