Detection of anomalous crop condition and soil variability mapping using a 26 year Landsat record and the Palmer crop moisture index

- Define normal crop condition based on 26 years of Landsat data.
- Crop condition assessed through NDVI, soil moisture from crop moisture index (CCMI).
- Responses in depressions were anti-correlated, and crests correlated, with CCMI.
- Normal condition was defined by a tolerance interval from the sample distribution.
- Broader application for crop condition monitoring and edaphological soil mapping.

Cost-effective and reliable vegetation monitoring methods are needed for applications ranging from traditional agronomic mapping, to verifying the safety of geologic injection activities. A particular challenge is defining baseline crop conditions and subsequent anomalies from long term imagery records (Landsat) in the face of large spatiotemporal variability. We develop a new method for defining baseline crop response (near peak growth) using the normalized difference vegetation index (NDVI) from 26 years (1986-2011) of Landsat data for 400 km2 surrounding a planned geologic carbon sequestration site near Jacksonville, Illinois. The normal score transform (yNDVI) was applied on a field by field basis to accentuate spatial patterns and level differences due to planting times. We tested crop type and soil moisture (Palmer crop moisture index (CMI)) as predictors of expected crop condition. Spatial patterns in yNDVI were similar between corn and soybeans - the two major crops. Linear regressions between yNDVI and the cumulative CMI (CCMI) exposed complex interactions between crop condition, field location (topography and soils), and annual moisture. Wet toposequence positions (depressions) were negatively correlated to CCMI and dry positions (crests) positively correlated. However, only 21% of the landscape showed a statistically significant (p < 0.05) linear relationship. To map anomalous crop conditions, we defined a tolerance interval based on yNDVI statistics. Tested on an independent image (2013), 63 of 1483 possible fields showed unusual crop condition. While the method is not directly suitable for crop health assessment, the spatial patterns in correlation between yNDVI and CCMI have potential applications for pest damage detection and edaphological soil mapping, especially in the developing world.