Forest Ecology and Management 109 (1998) 231±239
Site factors as multivariate predictors of the success of natural regeneration in Scots pine forests D.O. Tegelmark* Department of Forest Yield Research, Swedish University of Agricultural Sciences, Box 7061, S-750 07 Uppsala, Sweden Accepted 8 January 1998
Abstract Site factors, such as regional climate (temperature, humidity, frost and radiation) and ground conditions (topography, soil and water properties and vegetation) were measured in sample plots in 26 naturally regenerated Scots pine-dominated stands in southern Sweden. The data were employed to develop multivariate prediction models for assessing the suitability of particular sites for natural regeneration with seed trees. The suitability for natural regeneration was estimated as the number of established coniferous stems at normal precommercial thinning age. Cross validated Partial Least Squares (PLS) regression models with different types of independent variables were developed. The models revealed strong climatic correlation with regeneration the main variation in which can be expressed in terms of latitude and altitude, e.g. positive correlations with temperature sum, length of growing season and humidity, and negative correlation with frost frequency. The best model explained 94% of the variation in number of established stems in the stands. # 1998 Elsevier Science B.V. Keywords: Pinus sylvestris; Climate; Soil; Regeneration ability; Seedling establishment; PLS-modelling
1. Introduction While regeneration of Scots pine (Pinus sylvestris L.) with the use of seed- or shelter-trees for natural sawing can result in dense stands, there is also a large risk that regeneration success will be poor. One key to fully utilise the potential of the method is to determine the suitability of sites for natural regeneration. To make such an assessment, a prediction of the stand density would be necessary, on the assumption that the site was naturally regenerated. High initial density in Scots pine stands is an important requirement for achieving high timber qualities (Voukila, 1982). *Correspondence address: Tel.: +46 23 778668; fax: +46 23 778601; e-mail:
[email protected] 0378-1127/98/$19.00 # 1998 Elsevier Science B.V. All rights reserved. PII S0378-1127(98)00255-2
The natural regeneration is the result of a chain of component processes occurring over several years, from the initiating of ¯owers to the survival or mortality of established seedlings. The success of each of these processes depends in turn on many environmental factors. These are both climate factors as temperature, humidity and radiation regimes during the growing season and ground factors such as soil moisture, depth of humus layer and soil texture affecting the water and nutrient availability at the soil surface and in the soil horizon. Topography can also be important by its in¯uence on the temperature and water regime. The objective of the present study was to develop models for predicting the stand density of naturally regenerated conifer stands based on easily estimated site factors. The work focused on Scots pine forest on
232
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
moderately productive sites in the hemi-boreal and southern boreal zones of Sweden. The number of competitive seedlings that has a potential to play a role in the future development of the stand and the tree properties is changing due to the developing stages of the stand. The prediction variable used in this study was the number of established and competitive coniferous trees (Scots pine and Norway spruce, Picea abies L.(Karst.) which intended to quantify the number of established trees at the precommercial thinning stage (about 10±20 years depending on the site) with a potential to survive and in¯uence the future growth and tree properties of the stand. 2. Material and methods In the hemi-boreal and southern boreal zones of Sweden (see Ahti et al., 1968), 26 naturally regener-
ated pine dominated stands prior to ®rst thinning on moderately productive sites were selected to form a suitable range of site types, site index, and geographic locations for modelling (Fig. 1). No beeting had been carried out in the stands. Norway spruce (Picea abies (L.) Karst.) and birch (Betula sp.) were present to various extents. Stand borders were set in the ®eld so that each stand represented a single site type. In each stand, one to seven plots (depending on stand size) of 100 m2 were located randomly within the stand borders. The stands differed in age (time since regeneration cut), ranging in one group from about 10±17 years, and in another group 25±42 years. In the young group of stands, not subjected to precommercial thinning and typically less than 5 m in tree height, all pine and spruce trees not considered suppressed or too small for potential development were registered. In the older group of stands, typically subjected to precommercial thinning a number of
Fig. 1. Distribution of stands in terms of latitude, longitude and altitude.
