Requirements on spectral resolution of remote sensing data for crop stress detection

Hyperspectral data proved to be highly suitable to identify areas of crop growth anomalies resulting from stress impact (e.g., nitrogen deficiency, fungal infections etc.). Stress symptoms are changes in plant physiology, whose characteristics affect the spectral signature of crop canopies that are consequently detectable via spectral measurements. Typical stress-related spectral changes in plant canopy signatures do not cause narrow spectral features, but rather influence certain wavelength ranges in the VIS and NIR. Sensor-based crop stress detection has certain requirements on the minimum spectral resolution of sensor systems. However, the question of whether the plentitude of narrow spectral bands of hyperspectral sensors is needed for crop stress detection arises. To analyze on which spectral scales stress symptoms can be detected, HyMap data was stepwise spectrally resampled and used for spectral mixture analyses to estimate powdery mildew severity of wheat. Results from 7 spectrally resampled data sets were compared to in-field sampled stress severity data. Results showed that the highest spectral resolution is actually not necessary to detect stressed wheat areas. Even with spectral resolutions 3 times lower than the original data set, regression analysis of endmember fractions and disease severities showed a satisfying coefficient of determination of R² = 0.55.