Proximal and remote detection of wheat infested by wheat stem sawfly (Cephus cinctus Norton) via multitemporal reflectance measurements

Abstract

For more than a century, the wheat stem sawfly (WSS, Cephus cinctus Norton) has been one of the most important insect pests of wheat (Triticum aestivum L.) production in North America. Effective use of management tactics is impaired by the difficulty in monitoring impacts on WSS. Extensive monitoring of WSS populations is cost prohibitive because comprehensive stem dissection surveys are required. The efficacy of surveys may be enhanced with remotely sensed (RS) data, yet little work has been conducted on this approach. We assessed the potential for using RS to estimate WSS infestation at the end of the growing season at the canopy and sub-field spatial scales. We investigated RS of WSS at the canopy level by experimentally infesting wheat plants with WSS. We collected weekly hyperspectral measurements to identify the spectral and temporal scales relevant to RS of WSS. We used sparse multiway partial least squares regression to model variation in multitemporal hyperspectral reflectance of wheat canopies as a function of WSS infestation. This approach accurately estimated the proportion of WSS infested stems (R2 = 0.68, RMSE = 13.5%) and identified spectral readings from the near-infrared shortwave infrared spectral regions collected from the entire experimental period as important. Building off these findings, we evaluated RS of WSS at the sub-field scale using multitemporal images from Sentinel 2 to estimate WSS infestation across 9 production wheat fields. From these fields, we dissected 43,155 wheat stems collected from 1,158 unique locations. For each field, we produced a model to estimate the proportions of total WSS infestation, adequate WSS infestation, and WSS-cut stems. We then compared the performance of these models and found that on average, models describing total WSS (R2 = 0.57) or adequate WSS infestation (R2 = 0.57) were more accurate than models of WSS cut (R2 = 0.34). Our findings suggest that multitemporal passive RS of WSS infestation at the sub-field scale can be useful for mapping patterns of total WSS infestation, but more work is required to effectively map WSS stem cutting. RS shows promise for supplementing research of infield management tactics to better manage WSS.