The objective of this research was to evaluate the accuracy of remote sensing for detecting weed infestation levels during early-season Glycine max production. Weed population estimates were collected from two G. max fields approximately 8 wk after planting during summer 1998. Seedling weed populations were sampled using a regular grid coordinate system on a 10- by 10-m grid. Two days later, multispectral digital images of the fields were recorded. Generally, infestations of Senna obtusifolia, Ipomoea lacunosa, and Solanum carolinense could be detected with remote sensing with at least 75% accuracy. Threshold populations of 10 or more S. obtusifolia or I. lacunosa plants m−2 were generally classified with at least 85% accuracy. Discriminant analysis functions formed for detecting weed populations in one field were at least 73% accurate in identifying S. obtusifolia and I. lacunosa infestations in independently collected data from another field. Due to highly variable soil conditions and their effects on the reflectance properties of the surrounding soil and vegetation, accurate classification of weed-free areas was generally much lower. Current remote sensing technology has potential for in-season weed detection; however, further advancements of the technology are needed to insure its use in future prescription weed management systems.

Weed population estimates were collected from four Glycine max fields during the summers of 1997 and 1998. Seedling weed populations were sampled using a regular coordinate system on a grid either 50 by 50, 30 by 30, or 10 by 10 m. MSU-HERB and Mississippi Herbicide Application Decision Support System (HADSS) (yield loss prediction and herbicide recommendation models for G. max) were used to determine the estimated net gain resulting from simulated herbicide applications at each sample location in each field. When necessary, the appropriate data points from the 10- by 10-m grid were removed to form population data sets on grids 20 by 20, 40 by 40, and 80 by 80 m. The objectives of this research were to compare estimated economic returns of site-specific herbicide management and broadcast herbicide management in nontransgenic and glyphosate-tolerant G. max and to evaluate the effects of various weed sampling intensities on estimated economic returns from site-specific herbicide applications. Site-specific herbicide management was the compilation of simulated herbicide treatments giving the highest estimated net gains at each location within each field. Broadcast herbicide management was the simulated broadcast application giving the highest estimated net gain for each field. Sampling costs and the unattainable site-specific application costs were not included in the estimated net gain calculations. In nontransgenic G. max production, the estimated net gain for treating the four fields with site-specific technology was $104.76 ha−1 higher than when using the optimum broadcast herbicide. In glyphosate-tolerant G. max production, the average estimated net gain for site-specific treatment of the fields was $96.24 ha−1 higher than for treatment with the best broadcast herbicide application. In nontransgenic G. max, the estimated net gain resulting from site-specific applications on a 10-m grid was $77.17 ha−1 higher than from site-specific applications on a 20-m grid; however, in glyphosate-tolerant G. max, this difference was only $19.84 ha−1. Increased estimated net gain resulted primarily from the use of herbicides that maximized return for each field area and from the decrease of unnecessary herbicide applications because of below-threshold weed infestations.

The objective of this research was to assess the accuracy of remote sensing for detecting weed species in soybean based on two primary criteria: the presence or absence of weeds and the identification of individual weed species. Treatments included weeds (giant foxtail and velvetleaf) grown in monoculture or interseeded with soybean, bare ground, and weed-free soybean. Aerial multispectral digital images were collected at or near soybean canopy closure from two field sites in 2001. Weedy pixels (1.3 m2) were separated from weed-free soybean and bare ground with no more than 11% error, depending on the site. However, the classification of weed species varied between sites. At one site, velvetleaf and giant foxtail were classified with no more than 17% error, when monoculture and interseeded plots were combined. However, classification errors were as high as 39% for velvetleaf and 17% for giant foxtail at the other site. Our results support the idea that remote sensing has potential for weed detection in soybean, particularly when weed management systems do not require differentiation among weed species. Additional research is needed to characterize the effect of weed density or cover and crop–weed phenology on classification accuracies.

The adoption of precision technologies that spatially register measurements using global positioning systems (GPS) greatly facilitates conducting large-scale on-farm research by farmers. On-farm experiments that utilize producer equipment include variations in agronomic practices that occur in situations where we want to predict the effect of inputs on yield. The domain of inference for such on-farm studies therefore more closely matches that desired by researchers. To investigate the feasibility of on-farm research using GPS, a study was conducted to evaluate the potential benefit of site-specific weed management. The study utilized producer-maintained field-scale equipment on four Montana farms in dryland spring wheat production. Paired site-specific and whole-field herbicide treatment areas were established in 0.9 to 1.9-ha blocks using consultant weed maps and a geographic information system (GIS). Yield was unaffected by herbicide treatment strategy (site-specific or broadcast). Minimal detectable yield differences were evaluated for the experimental design (0.2 T ha−1). Net returns increased when the percentage of field infested by wild oat decreased. Visual ratings of wild oat density taken at harvest indicated no difference in wild oat control between treatments in two of four site-years. This research suggests that producer-owned equipment can be used to compare treatments, but the accuracy and subsequent power of such comparisons are likely to be low.