This study aimed to develop empirical pasture dry matter (DM) yield prediction models using an unmanned aerial vehicle (UAV)-borne sensor at four flying altitudes. Three empirical models were developed using features generated from the multispectral sensor: Structure from Motion only (SfM), vegetation indices only (VI), and in combination (SfM+VI) within a machine learning modelling framework. Four flying altitudes were tested (25 m, 50 m, 75 m and 100 m) and based on independent model validation, combining features from SfM+VI outperformed the other models at all heights. However, the importance of SfM-based features changed with altitude, with limited importance at 25 m but at all higher altitudes SfM-based features were included in the top 10 features in a variable importance plot. Based on the independent validation results, data generated at 25 m flying altitude reported the best model performances with model accuracy of 328 kg DM/ha. In contrast, at 100 m flying altitude, the model reported an accuracy of 402 kg DM/ha which demonstrates the potential of scaling up this technology at farm scale. The spatial-temporal maps provide valuable information on pasture DM yield and DM accumulation of herbage mass over the time, supporting on-farm management decisions.

One of the most valuable and nutritionally essential agricultural commodities worldwide is milk. The European Union and New Zealand are the second- and third-largest exporting regions of milk products and rely heavily on pasture-based production systems. They are comparable to the Australian systems investigated in this study. With projections of herd decline, increased milk yield must be obtained from a combination of animal genetics and feed efficiencies. Accurate pasture biomass estimation across all seasons will improve feed efficiency and increase the productivity of dairy farms; however, the existing time-consuming and manual methods of pasture measurement limit improvements to utilisation. In this study, Sentinel-2 (S2) band and spectral index (SI) information were coupled with the broad season and management-derived datasets using a Random Forest (RF) machine learning (ML) framework to develop a perennial ryegrass (PRG) biomass prediction model accurate to +/−500 kg DM/ha, and that could predict pasture yield above 3000 kg DM/ha. Measurements of PRG biomass were taken from 11 working dairy farms across southeastern Australia over 2019–2021. Of the 68 possible variables investigated, multiple simulations identified 12 S2 bands and 9 SI, management and season as the most important variables, where Short-Wave Infrared (SWIR) bands were the most influential in predicting pasture biomass above 4000 kg DM/ha. Conditional Latin Hypercube Sampling (cLHS) was used to split the dataset into 80% and 20% for model calibration and internal validation in addition to an entirely independent validation dataset. The combined internal model validation showed R² = 0.90, LCCC = 0.72, RMSE = 439.49 kg DM/ha, NRMSE = 15.08, and the combined independent validation had R² = 0.88, LCCC = 0.68, RMSE = 457.05 kg DM/ha, NRMSE = 19.83. The key findings of this study indicated that the data obtained from the S2 bands and SI were appropriate for making accurate estimations of PRG biomass. Furthermore, including SWIR bands significantly improved the model. Finally, by utilising an RF ML model, a single ‘global’ model can automate PRG biomass prediction with high accuracy across extensive regions of all seasons and types of farm management.

This study explored the relationships between moderate resolution imaging spectroradiometer (MODIS) NDVI observations with both measured and simulated fractional green cover (FGrC), leaf area index (LAI), and above ground biomass (AGB) for dryland wheat in Australia. A total of 37 paddocks in north-western Victoria, Australia, were sampled during 2003-2006 for AGB at anthesis, and for FGrC, NDVI (from an active optical sensor), and AGB during 2012. The 2012 FGrC and NDVI measurements were fitted to MODIS NDVI, resulting in positive, linear relationships when the MODIS NDVI values were ≤ 0.80. Measured AGB was also positively, linearly related to MODIS summed NDVI, resulting in an overall R ² of 0.81 and root mean square error (RMSE) of 1397 kg/ha. Crop simulations were run for the fourteen paddocks from 2003 to 2006, and six paddocks from 2012. Four crop phenological points were selected to extract corresponding NDVI and simulated crop parameters: emergence, peak LAI, the mid-point between emergence and peak LAI, and anthesis. Linear models were fit between the MODIS NDVI and simulated values of FGrC, LAI, and AGB. Overall, the highest R ² values corresponded to using all of the dates for FGrC (R ² = 0.82) and AGB (R ² = 0.92), and anthesis dates for LAI (R ² = 0.74). For FGrC and AGB, the RMSE with simulated parameters were comparable or better than the equivalent results from the in situ measurements (note that there were no LAI in situ measurements to compare with). The results support the notion for extending the value of the MODIS NDVI using crop simulation models. The combination of remotely sensed and simulation data might offer regional maps of spatial AGB and ultimately grain yield, which would have high value for research, resource management, policy, and potentially, crop management.

The nutrition of grazing ruminants can be optimized by allocating pasture according to its nutritive characteristics, provided that nutritive concentrations are determined in near-real time. Current proximal spectrometers can provide accurate predictive results but are bulky and expensive. This study compared an industry standard, ‘control’, proximal spectrometer, often used for scientific estimation of pasture nutrient concentrations in situ (350–2500 nm spectral range), with three lower-cost, ‘next-generation’, handheld spectrometers. The candidate sensors included a hyperspectral camera (397–1004 nm), and two handheld spectrometers (908–1676 nm and 1345–2555 nm respectively). Pasture samples (n = 145) collected from two paddocks on a working Australian dairy farm, over three timepoints, were scanned in situ by each instrument and then analysed for eight nutritive parameters. Chemometric models were then developed for each nutrient using data from each sensor (split into 80:20 calibration and validation sets). According to Lin’s Concordance Correlation Coefficient (LCCC) from independent validation (n = 29), the hyperspectral camera was the best candidate instrument (LCCC from 0.31 to 0.85, and 0.67 on average), rivalling the control sensor (LCCC from 0.41 to 0.84, and 0.67 on average). Consideration was given to whether the hyperspectral camera’s success was due to spectral range or data type/capture method. It was found that the 400–920 nm (trimmed) spectral region was slightly less sensitive in principle to nutrient concentrations than higher spectral ranges. Therefore, the predictive performance of the camera was attributed to the advantage of gathering data as hyperspectral images as opposed to single spectra.

Most image recognition and object detection models fail to perform with more than 90% accuracy and precision due to a lack of automated processes in image quality assessment (IQA). An image-based dataset may often contain poor-quality images that are not always feasible to identify manually based on visual judgment. This study aimed to develop an automated IQA method using parallel computing and a computer vision approach with GPU integration to identify high-quality images of dairy cattle that can be used to develop a prediction model based on biometric traits in the future. The images were collected from a commercial dairy farm in Victoria, Australia. Four key features, namely, structural similarity index (SSI), image gradient (g), image entropy (E), and Laplacian (L) filter, were considered in developing the IQA model. The features were mathematically calculated by MATLAB from reference images, with mean values of those images considered as thresholds to meet the IQA criteria. In addition, the images were rated one to five using a star system, with high-quality images rated as five-star images. The final IQA model was able to produce highly accurate outcomes for both reference-based (internal) and reference-free (independent) validations. The internal validation produced accuracy = 0.99 and precision = 0.97, whereas the independent validation demonstrated accuracy = 0.95 and precision = 0.93. This study has investigated the development and validation of the IQA approach for a future cattle recognition model in dairy farms.