Short talk to be held at the workshop on image analysis and spatial statistics in forestry on November 2, 1999 at KVL, Frederiksberg, Denmark

Gaps in young forests and their spatial distribution

Jan Naumburg (Jan.Naumburg@sh.slu.se), Department of Forest Management and Products, Swedish University of Agricultural Sciences, Uppsala, Sweden.

This project is a part of my thesis that deals with gaps in young forest stands, reasons for the origin of gaps and their distribution in young stands of Picea abies and Pinus sylvestris.

The forests, or stands, that are investigated are all planted with either Picea abies or Pinus sylvestris. Half of them are planted 1972 and the other half are planted between 1985 to 1992. One aim at the time of regeneration is to see to that the plants are evenly distributed. Due to production and other purposes it is important that each stand have as few gaps as possible, as gaps over a certain size causes a reduction of the total volume production.

The main hypothesis is that single trees do not die in a randomised pattern, they die in clusters. The clustered, dead trees form a gap, and the gaps are distributed in some way in the stands. The reasons for origin of gaps differ from gap to gap, and even within the gaps. This part of my thesis will only deal with the gaps as such and their spatial distribution.

There are 20 different stands that have been photographed in stereo with full colour from 1500 meters above each object. The scale of these photographs is 1:30.000. There have been an inventory of each stand, some of the stands have a very intensive sample plot frequency. In 5 of the stands planted 1972 there are between 60 - 120 sample plots, and the sizes of the square spacing differs between 17 up to 68 meters. The other stands have larger spacing between the sample plots. The surveyed area always exceeds 3% of the stand area, and each sample plot has a 5.64m radius.

I have studied the laser scanning technique, and I have a nice data set that I will be working with later on to ascertain to what extent it is possible to detect single trees in young forests. But for the time being, it is too expensive to scan all my forest stands, as they are spread out over a large geographical area.

The aerial photographs are scanned, and each pixel represents approximately 3x3 dm. The photos will be interpreted in a stereo analysing instrument, with respect to gaps and stand border. These gaps and borders will be digitised.

There are two levels that I will focus at. The first level is the gap itself and the relation between the gaps within the single stand. The second level is the stand with its gaps.

Gap level

For the single gap, I will look at the geometrical form, in terms of area and shape measures. I will also focus at the relative relation between the different gaps (a nearest-neighbour approach) and at different density measures.

Stand level

The focus will be at an attempt to find a distribution index, but I will also study the total gap area, some total nearest-neighbour figures, and edge and shape totals for the gaps within the stand.

To do these calculations I intend to use FRAGSTATS, a spatial pattern analyses program for quantifying landscape structure, with my digitised gaps and stand borders. The data from the different levels will be analysed and I will study the eventual resemblances that are between the stands. I want to know if it is possible to discern any patterns in the spatial distribution of gaps within different stands. If there is a pattern, it will be possible to develop a model that can estimate gaps and eventually gap distribution on a stand level. If a goal is to have forest mensuration with high accuracy it's necessary to have good knowledge about gaps as they can cause production losses.

The digitised gaps will serve as the truth. With ordinary forest inventory methods it is hard to detect rare objects. Depending of the gap size, some gaps will be missed and some not and the overall estimation of gap frequency will differ compared to reality. By comparing the "true" data with the data from the inventories, it is possible to estimate the error of gap estimation.

An image program like Erdas Imagine, can be used to classify images with respect to gaps automatically. The gaps that are digitised and therefor considered true can be used as training areas for classification of the images. By using only some of the gaps, it is possible to use the rest of the gaps as control. It is also possible to see how much information about gaps you miss if the image is interpreted without stereo-instrument, and only a few gaps are used as training area for the automatic classification. As a last thing, it would probably be possible to investigate how much ground information you need to classify the image correct.