Dissertation - Open Access
Doctor of Philosophy (PhD)
Agronomy, Horticulture, and Plant Science
Fusion, Landsat, Random Forest, South Dakota, Soybean, White Mold
Soybean is among the most important crops, cultivated primarily for beans, which are used for food, feed, and biofuel. According to FAO, the United States was the biggest soybeans producer in 2016. The main soybean producing regions in the United States are the Corn Belt and the lower Mississippi Valley. Despite its importance, soybean production is reduced by several diseases, among which Sclerotinia stem rot, also known as white mold, a fungal disease that is caused by the fungus Sclerotinia sclerotiorum is among the top 10 soybean diseases. The disease may attack several plants and considerably reduce yield. According to previous reports, environmental conditions corresponding to high yield potential are most conducive for white mold development. These conditions include cool temperature (12-24 °C), continued wet and moist conditions (70-120 h) generally resulting from rain, but the disease development requires the presence of a susceptible soybean variety. To better understand white mold development in the field, there is a need to investigate its spatiotemoral characteristics and provide accurate estimates of the damages that white mold may cause. Current and accurate data about white mold are scarce, especially at county or larger scale. Studies that explored the characteristics of white mold were generally field oriented and local in scale, and when the spectral characteristics were investigated, the authors used spectroradiometers that are not accessible to farmers and to the general public and are mostly used for experimental modeling. This study employed free remote sensing Landsat 8 images to quantify white mold in South Dakota. Images acquired in May and July were used to map the land cover and extract the soybean mask, while an image acquired in August was used to map and quantify white mold using the random forest algorithm. The land cover map was produced with an overall accuracy of 95% while white mold was mapped with an overall accuracy of 99%. White mold area estimates were respectively 132 km2, 88 km2, and 190 km2, representing 31%, 22% and 29% of the total soybean area for Marshall, Codington and Day counties. This study also explored the spatial characteristics of white mold in soybean fields and its impact on yield. The yield distribution exhibited a significant positive spatial autocorrelation (Moran’s I = 0.38, p-value < 0.001 for Moody field, Moran’s I = 0.45, p-value < 0.001, for Marshall field) as an evidence of clustering. Significant clusters could be observed in white mold areas (low-low clusters) or in healthy soybeans (high-high clusters). The yield loss caused by the worst white mold was estimated at 36% and 56% respectively for the Moody and the Marshall fields, with the most accurate loss estimation occurring between late August and early September. Finally, this study modeled the temporal evolution of white mold using a logistic regression analysis in which the white mold was modeled as a function of the NDVI. The model was successful, but further improved by the inclusion of the Day of the Year (DOY). The respective areas under the curves (AUC) were 0.95 for NDVI and 0.99 for NDVI+DOY models. A comparison of the NDVI temporal change between different sites showed that white mold temporal development was affected by the site location, which could be influenced by many local parameters such as the soil properties, the local elevation, management practices, or weather parameters. This study showed the importance of freely available remotely sensed satellite images in the estimation of crop disease areas and in the characterization of the spatial and temporal patterns of crop disease; this could help in timely disease damage assessment.
Number of Pages
South Dakota State University
In Copyright - Educational Use Permitted
Mfuka, Confiance L., "Characterizing Spatiotemporal Patterns of White Mold in Soybean across South Dakota Using Remote Sensing" (2019). Electronic Theses and Dissertations. 3396.