Modeling soil parameters using hyperspectral image reflectance in subtropical coastal wetlands

Developing spectral models of soil properties is an important frontier in remote sensing and soil science. Several studies have focused on modeling soil properties such as total pools of soil organic matter and carbon in bare soils. We extended this effort to model soil parameters in areas densely covered with coastal vegetation. Moreover, we investigated soil properties indicative of soil functions such as nutrient and organic matter turnover and storage. These properties include the partitioning of mineral and organic soil between particulate (>53 μm) and fine size classes, and the partitioning of soil carbon and nitrogen pools between stable and labile fractions. Soil samples were obtained from Avicennia germinans mangrove forest and Juncus roemerianus salt marsh plots on the west coast of central Florida. Spectra corresponding to field plot locations from Hyperion hyperspectral image were extracted and analyzed. The spectral information was regressed against the soil variables to determine the best single bands and optimal band combinations for the simple ratio (SR) and normalized difference index (NDI) indices. The regression analysis yielded levels of correlation for soil variables with R² values ranging from 0.21 to 0.47 for best individual bands, 0.28 to 0.81 for two-band indices, and 0.53 to 0.96 for partial least-squares (PLS) regressions for the Hyperion image data. Spectral models using Hyperion data adequately (RPD > 1.4) predicted particulate organic matter (POM), silt + clay, labile carbon (C), and labile nitrogen (N) (where RPD = ratio of standard deviation to root mean square error of cross-validation [RMSECV]). The SR (0.53 μm, 2.11 μm) model of labile N with R² = 0.81, RMSECV= 0.28, and RPD = 1.94 produced the best results in this study. Our results provide optimism that remote-sensing spectral models can successfully predict soil properties indicative of ecosystem nutrient and organic matter turnover and storage, and do so in areas with dense canopy cover.     

Predicting grain protein content of winter wheat using remote sensing data based on nitrogen status and water stress

Advanced site-specific knowledge of grain protein content of winter wheat from remote sensing data would provide opportunities to manage grain harvest differently, and to maximize output by adjusting input in fields. In this study, remote sensing data were utilized to predict grain protein content. Firstly, the leaf nitrogen content at winter wheat anthesis stage was proved to be significantly correlated with grain protein content (R² = 0.36), and spectral indices significantly correlated to leaf nitrogen content at anthesis stage were potential indicators for grain protein content. The vegetation index, VI_green, derived from the canopy spectral reflectance at green and red bands, was significantly correlated to the leaf nitrogen content at anthesis stage, and also highly significantly correlated to the final grain protein content (R² = 0.46). Secondly, the external conditions, such as irrigation, fertilization and temperature, had important influence on grain quality. Water stress at grain filling stage can increase grain protein content, and leaf water content is closely related to irrigation levels, therefore, the spectral indices correlated to leaf water content can be potential indicators for grain protein content. The spectral reflectance of TM channel 5 derived from canopy spectra or image data at grain filling stage was all significantly correlated to grain protein content (R² = 0.31 and 0.37, respectively). Finally, not only this study proved the feasibility of using remote sensing data to predict grain protein content, but it also provided a tentative prediction of the grain protein content in Beijing area using the reflectance image of TM channel 5.     

Improving the prediction of arsenic contents in agricultural soils by combining the reflectance spectroscopy of soils and rice plants

