By Adriana Niewiadomska
Summary
Remote sensing is a key tool for assessing vegetation condition over large areas. Several ecosystem processes i.e growth has been related to plant biochemical composition. Biomass is a key biophysical parameter for monitoring of natural vegetation succession which is essential for optimal management of river floodplains in the Netherlands and is often used for initialization and validation of ecological models.
Until now a lot of research has been done on application of a high spectral resolution remote sensing to estimate biochemical- (mostly nitrogen) and biophysical parameters of crops and monogamous vegetation and the results are promising. Nevertheless an estimation of the biochemical composition of natural vegetation from spectral data is more complicated due to high spatial variability of vegetation composition.
The scope of this study was to asses the predictive power of different Remote Sensing techniques for estimating vegetation biomass and biochemical content of heterogeneous vegetation.
The presented study has been carried out in the Millingerwaard floodplain in the Netherlands, which is a heterogeneous natural area along the river Waal in the Netherlands.
Two spectral datasets were used; 1) spectral reflectances collected by the HyMap sensor (airborne) and 2) in the field collected spectral reflectances using ASD FieldSpec Pro spectroradiometer.
The prediction capability for biomass of broad used vegetation indices like NDVI and WDVI were tested. The REP was tested as a predictor for biomass and biochemical content. New NDNI index was used for prediction of nitrogen. PLS regression was implemented as a feature reduction technique for prediction of all dependent variables.
To decrease heterogeneity within the dataset, the 21 data plots were divided into plant functional types; Because of the small sample size only PFT2 (n =17) was used for study.
The results show characteristic differences between both datasets: vegetation indices calculated from the HyMap spectral reflectances have better prediction power for biomass, while for estimation of the vegetation biochemical content, vegetation indices calculated using the Field Spec reflectances ended in better results.
In case of both datasets, a soil corrected WDVI index had the highest prediction power (R 2 = 0.46 and R 2 = 0.58 for FieldSpec and HyMap dataset respectively) for biomass. In case of vegetation nitrogen content, REP turn up to be the best predictor (R 2 = 0.34 ), while PLS regression performed the best while predicting P and K content (R 2 = 0.38 and 0.33 respectively). All those results are for the FieldSpec dataset.
In generally this study has show how difficult is the assessment of biophysical and biochemical characteristics of the vegetation from the hyperspectral data and that a lot of different aspects should be taken into account. Very important aspects to look at are i.e. vegetation heterogeneity, leaf area index, the biomass content etc.