Return to search

Principal Component Analysis and Spatial Regression Techniques to Model and Map Corn and Soybean Yield Variability with Radiometrically Calibrated Multitemporal and Multispectral Digital Aerial Imagery

<p> Remotely sensed data has been discussed as a possible alternative to the standard precision agriculture systems of combine-mounted yield monitors because of the burden, cost, end of season use, and inherent errors that are associated with these systems. Due to the potential quantitative use of remote sensing in precision agriculture, the primary focus of this study was to test the relationship between multitemporal/multispectral digital aerial imagery with corn (<i>Zea mays</i> L.) and soybean (<i>Glycine max </i> L.) yield. Digital aerial imagery was gathered on nine different dates throughout the 2015 growing season from two fields (one corn and one soybean) located on a farm in Story County, Iowa. To begin assessing this relationship, the digital aerial imagery was radiometrically calibrated. The radiometric calibration process used calibration tarps with known reflectance values (3, 6, 12, 22, 44, and 56 percent). The calibrated imagery was then used to calculate and output 12 different vegetation indices (VIs) and three calibrated wavebands (red, green, and near-infrared). </p><p> Next, the calibrated VIs and wavebands from the 2015 growing season were used to examine their relationship with the corn and soybean yield data collected from a combine yield monitor system. This relationship between multitemporal/multispectral digital aerial imagery with corn and soybean yield was investigated with principal component analysis and spatial modeling techniques. The results from spatial modeling of corn revealed that VIs utilizing the green waveband performed strongly. VIs such as, chlorophyll index-green, chlorophyll vegetation index, and green normalized difference vegetation index accounted for 81.6, 83.0, and 82.4 percent of the yield variability, respectively. Strong modeling relationships were also found in soybean using just the near-infrared waveband or VIs that utilized the near-infrared waveband. The near-infrared waveband captured 89.1 percent of the yield variation, while VIs such as, difference vegetation index, triangular vegetation index, soil adjusted vegetation index, and optimized soil adjusted vegetation index accounted for 87.3, 87.3, 83.9, and 83.8 percent of soybean yield variability, respectively. The temporal assessment of the remotely sensed data also identified certain VIs and wavebands that captured pivotal growth stages for detecting potential yield limiting factors. These specific growth stages varied for different VIs and wavebands for both corn and soybean. Overall, the results from this study identified that mid-to-late vegetative growth stages (prior to tasseling) and late-season reproductive stages were important parameters that provided unique information in the modeling of corn yield variability, while the later reproductive stages (just prior to senescence) were essential to capturing soybean yield variability. </p><p> Lastly, this research produced corn and soybean yield maps from the digital aerial imagery. The digital aerial imagery yield maps were then compared with maps that used kriging interpolation of the combine yield monitor data gathered from the same corn and soybean fields. The results indicated that both corn and soybean yield maps produced with multitemporal/multispectral digital aerial imagery were comparable with a standard method of kriging interpolation from yield monitor data.</p><p>

Identiferoai:union.ndltd.org:PROQUEST/oai:pqdtoai.proquest.com:10807753
Date08 June 2018
CreatorsPritsolas, Joshua
PublisherSouthern Illinois University at Edwardsville
Source SetsProQuest.com
LanguageEnglish
Detected LanguageEnglish
Typethesis

Page generated in 0.0019 seconds