Sensor data analytics is an essential part of agricultural decision support systems. Farmers and crop managers need fit for purpose environmental derived data and predictive models to help them monitor their crop and adjust control options for optimal production. However, these data products can be sensitive to environmental data captured with variable quality, format, location and time. With the continuing increase in the number of sensors deployed in our environment, selecting sensors that are fit for purpose is a growing challenge. Ontologies that represent sensors and observations can form the basis for semantic sensor data infrastructures. Such ontologies may help to cope with the problems of sensor discovery, data integration, and re-use, but need to be used in conjunction with algorithms for sensor selection and ranking. This paper describes a method for selecting and ranking sensors based on the requirements of predictive models. It discusses a Viticulture use case that demonstrates the complexity of semantic modelling and reasoning for the automated ranking of sensors according to the requirements on environmental variables as input to predictive analytical models. The quality of the ranking is validated against the quality of outputs of a predictive model using different sensors.   This talk will be an extended version of the paper presentation at ISWC 14. It will also include a description of the umbrella project Sense-T (http://www.sense-t.org.au ) and a demonstration of the Web application developed in the project.