|Ph.D Student||Rud Ronit|
|Subject||Spatial Spectral Indicative Pixel of Salinity Effects in|
Leaves: A Case Study of Three Crops
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Maxim Shoshany|
|Dr. Victor Alchanatis|
|Full Thesis text - in Hebrew|
This research involves the use of hyperspectral remote sensing for Precision Agriculture. Monitoring the status of crops is critical for responding to stress situations. The use of reclaimed water and soil salination causes salinity stress in crop-growth and reduces their yield. It is essential to provide information on the location and intensity of the negative effect of salinity. There are numerous multi-spectral methods in this field but they assume homogeneity in the leaf response, which is far from the real situation. The primary hypothesis of this research refers to the existence of 'indicative pixels' which best represent the status of plants.
This research aims were to search for a spectral index which may allow detection of salinity effects at the leaf level, to assess the spatial patterns of salinity affected areas on the leaf surface, and to develop a methodology which will extract 'indicative pixels' and will facilitate estimation of salinity effects from spatial and spectral parameterizations of patterns of indicative pixels.
The methodology for detecting ‘indicative pixels’ of salinity effects in crops using hyperspectral image was tested in cauliflower, eggplant and kohlrabi, grown in media with sodium chloride (NaCl) concentrations between 0 and 150 mmol.
A new spectral index was proposed, the Green-Indigo-Ratio (GIR) that utilizes spectral bands of 554 nm and 436 nm. The use of autocorrelation analysis, applying Getis-Ord-Gi index over the GIR images, enabled the discovery of a pattern of polar-points with local extreme value (positive or negative). These points represent the ‘indicative pixels’. The new developed methodology was then applied to segment local extreme points and map the ratio between the percentage of positive and negative values resulted from the autocorrelation analysis.
It was found for the three crop types that were monitored during two seasons that high salinity stress is characterized by patches of higher proportion of positive values and low salinity stress by higher proportion of negative values.
The integrated spatial-spectral method was compared with other methods, such as vegetation indices, and was found to provide significantly better results.
The new methodology presented here may be useful in better understanding the biophysical processes responsible for the formation of such spatial patterns in plants.
Within the wider context of remote sensing, the concept of 'indicative pixels' may provide an improved analysis methodology, especially when there is high spatial heterogeneity and when it is difficult to extract useful information using approaches based on spatial averaging.