Assessing Separability of UAS-Derived Vegetation Indices for Detecting Plant Stress Due to Iron Chlorosis

This study used an Unmanned Aircraft System (UAS) equipped with a 4-band multispectral sensor to survey and assess a small plot sorghum field to determine the level of plant stress caused by iron deficiency, otherwise called iron chlorosis. Iron chlorosis is a nutritional disorder that effects a variety of crops that are grown on high-pH, calcareous soils and greatly effects crop growth, and eventually yield. The aim of this project was to detect iron chlorosis, by segments within crop rows for iron chlorosis, and for levels of chlorosis. A series of field data collection methods were used to analyze stress levels, amount of chlorophyll, and grain yield, correlating them to aerial data for field control and precise crop examination. A total of 25 vegetation indices (VIs) were extracted from multispectral reflectance maps processed using Structure-from-Motion (SfM) photogrammetry to assess overall plant health. The SfM-derived reflectance maps were cropped to remove soil, and used to calculate a second set of VIs. Both soil-removed and soil-included VIs were compared to field-acquired chlorophyll, color, and yield data measurements to determine those most separable to segmenting chlorosis at various levels using the UAS imagery.

Data analysis included multiple methods, which supported the determination of iron chlorosis and provided both quantitative and qualitative analyses. Segments of the areas affected by iron chlorosis were georeferenced in the field and assessed by chlorosis 'level' for control. This data provided a basis for detecting areas of high, medium, and low stress via UAS imagery (levels of healthy vegetation were also georeferenced in comparison). With this data, the known plant segments were located directly on the imagery, assessed, and then used to extract VI values for stressed and non-stressed segments. Class separability of VIs were then ranked using a two-class (chlorosis and non-chlorosis) median absolute deviation metric to account for outliers and non-Gaussian distribution. Results showed that that the Near Green (NG) VI raked most separable for separating zones of iron chlorosis. Furthermore, the results showed that VIs that included a combination of red-edge and near-infrared bands performed best (top two out of the three top-ranked metrics) except for the NG metric. The highest-ranking VIs can be used in a classifier to automate iron chlorosis detection at plant level using UAS.

Key words: UAS, iron chlorosis, multispectral imaging, structure-from-motion, grain sorghum, vegetation indices