Predicting Key Grassland Characteristics from Hyperspectral Data (original) (raw)
Related papers
International Journal of Remote Sensing, 2015
The use of hyperspectral data to estimate forage nutrient content can be a challenging task, considering the multicollinearity problem, which is often caused by high data dimensionality. We predicted some variability in the concentration of limiting nutrients such as nitrogen (N), crude protein (CP), moisture, and non-digestible fibres that constrain the intake rate of herbivores. In situ hyperspectral reflectance measurements were performed at full canopy cover for C3 and C4 grass species in a montane grassland environment. The recorded spectra were resampled to 13 selected band centres of known absorption and/or reflectance features, WorldView-2 band settings, and to 10 nm-wide bandwidths across the 400-2500 nm optical region. The predictive accuracy of the resultant wavebands was assessed using partial least squares regression (PLSR) and an accompanying variable importance (VIP) projection. The results indicated that prediction accuracies ranging from 66% to 32% of the variance in N, CP, moisture, and fibre concentrations can be achieved using the spectral-only information. The red, red-edge, and shortwave infrared (SWIR) wavelength regions were the most sensitive to all nutrient variables, with higher VIP values. Moreover, the PLSR model constructed based on spectra resampled around the 13 preselected band centres yielded the highest sensitivity to the predicted nutrient variables. The results of this study thus suggest that the use of the spectral resampling technique that uses only a few but strategically selected band centres of known absorption or reflectance features is sufficient for forage nutrient estimation.
2010
In this study, we investigated the opportunities of hyperspectral based remote sensing to assess the biomass and nitrogen status of temperate natural grasslands with a high vegetation coverage. Spectral and agronomic measurements were collected for 40 locations in three grasslands under different management regimes (grazing, mowing and organic fertilization) in the summer of 2008. Grassland reflectance in the range of 400-2500 nm was acquired from both fieldradiometry and a hyperspectral dataset acquired with the HyMap sensor. Reflectance-based vegetation indices in combination with linear regression analysis and partial least squares (PLS) regression were used to predict canopy biomass (CBM, kg/m2), canopy water content (CWC, kg/m2), and nitrogen concentration (NC, %). Results were evaluated using the coefficient of determination (R 2) and root mean squared error (RMSE) in a cross-validation approach. Red-edge based indices yielded the best performance for CBM, while for CWC the water index and the first derivative over 940-950 nm gave the best result. Best model performances for NC were obtained using PLS regression. Resulting PLS regression vectors indicated that next to the region around the red-edge also the short wave infrared around 1600 nm was important for characterization of canopy N. Finally, the vegetation index models with the highest accuracy were used to prepare continuous maps for CBM, CWC, and NC for the classified natural grasslands in the HyMap image. The results of this study show that a broad spectral range with contiguous spectral bands will be required for the biochemical assessment of natural grasslands. Satellite based hyperspectral sensors are therefore an important next step for the assessment of grassland ecosystems at the landscape level.
Hyperspectral remote sensing is increasingly being recognized as a powerful tool to map ecosystem properties and functions through time and space. However, general information on the accuracy of this technology to assess the vegetation's biophysical and-chemical trait composition, and on the variables which are mediating this accuracy, is often lacking so far. Here, we addressed this knowledge gap for grass-and shrubland ecosystems and applied novel three-level meta-analytical regression equations to 77 studies that validated hyperspectral remote sensing data with field observations. Our results showed that the accuracy of hyperspectral sensors is generally high, but strongly depends on the trait being studied (leaf area index: R 2 = 0.79 and nRMSE = 0.19, chlorophyll: R 2 = 0.77 and nRMSE = 0.21, carotenoids: R 2 = 0.80 and nRMSE = 0.29, phosphorus: R 2 = 0.75 and nRMSE = 0.14, nitrogen: R 2 = 0.74 and nRMSE = 0.09, water: R 2 = 0.69 and nRMSE = 0.13, and lignin content: R 2 = 0.64 and nRMSE = 0.26). Moreover, they indicated that the use of multivariate signal processing techniques could improve these estimation accuracies (adjusted p < 0.06 for LAI, chlorophyll and nitrogen). Finally, estimations from air-and spaceborne imaging spectrometers, allowing for functional mapping at broader spatial scales, were found to be as accurate as estimations from ground-based spectral measurements. Despite these promising findings, we revealed that leaf morphological properties (e.g. specific leaf area and leaf dry matter content) and biochemical traits which are not growth-related (e.g. lignin and cellulose) remain under-explored in grass-and shrublands. Moreover there was a strong publication bias towards R 2 for assessing model performance. Our findings foster and direct further methodological and technological developments for a more accurate and complete functional characterization of these ecosystems worldwide.
