Modelling nitrogen content of pasture herbage using thermal images and artificial neural networks

Abstract

The first step in improving nitrogen fertiliser application efficiency in pastoral farming is to develop a model to estimate pasture nitrogen content. This study relies on the principle that the higher the N content in pasture grass, the higher will be the light energy absorption for photosynthesis. The model will support the application of N depending on the pasture's real time temperature and other environmental parameters. Field experiments were carried out to investigate the correlation between N concentration in pasture leaves and their temperature. For this, the pasture's infrared image was taken using a thermal camera to measure leaf temperature. In addition, pasture was cut and processed into dry matter to analyse the % N content in the grass. During the study, air and soil temperatures, soil moisture, air velocity, humidity and illumination were also measured. Subsequently, a model of an artificial neural network (ANN), was produced to estimate N concentration in pasture measured at leaf temperatures and its environmental parameters. The ANN model was finalised by using four input parameters: herbage surface temperature, soil moisture, illumination (LUX) and air temperature. The model showed a ±0.32 %N prediction error for the training data, which increased slightly to ±0.37 for the validation data. The results showed that the ANN model was capable of estimating the N content of pasture with a variance between the training and validation data of 94% and 93%, respectively. The ANN model could also be used to develop pasture sward nutrient content assessment technologies, with implications for precision fertiliser placement as part of efficient N fertiliser application demands on farm.