A Hadoop based Framework for Soil Parameters Prediction
Nowadays, in view of the exponential growth of the populations, the consumption of the fresh water resources rises more and more. One can distinct different irrigation technique that have been implemented to optimize the consumption of freshwater resources in agriculture, such as flood irrigation and drip irrigation. As demand for freshwater rises, water availability decreases due to climate change. For this reason, most of the researchers are currently working on the automation of irrigation systems. These automated systems rely on the advances of machine learning, massive data and IoT techniques for building new innovative and effective solutions. Thus, the integration of predictive process represents a vital step for anticipating and assuming the adaptation to the impact of climatic changes in agriculture, through an accurate prediction of soil and environment features, and analysis of its dependencies as well. In this paper, we propose an adaptive online learning (OL) framework for supporting irrigation decision by soil features diagnosis and forecast with a focus on the implementation of Three prediction methods the extreme gradient boosting, random forest and the Auto Regressive Moving Average based on Hadoop/Map-Reduce environment to predict soil moisture, depending on soil temperature and time in various depth. At the end we discussed the accuracy of these methods in different conditions.
Irrigation, Time-series, Random Forests, ARIMA and XGBoost.