| dc.description.abstract | The study of proper water management practices is of prime importance due to the ever-
increasing population and rapid industrialization which results in shortage of portable water
supplies throughout the world. The current water supplies are not expected to meet the increasing
demand in the upcoming decades which could in result affect the socio-economic stability and have a detrimental effect on human livelihood.
About 30% of the current municipal supplies in the world are used for outdoor irrigation activities such as gardening and landscaping purposes. These numbers are on the rise due to the ever increasing human population. Due to the current inefficient landscape practices, substantial
amount of water is lost in the form of runoff. This poses a great threat to the environment with its potential for transporting fertilizers and pesticides into storm sewers and, eventually, surface waters. Thus, this study focuses on designing a Machine Learning approach which would act as a Decision Support System (DSS) to irrigate turfgrasses to minimize runoff in the plots while maintaining the quality of the turfgrasses.
For this, a robust Machine Learning approach named as Radial Basis Function - Support Vector Machine (RBF-SVM) was proposed which was trained on the synthetic data generated from the datapoints recorded during the year 2015-16 and 2016-17 at the Turfgrass Laboratory in
Texas A & M University, College Station. For each of the approaches, the target variable was changed and the number of features were varied in each case to see which gives the best results.
Among all the target variables, predicting the Soil Wetting Efficiency Index, devised by Wherley, et. al.[33] was the most applicable as it is one of the most generic approaches since it is not site-specific and gave the highest validation testing accuracy of 90%. Thus, the latter approach was used in the ASIS controller to observe the robustness of the algorithm in controlling the effectiveness of the irrigation cycle. Until now, only few irrigation cycles have been scheduled and experimental data are still
being collected from the facility. Preliminary results suggested that the Machine Learning algorithm has the potential to save water as it helped in efficient regulation of irrigation cycles and even achieved a goal of zero runoff in two of the irrigation runs. The Green Cover percentage of the plots where the proposed ASIS controller was mounted showed an increment of about 12%, thereby validating the fact that the quality of turfgrasses was also maintained. With more irrigation
cycles which would be scheduled over time, the proposed Machine Learning approach is expected to perform better with increase in observations and may nullify runoff eventually. | en |
