Machine vision tools for Smarter Irrigation of dairy pasture and cotton

Machine vision tools for Smarter Irrigation of dairy pasture and cotton

Site-specific irrigation enables application of irrigation when and where it is needed using variable-rate hardware on centre pivot and lateral move irrigation machines. Commercial and research tools have been developed to automate the labour-intensive process of developing and uploading irrigation prescription maps to commercial variable-rate hardware. These software tools have different data and computational requirements, and potentially perform in different ways depending on irrigation availability and field variability scenarios. Common approaches include filling the soil-water deficits based on evapotranspiration, or using soil-water sensors and/or canopy temperature sensors to control irrigation. However, these approaches may not optimise the irrigation, if there is limited water, nor will they adapt to the different crop water requirements during different growth stages. Data analytics approaches (e.g. ‘Model Predictive Control’) involve automated analyses of predictive soil-plant-atmosphere models using in-field weather, soil and plant data, to consider water availability. These potentially enable improvements over existing approaches. Software has been developed to implement and simulate these irrigation control strategies for cotton (McCarthy et al. 2010, 2014). These strategies have been transferred to the Australian dairy industry using a pasture growth model, and consider grazing management differences across a field. This enables implementation of data analytics ‘Model Predictive Control’ irrigation optimisation in dairy pasture fields.

A field trial has been conducted to compare the performance of four automated irrigation strategies at the TIA Dairy Research Farm (TDRF) in the 2019/20 summer season. The four irrigation strategies include uniform irrigation application, variable-rate irrigation based on a fixed historical map, soil-water deficit irrigation based on real-time data, and Model Predictive Control optimisation using available weather, soil and plant data with predictive modelling. The trial: (i) compared irrigation applied, energy use and productivity for five replicates these four alternative irrigation strategies needing different data requirements; (ii) conducted a cost benefit analysis for each irrigation strategy evaluated; and (iii) evaluated the ability to automate irrigation prescription map development. A plot-based comparison was conducted with five replicates of the four strategies following the approach of reported variable-rate trials (e.g. Hedley et al. 2011; Burdette Barker et al. 2018). A flow meter and energy sensor were installed on the centre pivot irrigation machine, soil-water sensors were installed in each soil-water deficit plot, and infield cameras were installed in each Model Predictive Control plot for assessing grazing status and pasture growth as required to calibrate the predictive model. The mean average error from the image analysis system for detecting pasture height as an indicator of yield, was 2.3 ± 1.9 cm. Periods of grazing were also detected as the pasture height reduced. This presentation will report the comparative performance of the irrigation strategies, and provide recommendations for use of commercial and research software tools for variable-rate irrigation in the Australian dairy industry.

Files associated with this item cannot be displayed due to copyright restrictions.