Olsson, Häkan 
MONITORING FOREST GROWTH USING LONG TIME SERIES OF SATELLITE DATA 
Steve Joyce and Häkan Olsson 
Department of Forest Resource Management and Geomatics 
Swedish University of Agricultural Sciences 
SE-901 83 Umeä, SWEDEN 
Steve.Joyce @resgeom.slu.se, Hakan.Olson@resgeom.slu.se 
Working Group VII/A 
KEY WORDS: Forestry, Monitoring, Multitemporal, Landsat 
ABSTRACT 
Forest managers and policy makers require timely information about the current state of forest resources over large 
areas. One important factor is forest growth and compositional change, which currently can only be approximately 
modeled in the time between field surveys. This study investigates whether a sequence of satellite images from 
Landsat-5 TM can be used to monitor forest growth, and specifically compare growth rates between different forested 
plots from the Swedish National Forest Inventory. Seven Landsat Thematic Mapper scenes were acquired over a 12- 
year period during the vegetation season. An image-based relative calibration procedure was applied to normalize the 
images for differences in atmospheric clarity and other specific conditions of image acquisition, and pixel values were 
extracted at sample plot locations for each scene. Plots were then selected from the inventory data for comparison of 
their spectral profiles over the 12-year period. Longitudinal regression models were fit to the datasets to test the 
significance of site index as an explanatory variable. It was found that once the effect of age was removed, the recorded 
site index could not explain the residual variance in individual plot trajectories. This is probably due to problems with 
precise positioning of the plots in the image, but also the fact that the site index is a rather coarse predictor of actual 
growth. 
1 INTRODUCTION 
1.1 Background 
Effective forest management requires detailed information about the current state of the resource as well as tables or 
models for forecasting future conditions. With this information, forest managers are able to select appropriate 
management treatments to optimize economic output as well as achieve certain preservation and diversity goals. In 
Sweden, and typical in other countries, new management inventories start with delineation of homogeneous stands in 
aerial photographs followed by measurement of forest parameters at a number of sample plots within each stand. 
Inventory parameters, stored in a database, are updated in subsequent years following the inventory using growth 
models, from recorded harvesting activities, and from periodic re-measurement. After some time, it is typical for 
uncertainty to accumulate in the inventory data, either from errors in the initial estimates, from departures from 
predicted behavior in the models, from unexpected damages, and from inaccurate updating of management treatments. 
After some years of continuous update, it is typical for errors to be so large, that the entire inventory must be re-done 
from scratch. There is considerable economic benefit if the information in the inventories can be improved, or if the 
useful life can be extended by even a few years. 
On a larger scale, the National Forest Inventory (NFI) in Sweden collects information about the forest land for the entire 
country using a network of permanent and temporary sample plots. There are roughly 18000 sample plots distributed 
throughout the country with a higher sampling density in the south. Temporary plots are allocated and measured only 
once, while permanent plots are revisited on roughly a 5-year cycle. Information from the NFI is used to identify long- 
term trends in wood-supply, forest composition, and health, and is an important input for environmental monitoring and 
setting national forest policy. While the purely plot-based design does provide objective estimates of forest state, it is 
not particularly efficient for sampling certain attributes, such as annual cutting intensity or rare forest types. 
In both of these inventory applications, there is a need to monitor forest growth and changes in the years between field 
samples, and to generalize measurements taken on sample plots to areal units. Satellite remote sensing, from moderate 
resolution sensors such as SPOT or Landsat-TM, has always offered much promise in this field, since it can provide an 
  
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1081