Visible and near-infrared reflectance spectroscopy provides a beneficial tool for investigating soil heavy metal contamination. This study aimed to investigate mechanisms of soil arsenic prediction using laboratory based soil and leaf spectra, compare the prediction of arsenic content using soil spectra with that using rice plant spectra, and determine whether the combination of both could improve the prediction of soil arsenic content. A total of 100 samples were collected and the reflectance spectra of soils and rice plants were measured using a FieldSpec3 portable spectroradiometer (350–2500 nm). After eliminating spectral outliers, the reflectance spectra were divided into calibration (n = 62) and validation (n = 32) data sets using the Kennard-Stone algorithm. Genetic algorithm (GA) was used to select useful spectral variables for soil arsenic prediction. Thereafter, the GA-selected spectral variables of the soil and leaf spectra were individually and jointly employed to calibrate the partial least squares regression (PLSR) models using the calibration data set. The regression models were validated and compared using independent validation data set. Furthermore, the correlation coefficients of soil arsenic against soil organic matter, leaf arsenic and leaf chlorophyll were calculated, and the important wavelengths for PLSR modeling were extracted. Results showed that arsenic prediction using the leaf spectra (coefficient of determination in validation, R_v² = 0.54; root mean square error in validation, RMSE_v = 12.99 mg kg⁻¹; and residual prediction deviation in validation, RPD_v = 1.35) was slightly better than using the soil spectra (R_v² = 0.42, RMSE_v = 13.35 mg kg⁻¹, and RPD_v = 1.31). However, results also showed that the combinational use of soil and leaf spectra resulted in higher arsenic prediction (R_v² = 0.63, RMSE_v = 11.94 mg kg⁻¹, RPD_v = 1.47) compared with either soil or leaf spectra alone. Soil spectral bands near 480, 600, 670, 810, 1980, 2050 and 2290 nm, leaf spectral bands near 700, 890 and 900 nm in PLSR models were important wavelengths for soil arsenic prediction. Moreover, soil arsenic showed significantly positive correlations with soil organic matter (r = 0.62, p < 0.01) and leaf arsenic (r = 0.77, p < 0.01), and a significantly negative correlation with leaf chlorophyll (r = −0.67, p < 0.01). The results showed that the prediction of arsenic contents using soil and leaf spectra may be based on their relationships with soil organic matter and leaf chlorophyll contents, respectively. Although RPD of 1.47 was below the recommended RPD of >2 for soil analysis, arsenic prediction in agricultural soils can be improved by combining the leaf and soil spectra.     

Retrieval of leaf water content spanning the visible to thermal infrared spectra

The objective of this study was to investigate the entire spectra (from visible to the thermal infrared; 0.390–14.0 μm) to retrieve leaf water content in a consistent manner. Narrow-band spectral indices (calculated from all possible two band combinations) and a partial least square regression (PLSR) were used to assess the strength of each spectral region. The coefficient of determination (R²) and root mean square error (RMSE) were used to report the prediction accuracy of spectral indices and PLSR models. In the visible-near infrared and shortwave infrared (VNIR–SWIR), the most accurate spectral index yielded R² of 0.89 and RMSE of 7.60%, whereas in the mid infrared (MIR) the highest R² was 0.93 and RMSE of 5.97%. Leaf water content was poorly predicted using two-band indices developed from the thermal infrared (R² = 0.33). The most accurate PLSR model resulted from MIR reflectance spectra (R² = 0.96, RMSE = 4.74% and RMSE cross validation RMSE_CV = 6.17%) followed by VNIR–SWIR reflectance spectra (R² = 0.91, RMSE = 6.90% and RMSE_CV = 7.32%). Using thermal infrared (TIR) spectra, the PLSR model yielded a moderate retrieval accuracy (R² = 0.67, RMSE = 13.27% and RMSE_CV = 16.39%). This study demonstrated that the mid infrared (MIR) and shortwave infrared (SWIR) domains were the most sensitive spectral region for the retrieval of leaf water content.     

A visible band index for remote sensing leaf chlorophyll content at the canopy scale

Leaf chlorophyll content is an important variable for agricultural remote sensing because of its close relationship to leaf nitrogen content. The triangular greenness index (TGI) was developed based on the area of a triangle surrounding the spectral features of chlorophyll with points at (670 nm, R₆₇₀), (550 nm, R₅₅₀), and (480 nm, R₄₈₀), where R_λ is the spectral reflectance at wavelengths of 670, 550 and 480, respectively. The equation is TGI = −0.5[(670 − 480)(R₆₇₀ − R₅₅₀) − (670 − 550)(R₆₇₀ − R₄₈₀)]. In 1999, investigators funded by NASA's Earth Observations Commercialization and Applications Program collaborated on a nitrogen fertilization experiment with irrigated maize in Nebraska. Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data and Landsat 5 Thematic Mapper (TM) data were acquired along with leaf chlorophyll meter and other data on three dates in July during late vegetative growth and early reproductive growth. TGI was consistently correlated with plot-averaged chlorophyll-meter values at the spectral resolutions of AVIRIS, Landsat TM, and digital cameras. Simulations using the Scattering by Arbitrarily Inclined Leaves (SAIL) canopy model indicate an interaction among TGI, leaf area index (LAI) and soil type at low crop LAI, whereas at high LAI and canopy closure, TGI was only affected by leaf chlorophyll content. Therefore, TGI may be the best spectral index to detect crop nitrogen requirements with low-cost digital cameras mounted on low-altitude airborne platforms.     

