Spatial and Temporal Monitoring of Pasture Ecological Quality: Sentinel-2-Based Estimation of Crude Protein and Neutral Detergent Fiber Contents

Abstract

Frequent, region-wide monitoring of changes in pasture quality due to human disturbances or climatic conditions is impossible by field measurements or traditional ecological surveying methods. Remote sensing imagery offers distinctive advantages for monitoring spatial and temporal patterns. The chemical parameters that are widely used as indicators of ecological quality are crude protein (CP) content and neutral detergent fiber (NDF) content. In this study, we investigated the relationship between CP, NDF, and reflectance in the visible–near-infrared–shortwave infrared (VIS–NIR–SWIR) spectral range, using field, laboratory measurements, and satellite imagery (Sentinel-2). Statistical models were developed using different calibration and validation data sample sets: (1) a mix of laboratory and field measurements (e.g., fresh and dry vegetation) and (2) random selection. In addition, we used three vegetation indices (Normalized Difference Vegetative Index (NDVI), Soil-adjusted Vegetation Index (SAVI) and Wide Dynamic Range Vegetation Index (WDRVI)) as proxies to CP and NDF estimation. The best models found for predicting CP and NDF contents were based on reflectance measurements (R2 = 0.71, RMSEP = 2.1% for CP; and R2 = 0.78, RMSEP = 5.5% for NDF). These models contained fresh and dry vegetation samples in calibration and validation data sets. Random sample selection in a model generated similar accuracy estimations. Our results also indicate that vegetation indices provide poor accuracy. Eight Sentinel-2 images (December 2015–April 2017) were examined in order to better understand the variability of vegetation quality over spatial and temporal scales. The spatial and temporal patterns of CP and NDF contents exhibit strong seasonal dependence, influenced by climatological (precipitation) and topographical (northern vs. southern hillslopes) conditions. The total CP/NDF content increases/decrease (respectively) from December to March, when the concentrations reach their maximum/minimum values, followed by a decline/incline that begins in April, reaching minimum values in July.

Predicting grain protein content of field-grown winter wheat with satellite images and partial least square algorithm

“…These varieties were available in the experimental counties. At harvest time, wheat grains were sampled by five-point sampling method in the field, and then brought back to the laboratory for wheat GPC determination [31].…”

Evaluation of Grass Quality under Different Soil Management Scenarios Using Remote Sensing Techniques

“…Higher nitrogen uptake by grass, which leads to increased chlorophyll content and grass growth, reduces the reflectance of visible spectra particularly in the red and green range of the spectrum [74]. The increased reflectance of NIR and red-edge spectra is also associated with the improvement of CP content and grass BM [37]. The importance of these wavelength ranges was also confirmed by identifying the most effective spectral indices for PLSR prediction of BM and CP using MSI-UAV.…”

“…The absorbance spectra at 2060 nm, 2130 nm, 2180 nm, and 2240 nm are associated with N-H and C-H bonds of protein [79,80]. An accurate estimation of forage attributes was also reported using the SWIR region of the spectrum by Pullanagari et al [81] and Lugassi et al [37]. Due to the low accuracy of models developed using MSI-Sentinel-2 and the spectral range of HSI images ( Table 2, 450-950 nm), the efficiency of SWIR for predicting CP could not be evaluated in this study.…”

“…MSI-Sentinel-2 images are available at 10 and 20 m resolution. The 20 m dataset was resampled to 10 m using the "special resampling" procedure using the ENVI software [36,37] in order to handle the spatial resolution disparity and enable both datasets to be combined. Twenty-one spectral indices which had been reported as typical vegetation indices for assessing vegetation quality parameters in the published literature were selected for spectral prediction of measured grass BM and CP (Table 3).…”

“…Although a reasonable level of accuracy was not obtained for predicting CP using MSI-Sentinel-2, band 11 and band 3 did have the maximum impact on spectral modelling of CP ( Figure 6, CPM-3). Lugassi et al [37] found band 3, band 4, band 8a, and band 12 as the most useful bands of Sentinel 2 for estimating protein content. The PLSR technique could reliably be used for spectral prediction of BM with an excellent accuracy using HSI and a good accuracy using UAV and Sentinel images.…”