Design of a hyperspectral nitrogen sensing system for orange leaves

The orange (Citrus sinensis) is one of the most important agricultural crops in Florida. Heavy reliance on agricultural chemicals and low fertilizer use efficiencies in citrus production have raised environmental and economic concerns. In this study, a nitrogen sensor was developed to predict nitrogen concentrations in orange leaves. Four design criteria were chosen to maximize the sensing efficiency and reliability. They were: (1) coverage of the spectral N sensing range, (2) no moving parts, (3) single leaf detection, and (4) diffuse reflectance measurement. Based on chlorophyll and protein spectral absorption bands, the sensor's wavelength ranges were chosen to be 620–950 nm and 1400–2500 nm. A reflectance housing was designed to block environmental noise and to ensure single leaf measurement. A halogen light source, two detector arrays, two linear variable filters, and data acquisition cards with 16-bit analog-to-digital converters were used to collect data. The designed N sensor had a spectral resolution less than 30 nm. Test results showed that the nitrogen sensor had good linearity (r > 0.99) and stability. With averaged signal-to-noise ratio (SNR) of 299, the system was able to predict N content with a root mean square difference (RMSD) of l.69 g kg⁻¹ for the validation data set. Using the N sensor, unknown leaf samples could be classified into low, medium and high N levels with 70% accuracy.     

用MODIS数据预估森林可燃物湿度的研究

可燃物湿度(FMC）是决定森林火险和火行为的一个重要因子.为了快速有效地获得FMC,该文使用中分辨率成像光谱仪(MODIS)的短波红外(SWIR)和近红外(NIR)波段反射率、NIR和红光波段反射率估算了内蒙古乌素图林区的FMC,并用2004和2005年的实测数据进行了验证.结果表明,使用SWIR和NIR波段反射率能够更好地估算森林植被的FMC,其R2可以达到0.36.虽然该R2值比草地和灌木林地的要低,但是仍然能够提供有效的森林湿度的空间分布,在预测大面积的森林火灾行为时有一定的实用性.      

Continuum removal versus PLSR method for clay and calcium carbonate content estimation from laboratory and airborne hyperspectral measurements

Reflectance spectroscopy provides an alternate method to classical physical and chemical laboratory soil analysis for estimation of a large range of key soil properties. Techniques including classical chemometrics approaches and specific absorption features studies have been developed for deriving estimates of soil characteristics from visible and near-infrared (VNIR, 400-1200 nm) and shortwave infrared (SWIR, 1200-2500 nm) reflectance measurements. This paper examines the performances of two distinct methods for clay and calcium carbonate (CaCO₃) content estimation (two key soil properties for erosion prediction) by VNIR/SWIR spectroscopy: i) the Continuum Removal (CR) has been used to correlate spectral absorption bands centred at 2206 and 2341 nm with clay and CaCO₃ concentrations and ii) the partial least-squares regression (PLSR) method with leave-one-out cross-validation, which is a classical chemometrics technique, has been used to predict clay and CaCO₃ concentrations from VNIR/SWIR full spectra. We tried to respond to the question “should we use all bands in the 400-2500 nm range or should we focus our analysis on selected spectral absorption bands to determine soil properties from reflectance data?” In this paper, the CR and PLSR methods were applied to VNIR/SWIR laboratory and airborne HYMAP reflectance measurements collected over the La Peyne Valley area in southern France.This study shows that the performance of both techniques is dependent on the spectral feature for the soil property of interest and on the level data acquisition (lab or airborne) face to the instrument specifications. When airborne HYMAP reflectance measurements are used, the PLSR technique performs better than the CR approach. As well, when the soil property of interest has no well-identified spectral feature, which is the case of clay, the PLSR technique performs better than the CR approach. In this last situation, PLSR is able to find surrogate spectral features that retain satisfactory estimations of the studied soil properties. However, parts of these spectral features remain difficult to explain or relate to area-specific correlations between soil properties, which means that extrapolation to larger pedological contexts must be envisaged with care. In the near future, VNIR/SWIR airborne hyperspectral data processed by the PLSR technique will allow for accurate mapping of clay and CaCO₃ contents, which will contribute significantly to the digital mapping of soil properties.     

Quantification and depth distribution analysis of carbon to nitrogen ratio in forest soils using reflectance spectroscopy

Forest soils have large contents of carbon (C) and total nitrogen (TN), which have significant spatial variability laterally across landscapes and vertically with depth due to decomposition, erosion and leaching. Therefore, the ratio of C to TN contents (C:N), a crucial indicator of soil quality and health, is also different depending on soil horizon. These attributes can cost-effectively and rapidly be estimated using visible–near infrared–shortwave infrared (VNIR–SWIR) spectroscopy. Nevertheless, the effect of different soil layers, particularly over large scales of highly heterogeneous forest soils, on the performance of the technique has rarely been attempted. This study evaluated the potential of VNIR–SWIR spectroscopy in quantification and variability analysis of C:N in soils from different organic and mineral layers of forested sites of the Czech Republic. At each site, we collected samples from the litter (L), fragmented (F) and humus (H) organic layers, and from the A₁ (depth of 2–10 cm) and A₂ (depth of 10–40 cm) mineral layers providing a total of 2505 samples. Support vector machine regression (SVMR) was used to train the prediction models of the selected attributes at each individual soil layer and the merged layer (profile). We further produced the spatial distribution maps of C:N as the target attribute at each soil layer. Results showed that the prediction accuracy based on the profile spectral data was adequate for all attributes. Moreover, F was the most accurately predicted layer, regardless of the soil attribute. C:N models and maps in the organic layers performed well although in mineral layers, models were poor and maps were reliable only in areas with low and moderate C:N. On the other hand, the study indicated that reflectance spectra could efficiently predict and map organic layers of the forested sites. Although, in mineral layers, high values of C:N (≥ 50) were not detectable in the map created based on the reflectance spectra. In general, the study suggests that VNIR–SWIR spectroscopy has the feasibility of modelling and mapping C:N in soil organic horizons based on national spectral data in the forests of the Czech Republic.