Vegetation Ground-based Models in Crop State Monitoring

The development of efficient algorithms for multispectral and multitemporal data analysis is still one of the most essential issues of remote sensing. The importance of this issue is related to the ever-increasing quantity of data provided by numerous radiometric and imaging sensors. Besides, the necessity to use various geoinformation technologies incorporating remote sensing, in-situ observations, ancillary data etc., imposes information sharing and integration. This paper is devoted to ground-based spectral modeling as an integral part of vegetation remote sensing monitoring. It examines the relationship between agricultural crop spectral and biometrical features with consideration of growing conditions and plant ontogenesis. The influence of soil properties and anthropogenic factors (fertilization, heavy metal pollution) on crop spectral response has been studied. VIS and NIR ground-based reflectance measurements have been related to plant growth features to derive empirical models. Some results of crop state assessment using these models and airborne radiometric data are presented. Good agreement has been found between model estimates and ground-truth data. INTRODUCTION Nowadays the aerospace information gathered by different sensors and numerous Earth observation missions has become a genuine necessity in various investigations and application fields. Vegetation is among the priorities of these investigations which are related to many world significant problems such as environmental changes, anthropogenic impact on ecosystems, desertification processes, etc. In agriculture remote sensing is used for retrieving information about plant development and yield forecasting. Ground-based studies are an integral part of vegetation remote sensing technologies serving as a reference and verification source of remotely sensed data. Especially advantageous is the ability to vary and control experiment conditions getting a precise picture of plant spectral response to different factors as well as to track in detail temporal aspects of plant spectral properties during the ontogenetic process. A great number of papers is devoted to the possibility of deriving quantitative information about vegetation using reflective and emissive spectra. Many of them deal with plant growth evaluation, biomass estimation and yield prediction (i, ii, iii, iv, v). Empirical modelling is one of the most widely spread technique for vegetation assessment (i, vi, vii, viii, ix) although different conclusions have been made about the applicability of the obtained models, their dependence on local conditions and site-to-site or year-to year discrepancy (vi, x). This paper is further dedicated to spectral-biophysical modelling of agricultural vegetation. One of the objectives is to examine the impact of soil properties and anthropogenic factors (fertilization and heavy metal pollution) on plant spectral behaviour in relation to stress detection. The other goal is to test the applicability of spectral models for crop state assessment using airborne radiometric data. MATERIALS AND METHODS Reflectance, biometrical and phenological data were gathered from cereals throughout the growing season. A spring barley green-house experiment was conducted which consisted of two parts: NH4NO3 fertilization treatments over chernozem soil with different nitrogen concentrations (from 0 to 1000 mg/kg) and treatments with Ca(NO3)2 and KNO3 fertilizers for the nitrogen concentration of © EARSeL and Warsaw University, Warsaw 2005. Proceedings of 4th EARSeL Workshop on Imaging Spectroscopy. New quality in environmental studies. Zagajewski B., Sobczak M., Wrzesień M., (eds) 800 mg/kg, and a second part of Ni-polluted plots (with 100, 200, 300 and 400 mg/kg Ni concentration) grown over chernozem soil (pH=7.0-7.5) and grey forest soil (pH=5.0-5.5). The soils were chosen for two reasons their different reflectance spectra and different response to heavy metal pollution. A field experiment was carried out over winter wheat crops which state varied due to different growth conditions irrigation and nutrition regimen, soil erosion. Ground-based VIS and NIR spectral measurements were performed with a multichannel portable spectrometer from the nadir position in the wavelength range 0.4 0.8 μm and a 10 nm bandpass. Reflectance data were acquired at weekly intervals during plant development from emergence till full maturity for the barley plots and during wheat main phenological stages. Biometrical sampling included plant canopy cover, total fresh and dried above-ground phytomass, LAI, leaf biomass, plant height, stem and ear number, grain yield after harvest. At the time of ground data acquisition airborne multispectral measurements were performed within several transects over each winter wheat field. The data sets were statistically analysed to examine the correlations and establish empirical dependences between plant reflectance spectra, growth features and productivity. A regression analysis was run on vegetation spectral indices using band ratios, contrasts, normalized differences as a routinely implemented data transformation (iii, xi, xii, xiii, xiv). The wavelengths selected correspond to absorptions and high reflectance bands of vegetation spectra in the green (550 nm), red (670 nm) and near infrared (800 nm) range. Spectral indices were chosen from those having the best statistical correlation with vegetation bioparameters, the obtained empirical regressions being significant at the 95% level of confidence. Special attention was paid to temporal aspects of plant spectral properties throughout the growing period. The temporal behaviour of vegetation indices was regarded as a function of plant ontogenesis and used as a crop diagnostic feature and yield predictor. Significant variations in plant state, and consequently in spectral performance, were found associated with the impact of soil properties and anthropogenic factors. RESULTS AND DISCUSSION Various combinations of spectral ratios were examined for their correlation with plant bioparameters, nutrition conditions and Ni contamination. Many of them demonstrated high R values from 0.8 to 0.97 (viii, xv). Variations in vegetation reflectance are most attributed to green coverage. This parameter is at the same time a primary indicator of crop state. In Fig. 1 the statistical relationships of NIR/R and R/(G+R+NIR) spectral indices with barley canopy cover at pre-heading stage are shown. The dependences were derived separately for the grey (1) and chernozem soil (2) plots. Soil-integrated regression curves increase estimation errors almost twice, the canopy cover of the brighter grey soil treatments being systematically underestimated and overestimated for the dark chernozem soil treatments. Various combinations of spectral ratios were examined for their correlation with plant bioparameters, nutrition conditions and Ni contamination. Many of them demonstrated high R values from 0.8 to 0.97 (ix, xv). Variations in vegetation reflectance are most attributed to green coverage. This parameter is at the same time a primary indicator of crop state. In Fig. 1 the statistical relationships of NIR/R and R/(G+R+NIR) spectral indices with barley canopy cover at pre-heading stage are shown. The dependences were derived separately for the grey (1) and chernozem soil (2) plots. Soil-integrated regression curves increase estimation errors almost twice, the canopy cover of the brighter grey soil treatments being systematically underestimated and overestimated for the dark chernozem soil treatments. Similar spectral models were developed for barley leaf area index. It is explicable considering the high correlation between the two bioparameters described in this phenological stage by the equation: LAI = −0.052+5.74×canopy cover (R=0.95).