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
years before the survey and with tree heights of about 7±15 m for dominating and co-dominating trees, two types of trees were registered. First, trees exceeding 5 cm diameter at breast height (dbh) representing not suppressed or too small trees for potential development at the stage for precommercial thinning. Second, stumps originating from a previous precommercial thinning, representing the removed but established trees with potential to survive and develop. Hence, only stands where all stumps could be identi®ed were accepted. Trees smaller than sample trees at the survey were assumed not to be competitive nor to have been competitive at the age for pre-commercial thinning and, therefore, not in¯uencing the development of the stand. The prediction variable used was number of competitive coniferous stems per hectare at the pre-commercial thinning stage (CR) and was only intended for comparisons between stands for classi®cation purposes. The number of CRs ranged from 1500 to 9433 in the stands with 5472 CRs on average (88% Scots pine, 12% Norway spruce). No correlation was found between the percentage of spruce and the soil moisture, site index, regional frost frequency, temperature or humidity. Each plot was described in terms of the following site factors: soil type, soil texture, depth of the soil layer, soil moisture type, depth of the humus layer, depth of the leachate horizon, soil pro®le type, ground-cover percentages for species in the ®eld and ground layers (recalculated to vegetation type classes, HaÈgglund and Lundmark (1981)). A slope factor for solar radiation was calculated as the percentage of the radiation impinging on a corresponding horizontal surface at the same latitude. The calculation was made by interpolating slope-factor values calculated by MoreÂn and Perttu (1994) for latitudes 568N and 668N on slopes at 58 intervals with azimuths at 458 intervals. De®nitions are presented in Table 1. All plot values were averaged to stand level. Temperature sums, starting and ending dates for the growing season (Perttu et al., 1978) and frost frequencies (Perttu and HuszaÂr, 1976) for a period of 15 years corresponding to the seedling establishment and early growth period of the stands were obtained from the network stations of The Swedish Meteorological and Hydrological Institute (SMHI). Global radiation sums for the growing season were calculated by using
233
functions based on the latitude and altitude (MoreÂn and Perttu, 1994) of individual stands. All indices were calculated with 58C, 68C and 88C as threshold values, except for single month values where only 58C were used. To obtain climate indices for each stands, one to ®ve SMHI network stations were selected at a maximum distance of about 50 km and at elevations both above and below the stand. Special care was taken for coastal climates. To estimate climatic factors at the site, the following model was applied: Cs
n X
Ii C i
i1
where Cs and Ci are climatic factors at the site and network stations, respectively, n the number of network stations and Ii is a weighted index for the adjacency of each station, calculated as Pi Ii P Pi 1
1
Pi Pdi 1 Pai 100 1 di
ai 100
where di is the distance from site to network station (km), ai is the altitudinal distance between site and station (m) and Pi is the product of a network station's weight indices for distance and altitudinal distance to the site. To correct for the difference between the weighted mean altitude of the network stations and the altitude of the stand, the altitude component in functions for regional temperature indices (MoreÂn and Perttu, 1994) was applied to the calculated temperature indices for each stand. 2.1. Statistical analysis Stand data was analysed by the multivariate method called partial least squares regression or projection to latent structures (PLS) (e.g. Wold et al., 1982). The method is suitable for data sets where the X variables are internally correlated (Wold et al., 1984) and is a generalisation of principal component analysis (PCA) where a projection model is developed predicting Y from X via scores of X. The X variables are replaced by latent variables (components) similar but not identical to the latent variables computed in PCA. The PLS components are estimated to simultaneously describe the variation in X space and to correlate to the Y vector.
