The distance travelled by an animal, when determined by using global positioning system (GPS) coordinates, is usually calculated assuming linear movement between the recorded coordinates. When using long sample intervals, some movement may be overlooked if linear movement between each recorded position is assumed, because of the tendency of livestock to move in meandering paths. Conversely, overestimation of the true distance travelled could occur with short sample intervals because of the accumulation of extra distance due to GPS measurement error. Data from 10 experiments were used to explore the effect of paddock size and GPS sampling rate on the calculation of distance travelled by free-ranging cattle. Shortening the sample interval increased apparent distance travelled according to a power function. As paddock size increased from <1 ha to >450 ha, distance travelled increased according to a logarithmic relationship; however, other variation between experiments could have affected these results. It was concluded that selecting an optimal GPS sampling interval is critical to accurately determining the distance travelled by free-ranging cattle.

Abstract. Timely and accurate monitoring of pasture biomass and ground cover is necessary in livestock production systems to ensure productive and sustainable management. Interest in the use of proximal sensors for monitoring pasture status in grazing systems has increased, since data can be returned in near real time. Proximal sensors have the potential for deployment on large properties where remote sensing may not be suitable due to issues such as spatial scale or cloud cover. There are unresolved challenges in gathering reliable sensor data and in calibrating raw sensor data to values such as pasture biomass or vegetation ground cover, which allow meaningful interpretation of sensor data by livestock producers. Our goal was to assess whether a combination of proximal sensors could be reliably deployed to monitor tropical pasture status in an operational beef production system, as a precursor to designing a full sensor deployment. We use this pilot project to (1) illustrate practical issues around sensor deployment, (2) develop the methods necessary for the quality control of the sensor data, and (3) assess the strength of the relationships between vegetation indices derived from the proximal sensors and field observations across the wet and dry seasons. Proximal sensors were deployed at two sites in a tropical pasture on a beef production property near Townsville, Australia. Each site was monitored by a Skye SKR-four-band multispectral sensor (every 1 min), a digital camera (every 30 min), and a soil moisture sensor (every 1 min), each of which were operated over 18 months. Raw data from each sensor was processed to calculate multispectral vegetation indices. The data capture from the digital cameras was more reliable than the multispectral sensors, which had up to 67 % of data discarded after data cleaning and quality control for technical issues related to the sensor design, as well as environmental issues such as water incursion and insect infestations. We recommend having a system with both sensor types to aid in data interpretation and troubleshooting technical issues. Non-destructive observations of pasture characteristics, including above-ground standing biomass and fractional ground cover, were made every 2 weeks. This simplified data collection was designed for multiple years of sampling at the remote site, but had the disadvantage of high measurement uncertainty. A bootstrapping method was used to explore the strength of the relationships between sensor and pasture observations. Due to the uncertainty in the field observations, the relationships between sensor and field data are not confirmational and should be used only to inform the design of future work. We found the strongest relationships occurred during the wet season period of maximum pasture growth (January to April), with generally poor relationships outside of this period. Strong relationships were found with multispectral indices that were sensitive to the green and dry components of the vegetation, such as those containing the band in the lower shortwave infrared (SWIR) region of the electromagnetic spectrum. During the wet season the bias-adjusted bootstrap point estimate of the R2 between above-ground biomass and the normalized ratio between the SWIR and red bands (NVI-SR) was 0.72 (95 % CI of 0.28 to 0.98), while that for the percentage of green vegetation observed in three dimensions and a simple ratio between the near infrared and SWIR bands (RatioNS34) was 0.81 (95 % CI of 0.53 to 1.00). Relationships between field data and the vegetation index derived from the digital camera images were generally weaker than from the multispectral sensor data, except for green vegetation observations in two and three dimensions. Our successful pilot of multiple proximal sensors supports the design of future deployments in tropical pastures and their potential for operational use. The stringent rules we developed for data cleaning can be more broadly applied to other sensor projects to ensure quality data. Although proximal sensors observe only a small area of the pasture, they deliver continual and timely pasture measurements to inform timely on-farm decision-making.

Crop phenology modeling often involves determining variety‐specific growing degree day thresholds, or parameterizing mechanistic crop models. In this work, we used machine learning methods to develop models that provide daily predictions of the probability that rice (Oryza sativa) crops had reached the panicle initiation and flowering growth stages. These per‐date classifications were summarized into per‐paddock growth stage transition dates, which were then compared with field‐sampled reference data, encompassing 15 rice varieties, 10 years, and 380 sites. Leave‐one‐year‐out cross validation was used to provide realistic estimates of model errors. Compared with more complex and computationally intensive algorithms, logistic regression produced competitive results (mean cross‐season validation RMSE 3.9 and 5.2 days for panicle initiation and flowering, respectively). Logistic regression had additional advantages: providing confidence of growth stage predictions at each date (as it is a probabilistic algorithm), and straightforward explainability (as model parameters directly indicated how the various input variables contributed to growth stage predictions). Input variables included accumulated weather, rice variety, and sowing methods. The models were applied to forecasting phenology transition dates of the rice crops planted throughout the Murray and Murrumbidgee valleys. In addition, recommendations for optimal sowing dates were developed, using simulations involving more than 40 years of weather data, with the goal of minimizing the risk of cold‐temperatures during the microspore growth phase, which can severely degrade yield in temperate rice growing regions.