Work

GRAZE-FUSE: Grazing-Informed Integration of APSIM and Sentinel-2 for Pasture Biomass Monitoring

Year: 2026

Type: article

Abstract: Timely estimation of pasture biomass in rotational grazing systems is difficult because field measurements, cloud-affected satellite observations, and process-based simulations differ in temporal support and responsiveness to grazing. We developed GRAZE-FUSE (Grazing-Informed APSIM–Sentinel Fusion Framework), which combines Sentinel-2 biomass estimates from the SPECTRA-FOR model with APSIM simulations using a farm-specific grazing-interval smoothing window and a linear fusion model. The analysis used 7312 field biomass records from five intensively managed dairy farms in Tasmania. GRAZE-FUSE was evaluated using nested leave-one-farm-out (LOFO) validation, in which each farm was withheld in turn, and model fitting and parameter selection were conducted using only the remaining farms. Across 520 pooled out-of-fold weekly predictions, GRAZE-FUSE achieved R2 = 0.876, RMSE = 103.3 kg DM/ha, MAE = 82.5 kg DM/ha, and bias = −1.2 kg DM/ha. Farm-specific R2 ranged from 0.814 to 0.893. Ablation analysis showed that a simple 50:50 APSIM–SPECTRA-FOR mean achieved R2 = 0.819 and RMSE = 124.5 kg DM/ha, while OLS fusion without grazing information improved performance to R2 = 0.872 and RMSE = 104.6 kg DM/ha. Incorporating farm-specific grazing intervals produced a further, smaller improvement to R2 = 0.876 and RMSE = 103.3 kg DM/ha. These results indicate that most of the predictive improvement arose from integrating complementary process-model and satellite information, while grazing-informed temporal weighting provided additional refinement. GRAZE-FUSE therefore provides a transferable framework for continuous pasture biomass estimation across the study farms; however, validation across additional regions, management systems, and years is required before broader operational deployment.

Source: Remote Sensing

DOI: 10.3390/rs18172911

Authors: Michael Gbenga Ogungbuyi, Matthew Tom Harrison, Caroline Mohammed, Richard Azu Crabbe

Institutions: University of Tasmania, Charles Darwin University, Tasmanian Institute of Agriculture

Language: English

Topic: Remote Sensing in Agriculture

Subfield: Ecology

Field: Environmental Science

Domain: Physical Sciences

SDGs: Zero hunger, Life on land, Climate action

Open Access status: gold

Estimated APC paid: $3,258