As in the other applications discussed here, these remote sensing-derived 
area data can be combined with ground plot data to produce more precise 
forest parameter estimates and estimates for smaller administrative regions. 
An alternative, is to use the remote sensing data directly with ground data to 
estimate or simulate a complete set of measurements for all sample plots, 
satellite picture elements, or forest stands/compartments. Such an approach 
has been suggested by the work of Poso and Simila (1987, and personal 
communication), Tomppo (1987, and personal communication), and Hagner 
(1989, and personal communication). Estimates of many kinds could then be 
produced independent of an initial cover type stratification/classification 
strategy. 
The degree of detail in cover type specification will likely be limited initially 
as compared to the degree of detail currently obtained on NFI ground plots 
regarding tree species mix, stocking density, and tree size or cutting class. 
This detail should improve with further research on remote sensing 
information extraction and on related estimation techniques for linking plot 
and stand data with remote sensing data. Similar comments can be made 
about determining detail regarding the cause or type of change, when a 
change is detected. 
Based on this review of previous, and on-going work, it is likely that the 
sample variance of many forest parameter estimates now made will be 
lowered when remote sensing cover type information is used together with 
the current NFI ground plot data. Use of remote sensing data will also enable 
more localized estimates by, in effect, increasing the area subjected to some 
form of measurement. Recent satellite/ground data-based simulation results 
at the stand (e.g., Hagner 1989 and 1990) , and higher (e.g., commune - 
Tomppo 1990) levels of aggregation have shown reasonably good agreement 
will standard reference data. 
Other findings include (1) a trend towards the use of automatic image 
segmentation (e.g., Tomppo 1989 and Hagner 1990) to improve results over 
pixel-wise classification. (2) As with clear-felling, cloud cover has been 
found to be a major problem in some areas of Sweden, reducing the 
availability of cloud-free scenes from a given satellite sensor system to one 
every three to five years. Thus it is likely that a robust system should be 
flexible enough to mix sources of satellite data and to substitute other sources 
of cover type information, such as recent map, photo, and ground data, when 
required. (3) The capability exists in Sweden for use of very large scale 
photography (greater than 1:2,000) scale for extending the sample for 
computer class training and subsequent stratification verification. (4) 
Capability exists for detection of major cover type change in satellite data in 
Sweden. However, detection of more subtle changes will require more 
research in calibration for haze effects by band, understanding the 
relationship between canopy structure and spectral reflectance, changes in 
ground plot layout and measurements that will give data more directly 
corresponding to spectral reflectance or emission phenomena, image 
segmentation for change, change detection and estimation using different 
sources of data, and the development of an integrated procedure for 
detecting and labelling many types of change. (5) Radar and lidar hold 
potentially important possibilities for stand characterization, but require 
much more research before operational use can be planned. 
Forest Area Subjected to Thinning 
411