234
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
Table 1 Definitions of variables and original units Variable
Definition and original unit
Alt2 Lat2 Lat60p, Lat59 Alt190p Czon4
Altitude, m.a.s.l., transformed to power 2. Latitude, 8N in decimal form, transformed to power 2. Dummies, latitude >60.38N and latitude >57.88N and <59.48N Dummy, altitude 190 m.a.s.l. or above. Zone 4 according to the `Pomologiska foÈreningens zonkarta' (in Anonymous, 1975), climatic zones (rated in terms of severity) in Sweden. In the studied region, zones 1±7 are represented. The higher the zone the more severe the climate. Ê ngstroÈm (1974). Dummy for sites in the locally continental area `K3' in the middle of province of GoÈtaland, from A Length of growing season, days, threshold temperature 88C. (Perttu et al., 1978). Start of growing season and end of growing season. Julian day number, threshold temperature 88C (Perttu et al., 1978). Temperature sum during growing season, Degree days, threshold temperature 88C (Perttu et al., 1978). Temperature sum in August, Degree days, threshold temperature 58C. (Perttu et al., 1978). Global radiation sum during the growing season (threshold temp.88C) GJ mÿ2, (MoreÂn and Perttu, 1994). Number of hours with sunshine in June (Atlas oÈver Sverige, in Anonymous, 1975). Humidity index for the summer period calculated according to a formula suggested by Martonne (Eriksson, 1986). Frost frequency (below 08C) in May±June (25/4±25/6)*10 and September (25/8±25/9)*10 (Perttu and HuszaÂr, 1976). Soil moisture class (HaÈgglund and Lundmark, 1981). 1very dry, 2dry, 3intermediate, 4mesic, 5intermediate, 6moist, 7wet. Dummy, soil moisture class 5 or higher. Width of soil humus horizon, cm, (horizon A0, Troedsson and Nykvist, 1973). Soil texture class (HaÈgglund and Lundmark, 1981), Class 2gravel, 3-4coarse sand, 5-6fine sand, 7silt, 8clay. Texture dummy, class 4 or higher. Soil type dummy for sediments. Soil water flow class (HaÈgglund and Lundmark, 1981). Classification according to inclination and distance to top of the hill. Class 1never or seldom, 2short periods, 3longer periods. Slope factor for adjustment of the global radiation depending on declination, aspect and latitude (MoreÂn and Perttu, 1994). Vegetation type classification (HaÈgglund and Lundmark, 1981). Lichen-type class 1±2. EÂmpetrum type4, VaccõÂnium types 5±6, Grass types 8±9, Herb types11. Vegetation type dummies, VaccõÂnium myrtõÂllus type (class 6) and Grass types (class 8±9).
Cont3 GrSe Sta, Stop Tsum Tsag Grad SunJ HumId F0mj, F0sp Mois Mois5p HzA0 Tex Tex4P Sed Swat SloF Vtyp VTBlu, VTGrs
The components are estimated consecutively on the residuals of X and Y and they are used to regress the Y variables with linear regression. The predictive power of components were determined by cross validation (Stone and Brooks, 1990). In the cross-validation procedure the observations were divided into seven groups. The data set was temporarily reduced by one group while a model was ®tted to the remaining data set. The model was used for predicting Y values of the excluded observations. This was made seven times until all observations had been excluded ones and their Y values predicted. The fraction of the total variation of the Y's that could be predicted by a component (Q2), was calculated as Q2
1 ÿ PRESS=SStotal where PRESS is the Predictive Sum of Squares and
SStotal is the Sum of Squares of the deviation of the observed y values from the mean. PRESS was calculated as the squared differences between predicted and observed Y values, X PRESS
y ÿ ^y2 and SStotal, was calculated as X
y ÿ y2 SStotal New components were added to the model as long as they were considered statistically signi®cant, that is, if PRESS/SStotal was statistically smaller than 1.0 at the 5% level. The cumulative Q2 of all components (Q2cum , predictive ability of the model) was computed. Only models with a predictive ability >0.5 were accepted. Prior to calculations, the variables were centred and scaled to unit variance to give all variables the same relative importance. Statistical analysis were
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
