Please use this identifier to cite or link to this item:
Title: Efficient Forecasting of Crop Water Demand
Authors: Sidhu, Ravneet Kaur
Supervisor: Rana, Prashant Singh
Kumar, Ravinder
Keywords: Forecasting;Crop Water Demand;LSTM;Time Series;APSIM
Issue Date: 23-Jun-2021
Abstract: Water plays an important role in the creation of everything we produce. Around 70% of freshwater around the world and up to 95% in several developing countries is used for farming. Agriculture is the main sector for global water use. The growing water scarcity is one of the major challenges for agriculture. However, water resources need to be used more efficiently and sustainably in agricultural production as population growth, coupled with increasing urbanization and industrialization, will cause less and less water to become less available for cereal production. Traditional systems of irrigation and water use and management in agriculture are highly inefficient with low water productivity, and cannot ensure long-term sustainable food security. Many innovative technologies, such as micro-irrigation (low volume irrigation), offer an efficient alternative to traditional flood irrigation. Drip irrigation (surface and subsurface) systems provide water (and nutrients) to the crop root zone, where it can be utilized most efficiently. Recent advances in the science of sensor technologies and the internet of things (IoT) can be useful in the automation of irrigation systems, which can help in addressing the emerging challenges of labor shortages and inefficiency of water use in agriculture. The use of wireless sensor networks (WSNs), IoTs, and communication technology for the automation of irrigation in general and drip irrigation in cereal crops, in particular, is an entirely new and futuristic field. The purpose of this study has been to forecast the crop water demand efficiently. To sustain and improve agriculture production from the depleting freshwater resources. In this work, various parameters affecting the crop water demand have been identified. The problem of forecasting crop water demand has been formulated as classification, regression, and a deep learning problem. Hence, various machine learning models and two deep models have been implemented, and the most suitable for irrigation demand forecasting have been selected. Machine learning classification models have been found to predict the need for irrigation accurately. While the regression and deep learning models have predicted, the amount of water to be applied quite well. The best performing models were ensembled, and an algorithm defining the approach was proposed and implemented. We also used dynamic simulation model APSIM (Agriculture Production System Simulator) which is being used commercially for irrigation scheduling in many countries, however, it needs intensive data to calibrate and simulate the irrigation scheduling. We compare the APSIM results to our results to see if there is any loss of prediction accuracy if we used data-driven models which are based on readily available datasets.
Appears in Collections:Doctoral Theses@CSED

Files in This Item:
File Description SizeFormat 
PhD Thesis - Signed - Ravneet Kaur - Final.pdf2.05 MBAdobe PDFThumbnail

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.