In a world with ever increasing shortages of food production due to increasing populations and depletion of resources, the need for new technologies and techniques for sustainable and efficient agriculture with long term financial, environmental and cultural benefits are critical. An area of scientific study concerning crop-production management called Precision Agriculture (PA) is a concept based on integrating modern information technologies such as Big Data Analytics, GPS (Global Positioning Systems), remote sensing technology, and GIS (Geographic Information Systems) to monitor and control the precise application of agricultural inputs/outputs for optimized crop growth. One drawback of traditional approaches to PA is that it has been more of a reactionary approach in which only the current states of the field are provided to the growers without incorporating accurate predictive forecasts.
The research team at the University of St. Thomas (UST) strongly believe that a proactive approach is more important than a reactionary approach in PA applications. In the proactive approach, predictive models generate forecast reports to the growers, predicting the possible evolution of vegetation states and future risks in the field.
In order to provide highly accurate predictions to growers in real time, the predictive models must integrate tremendous amounts of information from numerous sources such as various sensor data (i.e. temperature, wind, soil pH, moisture), and aerial multi-spectral imagery data (i.e. vegetation index and current severity of infestation). Furthermore, historical ground and aerial data also needs to be integrated into the predictive models for trend analysis in the field. Due to the amount of information that needs to be processed and the complexity of predictive models, the research team at UST plans to utilize two cluster machines in its Center of Excellence for Big Data (CoE4BD) to facilitate this intensive information processing and prediction computation.
The first cluster machine (consisting of 24 nodes, where each node has 2 Quad-Core CPUs, 12GB RAM, and 1TB disk space) will be dedicated to process both the aerial multi-spectral image data and the ground sensor data; while the second cluster machine (which is under construction and will have magnitudes more processing power than the first) will build several predictive models that integrate current and historical data from the ground and air. More specifically, the first cluster machine will be dedicated to process multi-spectral images captured by small Unmanned Airborne Vehicles (sUAVs). This cluster machine will identify the health of the vegetation, status of the soil, infestations, and their geo-locations. A stitching map of the imaged data will also be produced from this cluster. The processed data identified by the first cluster machine will then be sent to the second cluster where a forecast of the vegetation and soil health as well as any risks such as future infestation areas will be predicted. The confidence of each prediction will be integrated in the report to the growers.
The UST team believes their system will ultimately shift the farming paradigm to more anticipatory rather than reactionary techniques, which in turn will increase production yield as well as reduce consumed resources (fertilizers, pesticides, fuel) by using resources more effectively.
Keywords: Big Data, Predictive Models, Unmanned Airborne Vehicles, Multi-Spectral Images, Clustered CPUs.
