Spectral assessment of senescent sourveld and sweetveld grass quality using continuum-removed absorption features from EnMAP spaceborne imaging spectrometer

• Senescent sourveld and sweetveld grass quality dynamics are poorly understood from a remote sensing perspective. • First study to explore spaceborne EnMAP hyperspectral data for estimating senescent grass quality constituents. • Study shows the effectiveness of applying continuum removal to absorption features in data acquired from the EnMAP for regional estimation of senescent grass quality. • Key wavelengths for senescent grass quality estimation in the different velds were identified. • Results are crucial for informing sustainable veld and grazing management strategies. Sourveld and sweetveld grasses exhibit interesting nutritional quality dynamics at the senescent stage, which have not been investigated from a remote sensing perspective. This study presents the first spatially explicit method(s) for estimating biochemical constituents of senescent grass across sour and sweet veld types using the spaceborne Hyperspectral Imager (HSI) on board the Environmental Mapping and Analysis Programme (EnMAP). Specifically, this study assessed the effectiveness of applying continuum removal to absorption features in data acquired from the EnMAP HSI. We also identified the key wavelengths relevant for senescent grass quality estimation in the different velds. Grass quality was quantified using three foliar biochemical indicators: nitrogen (N), acid detergent fibre (ADF) and acid detergent lignin (ADL). Data analysis was conducted using the random forest (RF) model and five broad continuum-removed absorption features which have been found to relate to foliar biochemicals in dry vegetation spectra. The study first evaluated the performance of four metrics derived from continuum-removed spectra, namely the continuum-removed derivative reflectance (CRDR), band depth (BD), band depth ratios (BDR) and the normalized band depth index (NBDI). The CRDR combined with RF regression was then used to identify key wavelengths from the absorption features for predicting senescent grass canopy N, ADF, and ADL across both veld types. Using the testing datasets, the selected wavebands yielded strong predictive performance for senescent sourveld, with R 2 values of 0.72, 0.71, and 0.69, RMSEs of 0.08, 6.96, and 3.68, and RMSE% (of mean) of 13.77, 12.64, and 16.38 for N, ADF, and ADL, respectively. Statistical agreement was highest in sweetveld grasses, yielding R 2 values of 0.73, 0.78, 0.82, corresponding RMSEs of 0.13, 5.82, and 2.01 while the RMSE% (of mean) were 10.95, 11.81 and 11.26 for N, ADF, and ADL, respectively. Overall, the R 550.69 – 741.83 absorption feature had the highest number of optimal wavelengths in estimating canopy N across both velds. Moreso, optimal wavebands for estimating fibre concentration of senescent grass were derived more from the SWIR with the R 2224.58 – 2376.95 absorption feature, yielding higher estimation accuracies. CRDR exhibited promising results, as it identified wavelengths causally linked to N, ADF and ADL in senescent vegetation. This work underscores the potential of spaceborne hyperspectral data with continuum removal on known absorption features for regional estimation of senescent grass quality.