Hypersnectral Remote Sensing Of Plant Pigments And Nitrogen:a Meta-analysis

Pigments and nitrogen are of tremendous significance in plants. The photosynthetic pigments control the amount of solar radiation absorbed by leaf and thus determine photosynthetic potential and primary production. Nitrogen provides crucial support for plant photosynthesis and acts as a limiting resource in vegetative growth of plants. Given the demonstrable importance of plant pigments and nitrogen, studies concerning the temporal dynamics and spatial variations of pigments and nitrogen can provide key contributions to the environmental/agricultural management.With the advent of remote sensing technology, there have been opportunities to quantitatively describe plant biochemical concentrations non-destructively. A number of approaches have arisen recently to remotely estimate the pigment and nitrogen concentration. These studies produced variable results, but there are significant differences between the estimation precision and there is lack of agreement on optimal wavelengths for pigment and nitrogen detection. This is caused by the different species, experimental conditions and analytical methods used in these studies.In order to provide a comprehensive evaluation, meta-analysis has been applied to integrate literatures which reported the detection of pigment and nitrogen using remotely sensed data. For the selected articles, adjusted coefficient of determination and wavelengths were extracted for further analysis:(1) Identify optimal wavelengths for the individual pigment detectionHistogram and quantile plot were used to identify the central tendency of wavelength distribution. The results show that most of the wavelengths are distributed in the red edge, green peak and NIR regions. Tuning spectral bands in accord with spectral properties of pigment of interest, wavelengths in the three regions could be used to estimate multiple pigment contents.To find out the most frequent wavelength combination, the association rule mining methods were used. The optimal wavelengths for pigment detection at the leaf scale are shown below:550-560nm and700-710nm,550-560nm,700-710nm and780-790nm for anthocyanin detection;530-540nm and570-580nm for carotenoid detection;700-710nm and750-760nm,550-560nm and750-760nm for chlorophyll detection; wavelengths for chlorophyll a detection are the same with chlorophyll;630-640nm and800-810nm,650-660nm,670-680nm and700-710nm for chlorophyll b detection. The optimal wavelengths for pigment detection were not identified at the canopy and landscape scales due to the multiple environmental factors.(2) Evaluate the estimation precision of pigment concentration using remotely sensed dataThe highest number of relationships was published for detections at the leaf scale, the pigment detection at the canopy scale was less frequent reported in the literature and a few studies were conducted at the landscape scale. For each scale, the highest number of relationships was published for chlorophyll detection, followed by chlorophyll a, carotenoid, chlorophyll b and anthocyanin. The estimation precision of pigment concentration was higher at the leaf scale than those at the canopy and landscape scales.(3) Identify the central tendency of wavelength distribution for estimation of nitrogen concentrationMost of the wavelengths for nitrogen detection at the leaf scale are distributed in the550nm,700nm and2160-2180nm. For fresh leaves, wavelengths concentrated in the550nm and700nm; for dried and ground leaves, wavelengths concentrated in the2050nm and2180nm. There are significant differences between the wavelength distribution of fresh and dried leaves. At the canopy scale, most of the wavelengths for nitrogen detection are distributed in the400-900nm, which is similar to that of fresh leaves. At the landscape scale, the central tendency of wavelength distribution is not obvious.Wavelengths selected for nitrogen detection were linked to the known nitrogen absorption features. The results show that less and less wavelengths were close to known nitrogen absorption features from the leaf to landscape scale. Among these nitrogen absorption features,640-660nm is chlorophyll absorption feature;2060nm is protein absorption feature. Both of them were selected for nitrogen detection at the three scales, which imply universal models for nitrogen detection could be built using both visible and near-infrared absorption features.(4) Evaluate the impact of concentration range of nitrogen and types of leaf on estimation of nitrogen concentrationConcentration range of nitrogen is divided into three levels, namely, small, medium and large range of nitrogen concentration. Random effects model was used to compute the weighted mean of adjusted coefficient of determination for the three levels. The highest adjusted coefficient of determination was for medium range of nitrogen concentration, followed by large and small range of nitrogen concentration. From small to medium level, the value of adjusted coefficient of determination increases with increasing range of nitrogen concentration. This is because the estimation models with the larger range of nitrogen concentration will also possess the higher value of slope, resulting in larger adjusted coefficient of determination. The actual relationship between reflectance and nitrogen concentration was nonlinear, however, linear models were used by91%of the studies inclusion in meta-analysis. This has resulted in lower adjusted coefficient of determination for large range of nitrogen concentration. The adjusted coefficient of determination for fresh leaves was lower than that of dried and ground leaves due to the multiple influencing factors.