done with the SIMCA-S 5.1 package (Anonymous, 1994). Four models were constructed that included different categories of X-variables; M1 (regional climate, ground, vegetation type variables), M2 (regional climate and ground variables), M3 (regional climate variable) and M4 (latitude, altitude, ground and vegetation type variables). Regional climatic variables include latitude and altitude, and ground variables include a slope factor variable. The dependent variable was transformed after analysis of residuals (Sabin and Stafford, 1990) in two models (M2 and M3) for improving the models prediction ability. To be able to compare the effects of the predictor variables with other studies, the models were utilised to compute regeneration results for regional climatic conditions independent of ground conditions and vice versa. The variables were divided into two blocks of data from the stands, one with climatic variables and the other with ground condition variables. For model M1, a matrix was formed with climatic conditions in the rows and ground conditions in the columns. In each cell of the matrix, a regeneration result was predicted representing the combined effects of one climatic condition and one ground condition. From this matrix, the row- and column-averages for each of the climate- and ground- conditions, were obtained. The average results were regarded as being independent of ground conditions or climatic conditions, respectively. To evaluate the correlation of regeneration with the closest possible to single variables, the variables in the two blocks were plotted against the obtained row- and column-averages of predicted regeneration results. In this way, only strong correlations can be seen since the simulated regeneration result is determined by the complete climate- or ground condition-variable block at the site rather than by only the one studied variable. 3. Results Models predicting a signi®cant amount of the variation of the success of natural regeneration in pine forests were constructed, using different numbers of climatic, ground, and vegetation site factors as predictors. The model using regional climate, ground and vegetation as predictors, model M1, explains the
235
Table 2 Statistical characteristics of the models. R2X and R2Y (goodness of fit) are the fraction of sum of squares of all the X- and Y-variables respectively explained by each component. Q2 (prediction ability) is the fraction of the total variation of the Y-variables that can be predicted by a component. All figures concern cumulative characteristics for all extracted components in the models Model
Comp.
R2X
R2Y
Q2
M1 M2 M3 M4
3 3 1 2
0.423 0.545 0.338 0.421
0.944 0.853 0.679 0.816
0.806 0.663 0.597 0.697
largest amount of the variation in the dependent variable, 94% (R2Y0.94) and have the highest predictive ability (Q2), 81%. The simpler models with fewer types of predictors have all lower values of Q2 and R2Y, see Table 2. The relative in¯uence of the predictor variables on the result can roughly be interpreted from the centred and scaled (to unit variance) variable coef®cients. But to make meaningful interpretations several coef®cients have to be interpreted at the same time. Each of the included variables has comparable in¯uence in all models. Among the climatic variables, there is a positive correlation in the models between the number of established stems and increasing latitude up to 608N, length of growing season on autumns, temperature sum and humidity. Increasing altitude and frost frequency are negatively correlated with the number of stems. Among the ground condition variables, there is a negative correlation between the number of stems and increasing moisture class and ®ner textured soils. Regression coef®cients for computations of new predictions from predictor variables in original units are presented in Table 3. Results from the predictions isolating the in¯uence of regional climatic factors from that of the ground condition factors show that the geographic location, de®ned as latitude and altitude, has a strong correlation with the regeneration result. In Fig. 2a and b, the stands between the latitudes 588N and 608N and below 160 m altitude have the best regeneration results and poor results are obtained on sites further south or north and above 200 m altitude. Also, the direct regional climatic indices show a good correlation with the regeneration result. The temperature sum (Fig. 2c),
236
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