Spectral resampling based on user-defined inter-band correlation filter: C3 and C4 grass species classification

In this paper, a user-defined inter-band correlation filter function was used to resample hyperspectral data and thereby mitigate the problem of multicollinearity in classification analysis. The proposed resampling technique convolves the spectral dependence information between a chosen band-centre and its shorter and longer wavelength neighbours. Weighting threshold of inter-band correlation (WTC, Pearson's r) was calculated, whereby r = 1 at the band-centre. Various WTC (r = 0.99, r = 0.95 and r = 0.90) were assessed, and bands with coefficients beyond a chosen threshold were assigned r = 0. The resultant data were used in the random forest analysis to classify in situ C₃ and C₄ grass canopy reflectance. The respective WTC datasets yielded improved classification accuracies (kappa = 0.82, 0.79 and 0.76) with less correlated wavebands when compared to resampled Hyperion bands (kappa = 0.76). Overall, the results obtained from this study suggested that resampling of hyperspectral data should account for the spectral dependence information to improve overall classification accuracy as well as reducing the problem of multicollinearity.     

Detecting Sirex noctilio grey-attacked and lightning-struck pine trees using airborne hyperspectral data,random forest and support vector machines classifiers

The visual progression of sirex (Sirex noctilio) infestation symptoms has been categorized into three distinct infestation phases, namely the green, red and grey stages. The grey stage is the final stage which leads to almost complete defoliation resulting in dead standing trees or snags. Dead standing pine trees however, could also be due to the lightning damage. Hence, the objective of the present study was to distinguish amongst healthy, sirex grey-attacked and lightning-damaged pine trees using AISA Eagle hyperspectral data, random forest (RF) and support vector machines (SVM) classifiers. Our study also presents an opportunity to look at the possibility of separating amongst the previously mentioned pine trees damage classes and other landscape classes on the study area. The results of the present study revealed the robustness of the two machine learning classifiers with an overall accuracy of 74.50% (total disagreement = 26%) for RF and 73.50% (total disagreement = 27%) for SVM using all the remaining AISA Eagle spectral bands after removing the noisy ones. When the most useful spectral bands as measured by RF were exploited, the overall accuracy was considerably improved; 78% (total disagreement = 22%) for RF and 76.50% (total disagreement = 24%) for SVM. There was no significant difference between the performances of the two classifiers as demonstrated by the results of McNemar’s test (chi-squared; χ² = 0.14, and 0.03 when all the remaining ASIA Eagle wavebands, after removing the noisy ones and the most important wavebands were used, respectively). This study concludes that AISA Eagle data classified using RF and SVM algorithms provide relatively accurate information that is important to the forest industry for making informed decision regarding pine plantations health protocols.     

Plant phenolics and absorption features in vegetation reflectance spectra near 1.66 μm

Past laboratory and field studies have quantified phenolic substances in vegetative matter from reflectance measurements for understanding plant response to herbivores and insect predation. Past remote sensing studies on phenolics have evaluated crop quality and vegetation patterns caused by bedrock geology and associated variations in soil geochemistry. We examined spectra of pure phenolic compounds, common plant biochemical constituents, dry leaves, fresh leaves, and plant canopies for direct evidence of absorption features attributable to plant phenolics. Using spectral feature analysis with continuum removal, we observed that a narrow feature at 1.66 μm is persistent in spectra of manzanita, sumac, red maple, sugar maple, tea, and other species. This feature was consistent with absorption caused by aromatic CH bonds in the chemical structure of phenolic compounds and non-hydroxylated aromatics. Because of overlapping absorption by water, the feature was weaker in fresh leaf and canopy spectra compared to dry leaf measurements. Simple linear regressions of feature depth and feature area with polyphenol concentration in tea resulted in high correlations and low errors (% phenol by dry weight) at the dry leaf (r² = 0.95, RMSE = 1.0%, n = 56), fresh leaf (r² = 0.79, RMSE = 2.1%, n = 56), and canopy (r² = 0.78, RMSE = 1.0%, n = 13) levels of measurement. Spectra of leaves, needles, and canopies of big sagebrush and evergreens exhibited a weak absorption feature centered near 1.63 μm, short ward of the phenolic compounds, possibly consistent with terpenes. This study demonstrates that subtle variation in vegetation spectra in the shortwave infrared can directly indicate biochemical constituents and be used to quantify them. Phenolics are of lesser abundance compared to the major plant constituents but, nonetheless, have important plant functions and ecological significance. Additional research is needed to advance our understanding of the spectral influences of plant phenolics and terpenes relative to dominant leaf biochemistry (water, chlorophyll, protein/nitrogen, cellulose, and lignin).     

