GrassLogic: A Grass Growth System for the Prediction of Grassland Production

Grasslands take up a significant part of land use in Europe covering more than a third of the European agricultural area. Grasslands have a key role in feeding herbivores and ruminants as well as providing essential ecosystem benefits, including erosion control, water management and water purification. Ireland is one of the most important players in European grassland production due to its privileged combination of climatic and soil conditions. In many Irish farms, grasslands pastures are the main crop to be cultivated throughout the years. The farming of grassland in Ireland has been a major economic pilar for thousands of years and nowadays account for the use of 92% of the agricultural land area in the country. In terms of feed for ruminant production systems, grass is considered to be the one of the best options and also the more cost-efficient alternative. In Irish farms there are significant variability on production of grass dry matter due to factors such as soil, fertilization and also management. The expansion of dry matter grass productivity is crucial for matching a higher feed demand on farms as stocking rate rises. Concerns about the effect of agriculture on climate change and the development of sustainable models are expanding and constitute a significant challenge nowadays. The increasing of stock to grass systems would also augment the generation of methane and excreta. The pasture-based scheme has the capacity to be sustainable, in part, by utilizing grass as a carbon sink and expanding the grazing season in a way to reduce slurry releases. Nevertheless, this sustainable prospect hangs on the use of precision technologies, such as machine learning and AI, for supervising variables such as grass growth, soil, and weather for the optimization of resources and minimization of climate impacts. Assessing pasture mass is the first step towards effective management decisions such as daily pasture allocation, surplus conservation, and supplementary feeding. The current work has proposed a grass growth prediction system which is able to estimate daily quantities of dry matter for individual paddocks based on canopy, rotation length, weather, location, and soil factors. The system makes use of a state-of-the-art based mechanistic model for predictions that are further adjusted by an extra layer of fuzzy logic and other machine learning regression algorithms. Validation was performed with a 28- year period of grass growth ground-truth data, where it was observed the system suitability in the prediction of grass growth throughout the years. Furthermore, in 5order to complement the mechanistic model predictions, it was proposed a system for measuring the amount of fresh weight directly from the paddock. Instead of using specific sensors, this system makes use of standard RGB images acquired with smartphones that are inputted to convolutional neural networks (CNN’s). A large dataset of images was acquired in different grass plots in Irish farms, which were used to train and validate this algorithm. By analysing the results, it was verified the feasibility of the grass fresh weight estimation algorithms as the proposed CNNs were able to accurately quantify the amount of grass in paddocks.