Precision variable rate nitrogen for dryland farming on waterlogging Riverine Plains of Southeast Australia?

Precision agriculture, using satellite navigation, has grown in popularity around the world. Because of its practical uses for land preparation, sowing, nutrient applications, pest (weed, pathogen, invertebrate) management, stress sensing and harvest recording, many major equipment manufacturers include precision variable rate capability as standard fittings. GIS (Geographic Information System) maps may include georeferenced soil chemistry information and detailed historical harvest yield data. We integrate such data to examine Precision Variable Rate Nitrogen (PVRN) applications with whole-farm management information and rainfall records in a district where waterlogging frequently reduces crop yields. Given wide variations in growing season rainfalls (GSR) and soils in the district, we test year-to-year stability of rainfed crop-yield rankings over time on 90x90m GIS grid-areas in large paddocks (over 100 ha). Variations in historical yield-quartile rankings of grid-areas across GSR levels over time are observed; some areas yield best at some GSR levels but not others, such that the best-yielding part of a paddock one year may be poorest in the next. We answer the question: "Why would a farmer in this district choose to apply a uniform moderate rate of N to a paddock at sowing even though in possession of precision variable rate-capable (PVR) equipment, georeferenced electromagnetic conductance (EM38) data and crop-yield map data for that paddock in many past seasons?" We show that soil conditions in the study district challenge the economic value of PVRN versus uniform rates in farming systems prone to waterlogging. If full-season GSR were reliably predictable early in the season, applications of N could be based on a rule calling for 40 kg/ha N/t of attainable yield at that GSR and grid-area EM38 level, minus sampled soil-N. Unfortunately, GSR is notoriously unpredictable. We simulate whole-farm financial risk profiles (CDFs of simulated decadal cash margins with varying prices and yields, minus all variable, fixed and capital costs) assuming moderate uniform N rates, as practiced in the study area, on two model farms; one with low and one with high-fixed-costs, given historical variations in GSR and prices. Assuming PVRN requires annual geo-referenced soil nutrient sampling of each hectare, these added costs could be covered by a 1% increase in yields across all wheat and canola crops or a 7% decrease in applied N. We cannot reject the null hypothesis that PVRN is no more profitable than uniform applications in this district. Near-real-time NDVI may lower the cost of PVRN for late applications.