Estimating soil salinity in Pingluo County of China using QuickBird data and soil reflectance spectra

Soil salinization is a worldwide environmental problem with severe economic and social consequences. In this paper, estimating the soil salinity of Pingluo County, China by a partial least squares regression (PLSR) predictive model was carried out using QuickBird data and soil reflectance spectra. At first, a relationship between the sensitive bands of soil salinity acquired from measured reflectance spectra and the spectral coverage of seven commonly used optical sensors was analyzed. Secondly, the potentiality of QuickBird data in estimating soil salinity by analyzing the correlations between the measured reflectance spectra and reflectance spectra derived from QuickBird data and analyzing the contributions of each band of QuickBird data to soil salinity estimation Finally, a PLSR predictive model of soil salinity was developed using reflectance spectra from QuickBird data and eight spectral indices derived from QuickBird data. The results indicated that the sensitive bands covered several bands of each optical sensor and these sensors can be used for soil salinity estimation. The result of estimation model showed that an accurate prediction of soil salinity can be made based on the PLSR method (R² = 0.992, RMSE = 0.195). The PLSR model's performance was better than that of the stepwise multiple regression (SMR) method. The results also indicated that using spectral indices such as intensity within spectral bands (Int1, Int2), soil salinity indices (SI1, SI2, SI3), the brightness index (BI), the normalized difference vegetation index (NDVI) and the ratio vegetation index (RVI) as independent model variables can help to increase the accuracy of soil salinity mapping. The NDVI and RVI can help to reduce the influences of vegetation cover and soil moisture on prediction accuracy. The method developed in this paper can be applied in other arid and semi-arid areas, such as western China.     

Deriving leaf mass per area (LMA) from foliar reflectance across a variety of plant species using continuous wavelet analysis

Leaf mass per area (LMA), the ratio of leaf dry mass to leaf area, is a trait of central importance to the understanding of plant light capture and carbon gain. It can be estimated from leaf reflectance spectroscopy in the infrared region, by making use of information about the absorption features of dry matter. This study reports on the application of continuous wavelet analysis (CWA) to the estimation of LMA across a wide range of plant species. We compiled a large database of leaf reflectance spectra acquired within the framework of three independent measurement campaigns (ANGERS, LOPEX and PANAMA) and generated a simulated database using the PROSPECT leaf optical properties model. CWA was applied to the measured and simulated databases to extract wavelet features that correlate with LMA. These features were assessed in terms of predictive capability and robustness while transferring predictive models from the simulated database to the measured database. The assessment was also conducted with two existing spectral indices, namely the Normalized Dry Matter Index (NDMI) and the Normalized Difference index for LMA (ND_LMA).Five common wavelet features were determined from the two databases, which showed significant correlations with LMA (R²: 0.51–0.82, p < 0.0001). The best robustness (R² = 0.74, RMSE = 18.97 g/m² and Bias = 0.12 g/m²) was obtained using a combination of two low-scale features (1639 nm, scale 4) and (2133 nm, scale 5), the first being predominantly important. The transferability of the wavelet-based predictive model to the whole measured database was either better than or comparable to those based on spectral indices. Additionally, only the wavelet-based model showed consistent predictive capabilities among the three measured data sets. In comparison, the models based on spectral indices were sensitive to site-specific data sets. Integrating the ND_LMA spectral index and the two robust wavelet features improved the LMA prediction. One of the bands used by this spectral index, 1368 nm, was located in a strong atmospheric water absorption region and replacing it with the next available band (1340 nm) led to lower predictive accuracies. However, the two wavelet features were not affected by data quality in the atmospheric absorption regions and therefore showed potential for canopy-level investigations. The wavelet approach provides a different perspective into spectral responses to LMA variation than the traditional spectral indices and holds greater promise for implementation with airborne or spaceborne imaging spectroscopy data for mapping canopy foliar dry biomass.     

