CHAPTER II
RESULTS AND DISCUSSION

The maps created are useful in that all data for each site are thematically assembled into one database and can be easily accessed and spatial patterns can be evaluated. Figure 3.2 shows a Hoodsport map assembled including tree mortality data (Thies personal communication 2003). Subsets of data such as all trees that died in a particular year, or all bulk density values in a certain range can be selected and viewed. The tree data will tell if the stumping successfully reduced the Phellinus mortality and if the changes in soil qualities noted in chapter one have affected tree growth. The ability to rapidly review data from all parameters at a specific location enables assimilation, analysis and educated decision making. Hoodsport is the only map where the tree data were available. No plot chacteristics other than the bulk density, nitrogen concentration and forest floor mass were obtained in the soils portion of the stumping study. Complete maps for soil data gathered at each site have been made (Appendix C) and can be altered as necessary to display desired information (Figure 3.3). Layers of data from other sources can be added to these maps. Topographical position of soil has important effects on soil properties, and can affect bulk densities and nitrogen levels. On these small sites, chosen for level terrain, this is not very important, but GIS software enables overlaying contour maps from digital elevation models (DEM) to assess the topograghical features of a site that may cause operational problems associated with factors such as slope or soil depth. Hand-held global positioning systems (GPS) can add local elevations to the maps to assess topology of these sites more accurately than with contours made from DEM layers.

Analyses of the data have already been done utilizing traditional statistical and mathematical methods (see Chapter II). Geostatistics adds position as an element for each measured variable. This allows comparing this measurement with other attributes of that position (such as slope or moisture content). Spatial location of the variables also allows predicting values for unmeasured locations. Figure 3.4 shows predicted bulk densities and mineral soil nitrogen levels at LaGrande.

An item of interest in the soil aspect of the stumping study is what effect pre-existing differences within the sites may have on the results. Using the post-treatment data, some idea of regional differences on the site can be presented using GIS and geostatistics. Figure 3.5 shows in a geostatistical map (ordinary kriging) that forest floor nitrogen (kg ha-1) tends to decrease in a northwesterly direction. GIS software enables analyses of trends in the spatial location of the data. In this case there is increasing elevation and a mesic toward xeric trend in the same northwesterly direction.Gates was the only site where the bulk densities of the no-stumping treatments were higher than the stumped treatments. Using GIS we can see that eight no-stumping plots were located close together in the center of the map in an area of higher bulk density (Figure 3.6). This may be a center of activity during the stumping operation, a landing area from the logging, or a local anomaly. Four more no-stumping plots are located to the north in an area that may be influenced by rocky cast from a road cut. The influence of these locations may have been enough to offset the probable increase in bulk density from stumping that was seen at the four other sites. Except at Gates, there does not seem to be another local factor that obscures the changes made by stumping.

GIS methods can relate the individual measurements spatially and give an idea of the overall effect on the site. As stated earlier, no measurements of soil variables were made before the study to give reference data. By using the data from the no-stumping plots as a base, comparisons can be made with the stumping effects. Figure 3.7 shows a probabilistic map of mineral soil nitrogen concentrations at Sweethome made geostatistically (ordinary kriging) using the no-stumping plots and comparing that with the data measured from all plots. It is known from the results presented earlier that the nitrogen concentrations were lower in the stumped plots. In Figure 3.7 GIS averaging makes it appear that nitrogen concentrations are lower over the entire site. Deterministic prediction layers utilizing methods such as inverse distance weighting are another way of predicting unmeasured values. These layers were also created (Fig. 3.8). These maps do not contain the statistical errors of the predicted values found in geostatistical methods. Each predicted location in the deterministic model is assigned a value calculated from its neighbor's value and distance.

Sweethome was measured both in 1991 and for this study in 2003. Using GIS to map these temporal changes can focus management or research attention on areas that might be problems for soil fertility or bulk density. Figure 3.9 shows that while most areas showed an increase in mineral soil total nitrogen, there is a cresent shaped area in the center of the site where nitrogen levels have decreased. This reduction in nitrogen may be because of the hydrology and increased leaching, or due to the lack of nitrogen fixing vegetation.

Spatial comparisons of site variability and autocorrelation can be made with semivariograms and covariances on the data using geostatistics. Five sites were chosen for the stumping study. Although all sites responded to the stumping with similar results, differences between the sites may have some effect on the results. GIS gives an opportunity to compare the five sites spatially. The correlation of neighboring measures of soil bulk density depends on many contributing factors, such as parent material, topology, organic content, and animal activity. These factors influence on differing scales at different sites. In this study the approximately 30 m spacing of the plots sets the scale being analyzed. The range in a semivariogram is the distance beyond which the variance of differences between data points are no longer related by the distance between them. The semivariograms give a measure of the scales of covariance or autocorrelation at each site. The differences in the semivariogram ranges show that for bulk density only Gates and LaGrande have a range of autocorrelation that falls within the site (Table 3.1). This implies some factor creating non-random dependence of these measures within this range. The LaGrande site is heavily utilized by domestic livestock, elk, deer, and rabbits. As we have seen, bulk densities at Gates may have been influenced by post treatment effects as well. For mineral soil nitrogen concentration, only Hoodsport and LaGrande show this same tendency of non random influence. These sites have much lower nitrogen levels than the other three sites, and factors that affect nitrogen levels may be more obvious. On all five sites the harvest and stumping treatments may have masked some variability. These site specific spatial correlations can be analyzed directionally (anisotropically) using GIS geostatistics.