Abstract

Managers of the nearly 0.5 million ha of public lands in North and South Dakota, USA rely heavily on manual measurements of vegetation properties to ensure conservation of grassland structure for wildlife and forage for livestock. Spectral imaging data may be useful in assessment of large (>100,000 ha) landscapes, as in the Grand River National Grassland (GRNG), South Dakota. Here, we examined the predictive potential for the Advanced High Resolution Spectrometer (AVIRIS) to estimate mixed-grass prairie canopy structural attributes (photosynthetically active vegetation (kg PV ha-1), non-photosynthetically active vegetation (kg NPV ha-1), total standing crop (kg PV+NPV ha-1), nitrogen content (kg N ha-1), and visual estimates of bare ground (%) in October 2010. We conducted the study on a 36,000-ha herbaceous area using 24 randomly selected plots divided into summit, midslope and toeslope positions. Field data were collected during the AVIRIS flyover, and three approaches for building a prediction model of canopy attributes based on spectra were evaluated based on R2 values. These approaches included Partial Least Squares Regression (PLS), a variable selection method with predictor variables based on functions of the AVIRIS spectra, and a variable selection method using individual bands or combinations of individual bands of spectra as predictors. All variable selection methods involved randomly partitioning the data into training and validations sets and choosing a final prediction model based on model selection frequency. PLS regression out-performed regression models (based on the variable selection methods) with R2 values of 0.73, 0.56, 0.62, 0.67, and 0.58, for PV, NPV, total standing crop, nitrogen content, and bare ground, respectively.

Creative Commons License


This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Share

COinS
 
Apr 29th, 3:40 PM

IDENTIFYING SPECTRA IMPORTANT FOR PREDICTION OF SENESCENT GRASSLAND CANOPY STRUCTURE

Managers of the nearly 0.5 million ha of public lands in North and South Dakota, USA rely heavily on manual measurements of vegetation properties to ensure conservation of grassland structure for wildlife and forage for livestock. Spectral imaging data may be useful in assessment of large (>100,000 ha) landscapes, as in the Grand River National Grassland (GRNG), South Dakota. Here, we examined the predictive potential for the Advanced High Resolution Spectrometer (AVIRIS) to estimate mixed-grass prairie canopy structural attributes (photosynthetically active vegetation (kg PV ha-1), non-photosynthetically active vegetation (kg NPV ha-1), total standing crop (kg PV+NPV ha-1), nitrogen content (kg N ha-1), and visual estimates of bare ground (%) in October 2010. We conducted the study on a 36,000-ha herbaceous area using 24 randomly selected plots divided into summit, midslope and toeslope positions. Field data were collected during the AVIRIS flyover, and three approaches for building a prediction model of canopy attributes based on spectra were evaluated based on R2 values. These approaches included Partial Least Squares Regression (PLS), a variable selection method with predictor variables based on functions of the AVIRIS spectra, and a variable selection method using individual bands or combinations of individual bands of spectra as predictors. All variable selection methods involved randomly partitioning the data into training and validations sets and choosing a final prediction model based on model selection frequency. PLS regression out-performed regression models (based on the variable selection methods) with R2 values of 0.73, 0.56, 0.62, 0.67, and 0.58, for PV, NPV, total standing crop, nitrogen content, and bare ground, respectively.