American Journal of Plant Sciences, 2014
Hyperspectral sensors provide the potential for direct estimation of pasture feed quality attributes. However, remote sensing retrieval of digestibility and fibre (lignin and cellulose) content of vegetation has proven to be challenging since tissue optical properties may not be propagated to the canopy level in mixed cover types. In this study, partial least squares regression on spectra from HyMap and Hyperion imagery were used to construct predictive models for estimation of crude protein, digestibility, lignin and cellulose concentration in temperate pastures. HyMap and Hyperion imagery and field spectra were collected over four pasture sites in southern Victoria, Australia. Co-incident field samples were analyzed with wet chemistry methods for crude protein, lignin and cellulose concentration, and digestibility was calculated from fiber determinations. Spectral data were subset based on sites and time of year of collection. Reflectance spectra were extracted from the hyperspectral imagery and collated for analysis. Six different transformations including derivatives and continuum removal were applied to the spectra to enhance absorption features sensitive to the quality attributes. The transformed reflectance spectra were then subjected to partial least squares regression, with full cross-validation "leave-one-out" technique, against the quality attributes to assess effects of the spectral transformations and post-atmospheric smoothing techniques to construct predictive models. Model performance between spectrometers, subsets and attributes were assessed using a coefficient of variation (CV),-the interquantile (IQ) range of the attribute values divided by the root mean square error of prediction (RMSEP) from the models. The predictive models with the highest CVs were obtained for digesti-S. Thulin et al. 998 bility for all spectra types, with HyMap the highest. However, models with slightly lower CVs were obtained for crude protein, lignin and cellulose. The spectral regions for diagnostic wavelengths fell within the chlorophyll well, red edge, and 2000-2300 nm ligno-cellulose-protein regions, with some wavelengths selected between the 1600 and 1800 nm region sensitive to nitrogen, protein, lignin and cellulose. The digestibility models with the highest CV's had confidence intervals corresponding to ±5% digestibility, which constitutes approximately 30% of the measured range. The cellulose and lignin models with the highest CV's also had similar confidence intervals but the slopes of the prediction lines were substantially less than 1:1 indicating reduced sensitivity. The predictive relationships established here could be applied to categorizing pasture quality into range classes and to determine whether pastures are above or below for example threshold values for livestock productivity benchmarks.
International Journal of Applied Earth Observation and Geoinformation, 2007
The main objective was to determine whether partial least squares (PLS) regression improves grass/herb biomass estimation when compared with hyperspectral indices, that is normalised difference vegetation index (NDVI) and red-edge position (REP). To achieve this objective, fresh green grass/herb biomass and airborne images (HyMap) were collected in the Majella National Park, Italy in the summer of 2005. The predictive performances of hyperspectral indices and PLS regression models were then determined and compared using calibration (n = 30) and test (n = 12) data sets. The regression model derived from NDVI computed from bands at 740 and 771 nm produced a lower standard error of prediction (SEP = 264 g m−2) on the test data compared with the standard NDVI involving bands at 665 and 801 nm (SEP = 331 g m−2), but comparable results with REPs determined by various methods (SEP = 261 to 295 g m−2). PLS regression models based on original, derivative and continuum-removed spectra produced lower prediction errors (SEP = 149 to 256 g m−2) compared with NDVI and REP models. The lowest prediction error (SEP = 149 g m−2, 19% of mean) was obtained with PLS regression involving continuum-removed bands. In conclusion, PLS regression based on airborne hyperspectral imagery provides a better alternative to univariate regression involving hyperspectral indices for grass/herb biomass estimation in the Majella National Park.
