Pasture potential estimation for dairy farms: A hybrid modelled approach using integrated industry and spatial data

Abstract

Pasture eaten (t DM/ha) is known to be a primary driver of profitability in New Zealand pastoral dairy farm systems. However, locally relevant estimates of achievable home-grown feed eaten (the “pasture potential”) are difficult to derive at industry scale. This study used a multi-stage hybrid methodology to estimate farm-specific pasture potential across a large industry dataset. Each farm was spatially enriched with environmental covariates to assign it to a farm typology
system. Within each farm type, a local nitrogen response slope was estimated based on pasture eaten values for nearby farms within an expanding radius (5-120 km); local pasture eaten was then estimated as a nominal zero-N figure, to remove the confounding impact of fertiliser usage. The pasture potential value was defined as the 90th percentile of home-grown feed eaten within farms of the same type, estimated via quantile regression. A machine learning model (XGBoost) was then developed to impute values for sparsely populated typologies. The analysis identified an average distance from pasture potential of 2.5-3.5 t DM/ha across the different regions. These findings highlight the potential
benefit to farms having land-specific insights into their pasture potential and support the identification of pathways for improving farm performance in terms of profitability and emissions footprint.