Using vegetation indices from nanosatellites for timely prediction of pasture biomass

• Generalised additive models (GAMs) outperformed linear regression in modelling green and total pasture biomass. • SAVI achieved the best prediction of green biomass, while NDVI performed better for total biomass estimation. • GAMs captured non-linear vegetation–biomass relationships, improving modelling accuracy across seasonal dynamics. • Planetscope imagery effectively predicted fractional cover components under variable cloud and grazing conditions. • Vegetation index responses varied with grazing intensity and rainfall, emphasising model and index selection in biomass monitoring. This study evaluates the potential of Planetscope nanosatellite imagery for predicting pasture biomass and estimating fractional cover under frequent cloud cover conditions. By integrating Normalised Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI), and Enhanced Vegetation Index (EVI) with field-measured total standing dry matter (TSDM), we compared the performance of linear regression and generalised additive models (GAMs) in modelling green and total biomass at the paddock scale. Results show that NDVI, SAVI, and EVI positively correlate with green biomass but exhibit negative correlations with total biomass due to the influence of senesced vegetation. While linear regression provides moderate accuracy, GAM outperforms it, capturing non-linear relationships and improving biomass estimation, particularly for green biomass. Among the indices, SAVI demonstrated the highest predictive performance for green biomass (R 2 = 0.52, RMSE = 487.24 kg DM/ha, MAE = 328.62 kg DM/ha), while NDVI was more effective in estimating total biomass (R 2 = 0.43, RMSE = 1123.81 kg DM/ha, MAE = 871.83 kg DM/ha). Seasonal variations revealed that NDVI responds more sensitively to grazing events, whereas SAVI and EVI offer more stable trends by mitigating soil and atmospheric effects. Furthermore, Planetscope imagery effectively captured fractional cover variability, supporting real-time livestock grazing and land management in the context of climate resilience and agricultural sustainability. These findings highlight the importance of model selection in pasture biomass modelling and reinforce the advantage of GAMs over traditional linear regression when handling non-linearity in vegetation indices.