Multi range spectral feature fitting for hyperspectral imagery in extracting oilseed rape planting area

Spectral feature fitting (SFF) is a commonly used strategy for hyperspectral imagery analysis to discriminate ground targets. Compared to other image analysis techniques, SFF does not secure higher accuracy in extracting image information in all circumstances. Multi range spectral feature fitting (MRSFF) from ENVI software allows user to focus on those interesting spectral features to yield better performance. Thus spectral wavelength ranges and their corresponding weights must be determined. The purpose of this article is to demonstrate the performance of MRSFF in oilseed rape planting area extraction. A practical method for defining the weighted values, the variance coefficient weight method, was proposed to set up criterion. Oilseed rape field canopy spectra from the whole growth stage were collected prior to investigating its phenological varieties; oilseed rape endmember spectra were extracted from the Hyperion image as identifying samples to be used in analyzing the oilseed rape field. Wavelength range divisions were determined by the difference between field-measured spectra and image spectra, and image spectral variance coefficient weights for each wavelength range were calculated corresponding to field-measured spectra from the closest date. By using MRSFF, wavelength ranges were classified to characterize the target's spectral features without compromising spectral profile's entirety. The analysis was substantially successful in extracting oilseed rape planting areas (RMSE ≤ 0.06), and the RMSE histogram indicated a superior result compared to a conventional SFF. Accuracy assessment was based on the mapping result compared with spectral angle mapping (SAM) and the normalized difference vegetation index (NDVI). The MRSFF yielded a robust, convincible result and, therefore, may further the use of hyperspectral imagery in precision agriculture.     

Assessing the impact of hydrocarbon leakages on vegetation using reflectance spectroscopy

This paper assesses the capability of hyperspectral remote sensing to detect hydrocarbon leakages in pipelines using vegetation status as an indicator of contamination. A field experiment in real scale and in tropical weather was conducted in which Brachiaria brizantha H.S. pasture plants were grown over soils contaminated with small volumes of liquid hydrocarbons (HCs). The contaminations involved volumes of hydrocarbons that ranged between 2 L and 12.7 L of gasoline and diesel per m³ of soil, which were applied to the crop parcels over the course of 30 days. The leaf and canopy reflectance spectra of contaminated and control plants were acquired within 350–2500 nm wavelengths. The leaf and canopy reflectance spectra were mathematically transformed by means of first derivative (FD) and continuum removal (CR) techniques. Using principal component analysis (PCA), the spectral measurements could be grouped into either two or three contamination groups. Wavelengths in the red edge were found to contain the largest spectral differences between plants at distinct, evolving contamination stages. Wavelengths centred on water absorption bands were also important to differentiating contaminated from healthy plants. The red edge position of contaminated plants, calculated on the basis of FD spectra, shifted substantially to shorter wavelengths with increasing contamination, whereas non-contaminated plants displayed a red shift (in leaf spectra) or small blue shift (in canopy spectra). At leaf scale, contaminated plants were differentiated from healthy plants between 550–750 nm, 1380–1550 nm, 1850–2000 nm and 2006–2196 nm. At canopy scale, differences were substantial between 470–518 nm, 550–750 nm, 910–1081 nm, 1116–1284 nm, 1736–1786 nm, 2006–2196 nm and 2222–2378 nm. The results of this study suggests that remote sensing of B. brizantha H.S. at both leaf and canopy scales can be used as an indicator of gasoline and diesel contaminations for the detection of small leakages in pipelines.     

Remote quantification of phycocyanin in potable water sources through an adaptive model

Cyanobacterial blooms in water supply sources in both central Indiana USA (CIN) and South Australia (SA) are a cause of great concerns for toxin production and water quality deterioration. Remote sensing provides an effective approach for quick assessment of cyanobacteria through quantification of phycocyanin (PC) concentration. In total, 363 samples spanning a large variation of optically active constituents (OACs) in CIN and SA waters were collected during 24 field surveys. Concurrently, remote sensing reflectance spectra (R_rs) were measured. A partial least squares–artificial neural network (PLS–ANN) model, artificial neural network (ANN) and three-band model (TBM) were developed or tuned by relating the R_rs with PC concentration. Our results indicate that the PLS–ANN model outperformed the ANN and TBM with both the original spectra and simulated ESA/Sentinel-3/Ocean and Land Color Instrument (OLCI) and EO-1/Hyperion spectra. The PLS–ANN model resulted in a high coefficient of determination (R²) for CIN dataset (R² = 0.92, R: 0.3–220.7 μg/L) and SA (R² = 0.98, R: 0.2–13.2 μg/L). In comparison, the TBM model yielded an R² = 0.77 and 0.94 for the CIN and SA datasets, respectively; while the ANN obtained an intermediate modeling accuracy (CIN: R² = 0.86; SA: R² = 0.95). Applying the simulated OLCI and Hyperion aggregated datasets, the PLS–ANN model still achieved good performance (OLCI: R² = 0.84; Hyperion: R² = 0.90); the TBM also presented acceptable performance for PC estimations (OLCI: R² = 0.65, Hyperion: R² = 0.70). Based on the results, the PLS–ANN is an effective modeling approach for the quantification of PC in productive water supplies based on its effectiveness in solving the non-linearity of PC with other OACs. Furthermore, our investigation indicates that the ratio of inorganic suspended matter (ISM) to PC concentration has close relationship to modeling relative errors (CIN: R² = 0.81; SA: R² = 0.92), indicating that ISM concentration exert significant impact on PC estimation accuracy.     

