Predicting spatiotemporal patterns of productivity and grazing from multispectral data using neural network analysis based on system complexity

Authors: Ashworth, Amanda; AVILA, ANGELE - University Of Texas At Arlington; SMITH, HARRISON - University Of Arkansas; WINZELER, EDWIN - University Of Texas At Arlington; Owens, Phillip; Flynn, Kyle; O'Brien, Peter; PHILIPP, DIRK - University Of Arkansas; SU, JIANZHONG - University Of Texas At Arlington

Submitted to: Agrosystems, Geosciences & Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/19/2024
Publication Date: 9/13/2024
Citation: Ashworth, A.J., Avila, A., Smith, H., Winzeler, E., Owens, P.R., Flynn, K.C., O'Brien, P.L., Philipp, D., Su, J. 2024. Predicting spatiotemporal patterns of productivity and grazing from multispectral data using neural network analysis based on system complexity. Agrosystems, Geosciences & Environment. 7(3). Article e20571. https://doi.org/10.1002/agg2.20571.
DOI: https://doi.org/10.1002/agg2.20571

Interpretive Summary: Silvopasture systems integrate trees and livestock in the same piece of land, which diversifies production, improves soil carbon storage, and improves animal welfare. Satellites and remote sensing tools can be used for identifying cattle grazing preference, movement, and environmental interactions. In addition, recent advancements of unmanned aerial vehicles (UAV) and UAV-based sensors provide data at unprecedented spatial and temporal scales Therefore, scientists set out to use novel tools for predicting forage yield and nutritive value, soil nutrient status, and poultry litter response in efforts to improve pasture management and identify factors driving animal grazing preference in silvopasture and pasture-only systems. In this study, authors constructed a model using UAV-based data for predicting forage nutrition and herbage mass for real time pasture and resource management and planning which eliminates the need for sample collection. Study results also found that cattle preferred to graze the native grass mix over the introduced forage in both pasture-only and silvopastoral systems, as well as soils receiving poultry litter as a fertility source. Overall, cattle grazing followed distinct soil-landscape patterns, namely reduced cattle grazing preference occurred in areas of water accumulation, which highlights linkages among terrain features—nutrient movement—soil properties—forage nutrition—and animal grazing response over time and space. Results can be used for sustainable intensification to meet growing demands for environmentally responsible protein.

Technical Abstract: Remote sensing tools along with Global Navigation Satellite System cattle collars and digital soil maps may help elucidate spatial relationships among soils, terrain, forages, and animals temporally; while standard and often dated computational procedures preclude systems-level evaluations across this continuum. However, neural network analysis, a subset of machine learning, may elucidate efficiency of livestock production and linkages within the livestock-grazing environment. Therefore, this study applied deep learning to 1) develop predictive equations for yield and forage nutrition based on vegetation indices; and 2) at pixel-level, identify how grazing is linked to soil properties, forage growth and quality, and terrain attributes in silvopasture and pasture-only systems. Remotely sensed data were able to rapidly and non-destructively estimate herbage mass and quality for enhanced management of net and primary productivity in livestock and grazing systems. Cattle grazed the big bluestem (Andropogon gerardii Vitman) mix with 182% greater frequency than orchardgrass (Dactylis glomerata L.) in the pasture-only system. Real-time estimates of vegetative bands may assist in predicting grazing pressure over time and space for more efficient management of pasture resources. Cattle grazing followed distinct soil-landscape patterns, namely reduced cattle grazing preference occurred in areas of water accumulation, which highlights linkages among terrain features—soil-water movement—soil properties—forage nutrition—and animal grazing response spatially and temporally. Results from this study could be scaled to greater extents to improve grazing management among the largest land-use category in the U.S., grasslands, which would allow for sustainable intensification of forage-based livestock production to meet growing demands for environmentally responsible protein.