Estimating crown defoliation and chemical constituents in the needles of Scots pine (Pinus sylvestris L.) trees by means of in-situ acquired hyperspectral data

Abstract

Methods of principal component and linear discriminant analysis were used to classify the needle samples into the defoliation classes and the partial least squares regression was used to predict the concentration of chemical constituents by means of hyperspectral reflectance data. Spectral reflectance data was found to poorly discriminate the samples of needles into the classes of defoliation assessed using 5% steps (kappa statistic was 0.29 and 0.26 for the last and current year needles respectively). However, combining the samples into 4 damage classes, according to UN-ECE/FAO definition (none – under 10 %; slight – 15 – 25 %, moderate - 30 – 60 % and severe – over 60%) the spectral reflectance data discrimination ability improved significantly for last year (kappa statistic 0.50) and not significantly for current year (kappa statistic 0.35) needles. Classification using 3 damage classes (under 30%; 35-50 % and over 50%) yielded precise classification accuracy (kappa statistic 1.0). Moderate prediction potential was found for the concentrations of nitrogen (correlation coefficient between actual and predicted values, estimated using the cross-validation, R=0.61), phosphorus (R=0.57), zinc (R=0.57), calcium (R=0.56), manganese (R=0.49), potassium (R=0.40) while it was poor for boron (R=0.33), iron (R=0.26), magnesium (R=0.20), and copper (R=0.20) in the current year needles. The study was carried out within the framework of the national project No VP1-3.1-ŠMM-08-K-01-025 entitled "Specific, genetic diversity and sustainable development of Scots pine forest to mitigate the negative effects of increased human pressure and climate change" supported by the EU Social Fund. The most successful studies in forest health assessment are those which involve using the potential of imaging spectrometry and concentrating on measuring of the spectral reflectance beyond the ability of human vision. Thus, we consider the hyperspectral imaging as a technique having a great potential for forest health assessment. This study is dealing with the estimation of crown defoliation and concentrations of some chemical constituents in the needles of Scots pine (Pinus sylvestris L.) trees by means of in-situ acquired hyperspectral data. Needle samples from 73 Scots pine trees having crown defoliation in the range from 5% to 80% (using 5% gradation) were collected in two mature stands located in eastern Lithuania. The concentrations of ten chemical elements in the needles were also measured. The hyperspectral reflectance data of the needles samples was recorded under laboratory conditions using Themis Vision Systems VNIR 400H portable scanning hyperspectral imaging camera operating in the 400-1000 nm range.