Assessing plantation canopy condition from airborne imagery using spectral mixture analysis and fractional abundances

Pine plantations in Australia are subject to a range of abiotic and biotic damaging agents that affect tree health and productivity. In order to optimise management decisions, plantation managers require regular intelligence relating to the status and trends in the health and condition of trees within individual compartments. Remote sensing technology offers an alternative to traditional ground-based assessment of these plantations. Automated estimation of foliar crown health, especially in degraded crowns, can be difficult due to mixed pixels when there is low or fragmented vegetation cover. In this study we apply a linear spectral unmixing approach to high spatial resolution (50 cm) multispectral imagery to quantify the fractional abundances of the key image endmembers: sunlit canopy, shadow, and soil. A number of Pinus radiata tree crown attributes were modelled using multiple linear regression and endmember fraction images. We found high levels of significance (r² = 0.80) for the overall crown colour and colour of the crown leader (r² = 0.79) in tree crowns affected by the fungal pathogen Sphaeropsis sapinea, which produces both needle necrosis and chlorosis. Results for stands associated with defoliation and chlorosis through infestation by the aphid Essigella californica were lower with an r² = 0.33 for crown transparency and r² = 0.31 for proportion of crown affected. Similar analysis of data from a nitrogen deficient site produced an outcome somewhat in between the other two damaging agents. Overall the sunlit canopy image fraction has been the most important variable used in the modelling of forest condition for all damaging agents.     

Predicting Thaumastocoris peregrinus damage using narrow band normalized indices and hyperspectral indices using field spectra resampled to the Hyperion sensor

Thaumastocoris peregrinus (T. peregrinus) is a sap sucking insect that feeds on Eucalyptus leaves. It poses a threat to the forest industry by reducing the photosynthetic ability of the tree, resulting in stunted growth and even death of severely infested trees. Remote sensing techniques offer the potential to detect and map T. peregrinus infestations in plantation forests using current operational hyperspectral scanners. This study resampled field spectral data measured from a field spectrometer to the band settings of the Hyperion sensor in order to assess its potential in predicting T. peregrinus damage. Normalized indices based on NDVI ratios were calculated using the resampled visible and near-infrared bands of the Hyperion sensor to assess its utility in predicting T. peregrinus damage using Partial Least Squares (PLS) regression. The top 20 normalized indices were based on specific biochemical absorption features that predicted T. peregrinus damage with a mean bootstrapped R² value of 0.63 on an independent test dataset. The top 20 indices were located in the near-infrared region between 803.3 nm and 894.9 nm. Twenty three previously published hyperspectral indices which have been used to assess stress in vegetation were also used to predict T. peregrinus damage and resulted in a mean bootstrapped R² value of 0.59 on an independent test dataset. The datasets were combined to assess its collective strength in predicting T. peregrinus damage and significant indices were chosen based on variable importance scores (VIP) and were then entered into a PLS model. The indices chosen by VIP predicted T. peregrinus damage with a mean bootstrapped R² value of 0.71 on an independent test dataset. A greedy backward variable selection model was further tested on the VIP selected indices in order to find the best subset of indices with the best predictive accuracy. The greedy backward variable selection model identified 3 indices and performed the best by predicting damage with an R² value of 0.74 with the lowest RMSE of 1.30% on an independent test dataset. The best three indices identified include the anthocyanin reflectance index, carotenoid reflectance index and the normalized index calculated at 864.4 and 884.7 nm. Individual relationships between these indices and T. peregrinus damage indicate that high correlations are obtained with the inclusion of a few severely infested trees in the sample size. When the severely infested trees were removed from the study, the normalized index (864.4 and 884.7 nm) and the anthocyanin reflectance index still yielded significant correlations at the 99% confidence interval. This study indicates the significance of normalized indices and spectral indices calculated from the visible and near-infrared bands in hyperspectral data for the prediction of T. peregrinus damage.     

