Development and examination of mobile sensor systems and software applications for use in estimation of forage dry matter biomass and crude protein

The use of field-based sensors can generate large amounts of data rapidly for phenomic modeling and management decisions; however some challenges may be encountered. AgriLogger software developed to rapidly acquire data for predictive model construction and implementation. AgriLogger features include user controls for data acquisition rate and a single output file for multiple sensors. Temporal and spatial data parsing was achieved from position and time stamps. Non-destructive biomass estimation of vegetation has been performed via remote sensing. This study examined several types of ground-based mobile sensing strategies for forage biomass estimation in alfalfa (Medicago sativa L.), bermudagrass [Cynodon dactylon (L.) Pers.], and wheat (Triticum aestivum L.). Forage quality analysis has historically been performed on physically collected samples through laboratory methods. Developing a sensor system which can collect data and provide estimates for crude protein (CP) in a more timely manner will allow near real time decision making by mangers. To evaluate the feasibility of such a system bermudagrass tall fescue (Festuca arundinacea Schreb.), and wheat were examined. AgriLogger reduced the post-processing time by a factor of 10 and data acquisition time by a factor of 60 as compared to commercially available alternatives which could be used for sensor data acquisition on vegetation. Predictive models were constructed via partial least squares regression and modeled estimates were compared to the physically measured biomass and CP. Differences between methods were minimal (average percent error of 11.2% for difference between predicted values versus machine and quadrat harvested biomass values (1.64 and 4.91 t ha-1, respectively). The predicted CP regressed with those measured in a laboratory using NIRS produced an R2 of 0.75 for a hyperspectral model. Wheat model prediction of crude protein bore n R2 of 0.65 and tall fescue R2=0.83. These data suggest that using mobile sensor-based biomass and CP estimation models could be an effective alternative to the traditional clipping and laboratory methods for rapid, accurate in-field estimation.