LAI and chlorophyll estimation for a heterogeneous grassland using hyperspectral measurements
ISPRS Journal of Photogrammetry and Remote Sensing, 2008
The study shows that leaf area index (LAI), leaf chlorophyll content (LCC) and canopy chlorophyll content (CCC) can be mapped in a heterogeneous Mediterranean grassland from canopy spectral reflectance measurements. Canopy spectral measurements were made in the field using a GER 3700 spectroradiometer, along with concomitant in situ measurements of LAI and LCC. We tested the utility of univariate techniques involving narrow band vegetation indices and the red edge inflection point, as well as multivariate calibration techniques, including stepwise multiple linear regression and partial least squares regression. Among the various investigated models, CCC was estimated with the highest accuracy (R 2 cv = 0.74, nRMSE cv = 0.35). All methods failed to estimate LCC (R 2 cv ≤ 0.40), while LAI was estimated with intermediate accuracy (R 2 cv values ranged from 0.49 to 0.69). Compared with narrow band indices and red edge inflection point, stepwise multiple linear regression generally improved the estimation of LAI. The estimations were further improved when partial least squares regression was used. When a subset of wavelengths was analyzed, it was found that partial least squares regression had reduced the error in the retrieved parameters. The results of the study highlight the significance of multivariate techniques, such as partial least squares regression, rather than univariate methods such as vegetation indices in estimating heterogeneous grass canopy characteristics.
Yield Estimates by a Two-Step Approach Using Hyperspectral Methods in Grasslands at High Latitudes
Remote Sensing
Ruminant fodder production in agricultural lands in latitudes above the Arctic Circle is constrained by short and hectic growing seasons with a 24-hour photoperiod and low growth temperatures. The use of remote sensing to measure crop production at high latitudes is hindered by intrinsic challenges, such as a low sun elevation angle and a coastal climate with high humidity, which influences the spectral signatures of the sampled vegetation. We used a portable spectrometer (ASD FieldSpec 3) to assess spectra of grass crops and found that when applying multivariate models to the hyperspectral datasets, results show significant predictability of yields (R2 > 0.55, root mean squared error (RMSE) < 180), even when captured under sub-optimal conditions. These results are consistent both in the full spectral range of the spectrometer (350–2500 nm) and in the 350–900 nm spectral range, which is a region more robust against air moisture. Sentinel-2A simulations resulted in moderately r...
Using Proximal Hyperspectral Sensing to Predict Herbage Nutritive Value for Dairy Farming
Agronomy
This study focuses on calibrating and validating models for hyperspectral canopy reflectance data that are useful to predict the nutritive value of ryegrass-white clover mixed herbage available to the grazing cow. Hyperspectral measurements and herbage cuts were collected from 286 sampling plots from a dairy farm from July 2017 to May 2018. Hyperspectral data were pre-treated by applying a Savitzky-Golay filter followed by a Gap-segment derivative algorithm. Herbage samples were analyzed for determination of herbage nutritive value traits, digestible organic matter in dry matter (DOMD), metabolizable energy (ME), crude protein (CP), neutral detergent fiber (NDF) and acid detergent fiber (ADF). Partial least squares regression was performed to calibrate the spectra against the five nutritive value traits. Results indicate that accuracy was moderately high for the CP model (R2 = 0.78) and moderate for the DOMD, ME, NDF and ADF models (0.54 < R2 < 0.67). The possibility of being ...
Giscience & Remote Sensing, 2016
The challenge of assessing and monitoring the influence of rangeland management practices on grassland productivity has been hampered in southern Africa, due to the lack of cheap earth observation facilities. This study, therefore, sought to evaluate the capability of the newly launched Sentinel 2 multispectral imager (MSI) data, in relation to Hyperspectral infrared imager (HyspIRI) data in estimating grass biomass subjected to different management practices, namely, burning, mowing and fertilizer application. Using sparse partial least squares regression (SPLSR), results showed that HyspIRI data exhibited slightly higher grass biomass estimation accuracies (RMSE = 6.65 g/m 2 , R 2 = 0.69) than Sentinel 2 MSI (RMSE = 6.79 g/m 2 , R 2 = 0.58) across all rangeland management practices. Student t-test results then showed that Sentinel 2 MSI exhibited a comparable performance to HyspIRI in estimating the biomass of grasslands under burning, mowing and fertilizer application. In comparing the RMSEs derived using wave bands and vegetation indices of HyspIRI and Sentinel, no statistically significant differences were exhibited (α = 0.05). Sentinel (Bands 5, 6 and 7) and HyspIRI (Bands 730 nm, 740 nm, 750 nm, 710 nm), as well as their derived vegetation indices, yielded the highest predictive accuracies. These findings illustrate that the accuracy of Sentinel 2 MSI data in estimating grass biomass is acceptable when compared with HyspIRI. The findings of this work provide an insight into the prospects of large-scale grass biomass modeling and prediction, using cheap and readily available multispectral data.