Quantification and attribution of spectral variation in irrigated perennial tree crops

Abstract

Satellite reflectance data are used for crop classification models globally, yet research on their robustness under varying temporal and spatial conditions is limited. We examined the effects of space, time, water availability and ontogeny on crop spectral distributions cross a large, heterogenous landscape, Australia's Murray-Darling Basin. We aimed for broad generality of our findings, and so used a large ground-truthed dataset from 2015, 2018, and 2021, covering multiple catchments and 12 crops. We characterised spectral distributions for each crop in each year and catchment before testing for differences due to space and time, and associated with the amount of water available from irrigation and rainfall. We also tested bareness metrics (as a surrogate for perennial crop ages) in almond plantations. We found that crop type explained the most variation, confirming the utility of satellite imagery for crop classification. Catchment and year both explained small but significant variation, emphasising the need for data collected over a range of spatial and temporal contexts. Water availability explained a significant but small proportion of variation in the data set (<1 %), suggesting that crops were receiving sufficient water across the observed range or that spectral signatures did not vary much as a result. The effect of bareness metrics suggested possible significant variation caused by ontogeny. This study affirms the validity of spectral imagery for crop classification studies, whilst underscoring the importance of spatial, hydrologic and ontogenetic context. Future studies in crop classification should consider these factors to enhance the robustness of their models.