Estimate of heavy metals in soil and streams using combined geochemistry and field spectroscopy in Wan-sheng mining area,Chongqing, China

Heavy metals contaminated soils and water will become a major environmental issue in the mining areas. This paper intends to use field hyper-spectra to estimate the heavy metals in the soil and water in Wan-sheng mining area in Chongqing. With analyzing the spectra of soil and water, the spectral features deriving from the spectral of the soils and water can be found to build the models between these features and the contents of Al, Cu and Cr in the soil and water by using the Stepwise Multiple Linear Regression (SMLR). The spectral features of Al are: 480 nm, 500 nm, 565 nm, 610 nm, 680 nm, 750 nm, 1000 nm, 1430 nm, 1755 nm, 1887 nm, 1920 nm, 1950 nm, 2210 nm, 2260 nm; The spectral features of Cu are: 480 nm, 500 nm, 610 nm, 750 nm, 860 nm, 1300 nm, 1430 nm, 1920 nm, 2150 nm, 2260 nm; And the spectral features of Cr are: 480 nm, 500 nm, 610 nm, 715 nm, 750 nm, 860 nm, 1300 nm, 1430 nm, 1755 nm, 1920 nm, 1950 nm. With these features, the best models to estimate the heavy metals in the study area were built according to the maximal R². The R² of the models of estimating Al, Cu and Cr in the soil and water are 0.813, 0.638, 0.604 and 0.742, 0.584, 0.513 respectively. And the gradient maps of these three types of heavy metals’ concentrations can be created by using the Inverse distance weighted (IDW).The gradient maps indicate that the heavy metals in the soil have similar patterns, but in the North-west of the streams in the study area, the contents are of great differences. These results show that it is feasible to predict contaminated heavy metals in the soils and streams due to mining activities by using the rapid and cost-effective field spectroscopy.     

Assessment of leaf carotenoids content with a new carotenoid index: Development and validation on experimental and model data

Leaf carotenoids content (LCar) is an important indicator of plant physiological status. Accurate estimation of LCar provides valuable insight into early detection of stress in vegetation. With spectroscopy techniques, a semi-empirical approach based on spectral indices was extensively used for carotenoids content estimation. However, established spectral indices for carotenoids that generally rely on limited measured data, might lack predictive accuracy for carotenoids estimation in various species and at different growth stages. In this study, we propose a new carotenoid index (CARI) for LCar assessment based on a large synthetic dataset simulated from the leaf radiative transfer model PROSPECT-5, and evaluate its capability with both simulated data from PROSPECT-5 and 4SAIL and extensive experimental datasets: the ANGERS dataset and experimental data acquired in field experiments in China in 2004. Results show that CARI was the index most linearly correlated with carotenoids content at the leaf level using a synthetic dataset (R² = 0.943, RMSE = 1.196 μg/cm²), compared with published spectral indices. Cross-validation results with CARI using ANGERS data achieved quite an accurate estimation (R² = 0.545, RMSE = 3.413 μg/cm²), though the RBRI performed as the best index (R² = 0.727, RMSE = 2.640 μg/cm²). CARI also showed good accuracy (R² = 0.639, RMSE = 1.520 μg/cm²) for LCar assessment with leaf level field survey data, though PRI performed better (R² = 0.710, RMSE = 1.369 μg/cm²). Whereas RBRI, PRI and other assessed spectral indices showed a good performance for a given dataset, overall their estimation accuracy was not consistent across all datasets used in this study. Conversely CARI was more robust showing good results in all datasets. Further assessment of LCar with simulated and measured canopy reflectance data indicated that CARI might not be very sensitive to LCar changes at low leaf area index (LAI) value, and in these conditions soil moisture influenced the LCar retrieval accuracy.     