Table 3 Coefficients for original values for model M1±M4. The dependent variable CR raised to the power 0.5 is denoted by CR0.5. Figures in parentheses behind variable names indicate cantering values that should be withdrawn from the measured values when used with the coefficients M1 (CR) Const Alt2 (-27000) Lat2 (-3461) Lat60p Lat59 Alt190p Czon4 Cont3 GrSe (-160) Sta (-122) Stop (-283) Tsum (-800) Tsag (-271) Grad (-2.3) SunJ (-294) HumId (-38) F0sp F0mj Mois Mois5p HzA0 Tex Tex4p Sed SWat SloF Vtyp VTBlu VTGrs Sta*Sta Sta*Sta*Sta SunJ*SunJ F0mj*F0mj Mois*Mois Vtyp*VTyp Alt2*SloF Lat2*GrSe Stop*TSag Grad*SloF Grad*SunJ Mois*HzA0 Mois*VTyp Tex*Sed Tex*VTyp
36316 ÿ0.02668 2.0617 ÿ331.29 ÿ965.32
M2 (CR0.5) 150.93 0.013676 ÿ3.1797 1.8326 ÿ9.9884 ÿ5.8982
6.8113 51.649 9.2888 4.1509 2.5424 ÿ1352.6
0.069985 0.87132 0.12769 0.021778 ÿ0.03214
53.384 ÿ56.707 ÿ163.23 ÿ4059.6
0.30861 ÿ0.39311 ÿ1.0525 ÿ2.5333
259.57 ÿ710.7 4271.8
0.90629 ÿ5.8293 34.604
80.464 ÿ3971.5 ÿ1073.2 710.31 ÿ51.133
ÿ0.5817 0.11025
0.84323 123.1 ÿ0.00637 0.17884 ÿ6.3687 1920.4 572.87 ÿ918.53
0.005549
M3 (CR0.5)
M4 (CR)
116.35 ÿ8.55Eÿ05 0.004161
20811 ÿ0.02381 1.2115
ÿ5.4529 ÿ4.1547 ÿ5.5915
900.1 ÿ1612.5
0.25936 0.12159 2.3791 0.12028 ÿ0.64501
ÿ1490.5 ÿ75.451 ÿ261.47 ÿ1195.8 3678.1 ÿ17.673
ÿ0.38574 0.038869 ÿ0.00965 0.00316
ÿ2798.7 ÿ1105.5 712.95
156.98 141.75
0.00153 ÿ0.035 1.8985 ÿ7.2052
length of the growing season and regional humidity are positively correlated with the regeneration result. The regional frost frequency are negatively correlated with the regeneration result.
42.127 ÿ685.69 199.37
The predictions also show correlations between the regeneration result and ground conditions. Among soil types, coarse texture and dryness are positively correlated with the regeneration result (Fig. 2d and e).
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
237
An increase in the richness of the vegetation type also have a positive in¯uence in the studied interval. 4. Discussion
Fig. 2. Single predictor variables versus simulated CR-values for climatic conditions (a±c) and ground conditions (d,e) of stands. Each CR-value is an average for all ground conditions (a±c) or climatic conditions (d,e), respectively, see text. (a) altitude, (b) latitude, (c) temperature sum (threshold 88C), (d) soil texture class, (e) soil moisture class. See Table 1 for definitions of variables. Simulations with model M3 (a,b), M1 (c) and M2 (d,e).
Since the models were developed from survey material, they are not causal. Although the variable coef®cients can be interpreted, they are all internally correlated. Models used for prediction can be descriptive or causal, the choice of approach depending on the purpose of the model and data availability. If the purpose is to predict responses in different environments, a descriptive model can be suf®cient. An advantage is that empirical data can be used while causal models require data from statistically designed experiments. As an empirical material is multivariate, it is impossible to know whether a change in the measured response is due to changes in the measured parameters or to something not measured. Therefore, it is important to compare how well the descriptive model parameters agree with earlier causal models for related phenomena. Many earlier studies has shown similar climatic in¯uences on various stages in the regeneration process of Scots pine in Sweden as expressed in the presented models. The in¯uences of latitude and altitude correspond to the in¯uence on the cone production of Scots pine for southern Sweden (Hagner, 1958) and the negative in¯uence of altitude on seedling establishment for northern Sweden (TireÂn, 1945). A (¯at) latitudinal maximum at 61±628N in Sweden and a negative altitudinal in¯uence has been shown (StaÊhl et al., 1990) for the frequency of planted Scots pine stems without weather defects, which is closely connected with survival. Persson and StaÊhl (1993) expressed a decrease in the survival of planted Scots pine as a function of increasing altitude and latitude above 608N in Sweden. The length of growing season and temperature are also important factors for seed maturation and germination (SahleÂn, 1992), the size of the seed crop (Leikola et al., 1982), the growth rate (Brand, 1990; Linder and Flower-Ellis, 1992) and the survival of seedlings (Saarenmaa, 1990) and planted Scots pine (Persson, 1994). Frost can reduce seedling survival (Cristersson, 1971) and damage the seeds at the end of the growing season (Keefe and Moore, 1982). A well-known regeneration problem is
238
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239
