Establishing grassland yield models using Projection Pursuit Regression Method

Abstract

There is a great deal of interest in grassland yields as they are a required component in the calculation of carrying capacity, which is very important in grassland management practices. Traditional approaches to yield information are often time‐consuming, expensive, and limited in areal samplings. It is therefore desirable to develop new methods that can provide quick, easy, and cost‐effective estimates of grassland yields over large areas. In this study, an experiment was conducted at Fukang County, Xinjiang, China, to collect in situ data and remote sensing imagery. The in situ data included green herbaceous forage yields and weather information at four grassland types (plain desert, saline steppe, hill desert steppe, and mountain meadow), while the remote sensing images were acquired by the Landsat satellite TM sensors at 30‐m resolution. Analysis of this dataset resulted in the development of a yield model using the Projection Pursuit Regression (PPR) method. The PPR model was compared with traditional multivariate linear statistical methods and was found to be much more accurate in estimating grassland yields. When applied to the four grassland types, the PPR models resulted in an accuracy of 81.76, 88.61, 83.50, and 92.35% for plain desert, saline steppe, hill desert steppe, and mountain meadow, respectively. Therefore, the new PPR model has the potential to provide an effective method to estimate grassland yields over large geographic areas.

Keywords:

• grassland
• estimated grassland yield
• Projection Pursuit Regression Method
• PPR model
• assessment of grassland resources