Quantifying Vegetation Change by Point Sampling Landscape Photography Time Series

Abstract

Quantitative assessment of vegetation change is often conducted by digitally analyzing time series of aerial or vertical photography. Change analysis conducted using repeated oblique or landscape photography, however, has been limited to qualitative assessments. The purpose of this study was to develop sampling and analysis techniques for using a time series of digitized landscape photography to quantify vegetation change on rangeland landscapes. Digital images were created from black-and-white landscape photographs acquired in 1917, 1962, and 2000 near Whiskey Mountain in the Reynolds Creek Experimental Watershed of southwestern Idaho. Images were spatially registered to each other using control points and a polynomial transformation algorithm. Thirty random pixels along each of 30 random image lines were selected as point samples (n = 900) from within each image. The landscape feature represented in each selected pixel was classified into 1 of 15 cover types. Cover-type classification accuracy for the 2000 image was estimated to be 92.2% based on ground-truth data collected in the field. Classification accuracy was increased to 98.9% by combining rare or poorly separable cover-type classes. Image cover of vegetation cover types was quantified for each photography acquisition date. Changes in image cover of each cover type and direction of cover-type conversions were determined for each intervening time period. Analysis of image cover using repeated landscape photography is constrained by limitations imposed by oblique view angles and variable image quality. Repeat landscape photography, however, can be used to quantitatively assess long-term dynamics of vegetation cover on rangeland landscapes with visually distinct vegetation types.