Detection of rice sheath blight for in-season disease management using multispectral remote sensing

Timely diagnosis of crop diseases in fields is critical for precision on-farm disease management. Remote sensing technology can be used as an effective and inexpensive method to identify diseased plants in a field scale. However, due to the diversity of crops and their associated diseases, application of the technology to agriculture is still in research stage, which needs to be elaborately investigated for algorithm development and standard image processing procedures. In this paper, we examined the applicability of broadband high spatial-resolution ADAR (Airborne Data Acquisition and Registration) remote sensing data to detect rice sheath blight and developed an approach to further explore the applicability. Based on the field symptom measurements, a comprehensive field disease index (DI) was constructed to measure infection severity of the disease and to relate to image sampled infections. In addition to direct band digital number (DN) values, band ratio indices and standard difference indices were used to examine possible correlations between field and image data. The results indicated that the broadband remote sensing imagery has the capability to detect the disease. Some image indices such as RI₁₄, SDI₁₄ and SDI₂₄ worked better than others. A correlation coefficient above 0.62 indicated that these indices would be valuable to use for identification of the rice disease. In the validation analysis, we obtained a small root mean square error (RMS = 9.1), confirming the applicability of the developed method. Although the results were encouraging, it was difficult to discriminate healthy plants from light infection ones when DI < 20 because of their spectral similarities. Hence, it was clear that identification accuracy increases when infection reaches medium-to-severe levels (DI > 35). This phenomenon illustrated that remote sensing images with higher spectral resolution (more bands and narrower bandwidth) were required in order to further examine the capability of separating the light diseased plants from healthy plants.     

Comparative analysis of different retrieval methods for mapping grassland leaf area index using airborne imaging spectroscopy

Fine scale maps of vegetation biophysical variables are useful status indicators for monitoring and managing national parks and endangered habitats. Here, we assess in a comparative way four different retrieval methods for estimating leaf area index (LAI) in grassland: two radiative transfer model (RTM) inversion methods (one based on look-up-tables (LUT) and one based on predictive equations) and two statistical modelling methods (one partly, the other entirely based on in situ data). For prediction, spectral data were used that had been acquired over Majella National Park in Italy by the airborne hyperspectral HyMap instrument. To assess the performance of the four investigated models, the normalized root mean squared error (nRMSE) and coefficient of determination (R²) between estimates and in situ LAI measurements are reported (n = 41). Using a jackknife approach, we also quantified the accuracy and robustness of empirical models as a function of the size of the available calibration data set. The results of the study demonstrate that the LUT-based RTM inversion yields higher accuracies for LAI estimation (R² = 0.91, nRMSE = 0.18) as compared to RTM inversions based on predictive equations (R² = 0.79, nRMSE = 0.38). The two statistical methods yield accuracies similar to the LUT method. However, as expected, the accuracy and robustness of the statistical models decrease when the size of the calibration database is reduced to fewer samples. The results of this study are of interest for the remote sensing community developing improved inversion schemes for spaceborne hyperspectral sensors applicable to different vegetation types. The examples provided in this paper may also serve as illustrations for the drawbacks and advantages of physical and empirical models.     

Determination of the factors governing soil erodibility using hyperspectral visible and near-infrared reflectance spectroscopy

Soil erodibility, which is difficult to estimate and upscaling, was determined in this study using multiple spectral models of soil properties (soil organic matter (SOM), water-stable aggregates (WSA) > 0.25 mm, the geometric mean radius (D_g)). Herein, the soil erodibility indicators were calculated, and soil properties were quantitatively analyzed based on laboratory simulation experiments involving two selected contrasting soils. In addition, continuous wavelet transformation was applied to the reflectance spectra (350–2500 nm) of 65 soil samples from the study area. To build the relationship, the soil properties that control erodibility were identified prior to the spectral analysis. In this study, the SOM, D_g and WSA >0.25 mm were selected to represent the most significant soil properties controlling erodibility and describe the erodibility indicator based on a logarithmic regression model as a function of SOM or WSA > 0.25 mm. Five, six and three wavelet features were observed to calibrate the estimated soil properties model, and the best performance was obtained with a combination feature regression model for SOM (R² = 0.86, p < 0.01), D_g (R² = 0.79, p < 0.01) and WSA >0.25 mm (R² = 0.61, p < 0.01), respectively. One part of the wavelet features captured amplitude variations in the broad shape of the reflectance spectra, and another part captured variations in the shape and depth of the soil dry substances. The wavelet features for the validated dataset used to predict the SOM, WSA >0.25 mm and D_g were not significantly different compared with the calibrated dataset. The synthesized spectral models of soil properties, and the formation of a new equation for soil erodibility transformed from the spectral models of soil properties are presented in this study. These results show that a spectral analytical approach can be applied to complex datasets and provide new insights into emerging dynamic variation with erodibility estimation.