connected with low humidity in early summer in Sweden. The narrow range of ground conditions in the study, i.e. normal forest land on moderately productive sites, explains the relatively small in¯uence of ground conditions in the models. Simulations based on ground conditions, assumed to be independent of climatic conditions, are not absolutely accurate owing to the considerable in¯uence of climate on soil moisture, humus layer and vegetation, etc. However, the most important factors affecting soil moisture, humus layer formation and development of the vegetation type are the soil texture, nutrient content and topography. The assumption behind the simulations is that the internal variable combinations are independent enough to give indications of the ground condition in¯uences in the models. Earlier studies has shown similar ground in¯uences on regeneration as expressed in the presented models. A coarse texture was shown by Ferrel (1953) to be positive for survival of Loblolly pine. Site dryness was shown to be positively correlated with the number of regenerated Scots pine seedlings, by Ackzell (1994) in northern Sweden, but this contradicts results by Ferrel (1953). Ackzell (1994) also found sediments to be positive for natural regeneration of Scots pine. Vegetation characteristics are important for seed germination and early seedling performance (Steijlen et al., 1994) and strengthens the models. However, the vegetation-type variables used in the models were estimated in young stands 25±42 years after regeneration, and may differ from those assessed in older stands before regeneration, since the vegetation type and species distribution changes after clear-felling (Olsson, 1995). The most important vegetation characteristic is most likely quantity after overstory release. 5. Practical implications The models could serve as a valuable tool for ranking different sites in the hemi-boreal and southern boreal zones of Sweden in terms of their potential to support natural Scots pine regeneration. The presented models are valid in the hemi-boreal and southern boreal zones of Sweden around latitude 56.5±
60.58N up to about 250 m altitude on moderately productive forest land. This corresponds with a regional climate with a growing season length of 140±170 days and a temperature sum of 650±920 d.d. (both 88C threshold), a site index for Scots pine of 21±26 m (HaÈgglund and Lundmark, 1981), dry to moist, ®ne and coarse sands and vegetation types ranging from EÂmpetrum-type to various grass types. Similar models for different areas or species could be similarly constructed. Acknowledgements I am grateful to all the people involved in the ®eld work to ®nd suitable stands for the study and to my colleagues for offering valuable criticism. David Tilles corrected the English. The study was ®nancially supported by Grants from the Swedish Council for Forestry and Agricultural Research (SJFR).
References Ackzell, L., 1994. Natural regeneration on planted clear cuts in boreal Sweden. Scand. J. For. Res. 9, 245±250. Ahti, T., HaÈmet-Ahti, L., Jalas, J., 1968. Vegetation zones and their sections in northwestern Europe. Ann. Bot. Fenn. 5, 169±211. Ê ngstroÈm, A., 1974. Sveriges klimat (The climate of Sweden). A Generalstaben Litografiska Anstalts FoÈrlag. Stocholm. 188 pp. (in Swedish). Anonymous, 1975. BestaÊndsanlaÈggning (Stand establishment). Skogsstyrelsen. JoÈnkoÈping. 406 pp. ISBN 91-38-02270-2. Anonymous, 1994. User's guide to Simca-S. Version 5.1. Umetri AB, UmeaÊ. Brand, D.G., 1990. Growth analysis of responses by planted white pine and white spruce to changes in soil temperature, fertility, and brush competition. For. Ecol. Manage. 30, 125±138. Cristersson, L., 1971. Frost damage resulting from ice crystal formation in seedlings of spruce and pine. Physiol. Plant. 25, 273±278. Eriksson, B., 1986. NederboÈrds och humiditetsklimatet i Sverige under vegetationsperioden (The precipitation and humidity climate of Sweden during the growing season). SMHI Reports, Meteorology and Climatology 46, Swedish Meteorological and Hydrological Institute, NorrkoÈping. (in Swedish with English abstract). Ferrel, W.K., 1953. Effect of environmental conditions on survival and growth of forest tree seedlings under field conditions in the Piedmont region of North Carolina. Ecology 34, 667±688. HaÈgglund, B., Lundmark, J.-E., 1981. Handledning i bonitering med SkogshoÈgskolans boniteringssystem, Del 1-3 (Site index
D.O. Tegelmark / Forest Ecology and Management 109 (1998) 231±239 determination, Part 1-3). Nat. Board of Forestry, JoÈnkoