Case Study
Disease in Trees
Using hyperspectral remote sensing to find ash dieback in Northamptonshire.
Overview
The aim of this campaign was to determine the capabilities of airborne hyperspectral imagery in the assessment of ash dieback (Hymenoscyphus fraxineus). The particular site of interest contained mature European ash (Fraxinus excelsior) situated in a roadside hedgerow environment located in Northamptonshire, United Kingdom.
To conduct the analysis for individual ash trees, tree crowns were automatically identified with an accuracy of 89%. The results of the campaign demonstrated the application of narrow-band vegetation indices for the classification of dieback presence in ash trees with an overall accuracy of 87%. The suppressed reflectance in the NIR for ash crowns affected by dieback was a key factor in this successful classification.
These results highlight the detection capabilities of airborne hyperspectral imagery for the dieback of individual ash trees in hedgerow environments.
Method
To conduct the analysis, 51 individual ash trees situated in a roadside hedgerow environment in Northamptonshire, United Kingdom (Figure 1), were surveyed to determine the presence and severity of crown dieback. Airborne hyperspectral and true-colour imagery were also acquired simultaneously for the study site with spatial resolutions of 0.33m and 0.07m respectively.
The hyperspectral imagery was processed to provide georectified surface reflectance. In addition, the high resolution true-colour imagery was subject to photogrammetric processing to provide a digital surface model (DSM) and digital terrain model (DTM).
The surface height information (Figure 1b) was subsequently applied to identify and isolate individual tree crowns in the hedgerow with an accuracy of 89%. Narrow-band vegetation indices calculated from the hyperspectral imagery were extracted for the automatically identified ash tree crowns for the dieback analysis.
Figure 1: Study site characteristics.
1a (from left to right): True-colour image from the study site. 1b (from right to left): Canopy height model for the study site created by the subtraction of the DTM from the DSM. Green polygons represent a sample of the ash trees surveyed at the study site.
Narrow-band vegetation indices calculated from the hyperspectral imagery using wavelengths from these dieback sensitive regions of the electromagnetic spectrum performed best in the dieback analysis. The classification of dieback presence (no dieback/dieback <5%) using these vegetation indices extracted from individual ash tree crowns automatically isolated in the hedgerow yielded an overall accuracy of 87%.
The capabilities of narrow-band vegetation indices were also evaluated for the classification of ash crown dieback severity. The three-class classification classified the crowns as healthy, light dieback (0-20%) and heavy dieback (20-50%) with an overall accuracy of 64%. Confusion in the classification was largely associated with the separation of ash crowns subject to light and heavy dieback. The normalised difference vegetation index (NDVI) which has commonly been applied to the assessment of vegetation condition was outperformed by of the narrow-band indices calculated from the hyperspectral imagery.
Figure 2: Median reflectance of ash tree crowns with and
without dieback..
Conclusion
The study has demonstrated the potential applications of airborne hyperspectral imagery for the detection of dieback in ash tree crowns. Individual trees crowns were identified from the hedgerow with an accuracy of 89%. Narrow-band vegetation indices calculated from hyperspectral imagery were applied to classify the presence of dieback with an overall accuracy of 87%. These indices were also applied classify the severity of ash crown dieback with an accuracy of 64%.
