Predicting in situ pasture quality in the Kruger National Park, South Africa, using continuum-removed absorption features

O. Mutanga*, A. K. Skidmore, H. H.T. Prins

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

243 Citations (Scopus)

Abstract

The remote sensing of pasture quality as determined by nitrogen, phosphorous, potassium, calcium and magnesium concentration is critical for a better understanding of wildlife and livestock feeding patterns. Although remote sensing techniques have proved useful for assessing the concentration of foliar biochemicals under controlled laboratory conditions, more investigation is required to assess their capabilities in the field, where inconsistent results have been obtained so far. We investigated the possibility of determining the concentration of in situ biochemicals in a savanna rangeland, using the spectral reflectance of five grass species. Canopy spectral measurements were taken in the field using a GER 3700 spectroradiometer. We tested the utility of using four variables derived from continuum-removed absorption features for predicting canopy nitrogen, phosphorus, potassium, calcium and magnesium concentration: (i) continuum-removed derivative reflectance (CRDR), (ii) band depth (BD), (iii) band depth ratio (BDR) and (iv) normalised band depth index (NBDI). Stepwise linear regression was used to select wavelengths from the absorption-feature-based variables. Univariate correlation analysis was also done between the first derivative reflectance and biochemicals. Using a training data set, the variables derived from continuum-removed absorption features could predict biochemicals with R 2 values ranging from 0.43 to 0.80. Results were highest using CRDR data, which yielded R2 values of 0.70, 0.80, 0.64, 0.50 and 0.68 with root mean square errors (RMSE) of 0.01, 0.004, 0.03, 0.01 and 0.004 for nitrogen, phosphorous, potassium, calcium and magnesium, respectively. Predicting biochemicals on a test data set, using regression models developed from a training data set, resulted in R2 values ranging from 0.15 to 0.70. The error of prediction (RSE) in the test data set was 0.08 (±10.25% of mean), 0.05 (±5.2% of mean), 0.02 (±11.11% of mean), 0.05 (±11.6% of mean) and 0.03 (±15% of mean) for nitrogen, potassium, phosphorous, calcium and magnesium, respectively, using CRDR. When data was partitioned into species groups, the R2 increased significantly to >0.80. With high-quality radiometric and geometric calibration of hyperspectral imagery, the techniques applied in this study (i.e. continuum removal on absorption features) may also be applied on data acquired by airborne and spaceborne imaging spectrometers to predict and ultimately to map the concentration of macronutrients in tropical rangelands.

Original languageEnglish
Pages (from-to)393-408
Number of pages16
JournalRemote Sensing of Environment
Volume89
Issue number3
DOIs
Publication statusPublished - 15 Feb 2004
Externally publishedYes

Keywords

  • absorption features
  • bootstrapping
  • continuum removal
  • field spectra
  • savanna rangelands

Fingerprint

Dive into the research topics of 'Predicting in situ pasture quality in the Kruger National Park, South Africa, using continuum-removed absorption features'. Together they form a unique fingerprint.

Cite this