Combined use of spectral signatures and ultrasonic sward height for the assessment of biomass and quality parameters in heterogeneous pastures

Precision management of grasslands requires accurate information on sward characteristics at a spatial and temporal scale. Multi-sensor data fusion provides data for site-specific grassland management. The present study aimed to demonstrate the potential of using a combination of ultrasonic and hyper-spectral sensor data fusion to predict forage quality (crude protein and acid detergent fiber) and biomass in heterogeneous pastures. The ability of a mobile sensing system equipped with both sensors to quantify within-field variations was also evaluated. A field experiment with paddocks continuously stocked by cows and three levels of grazing intensity (moderate, lenient and very lenient) was used providing a broad range of sward characteristics for a sensor alignment. Different variables derived from hyper-spectral data including normalized difference spectral indices (NDSIs), multi-spectral satellite bands and PCA (principle component analysis) derived components were tested exclusively and combined with ultrasonic sward height (USH) to identify measurement options of high predictive capability and compared to modified partial least squares regression (MPLSR). Exclusive USH or spectral variables could hardly predict yield and quality parameters of heterogeneous pastures, whereas sensor data fusion by combining USH with narrow band NDSIs or WorldView2 satellite broad bands increased the prediction accuracy significantly, so that most calibration models exceeded an RPD value of 1.4 (residual predictive value), which is considered as an acceptable predicting capability for variable field condition. These combinations can be on par or even better than the use of the full hyperspectral information. Spectral regions related to water content were found to be most important for prediction of biomass while for estimating quality parameters both visible-near infrared regions were found to be important. The presence of a high proportion of senesced material in pastures influences the performance of the sensor systems and may limit the applicability of such concepts in the second half of the growing season. Prediction of biomass by mobile application of sensors explained > 63 % of the variation in manually determined reference plots representing the biomass range of each paddock. Accuracy of biomass prediction improved with increasing grazing intensity. Prediction accuracy with a mobile application of sensors was always lower than when sensors were applied statically. Differences between mobile and static measurements may be caused by position errors, which accounted for 8.5 cm on average. Even if future research is necessary to identify the limitations and improve sensor configurations, the present research of applying combined sensing technique on pasture canopies, could be seen as a step towards being able to measure and map pasture properties of interest in real time and at the field scale.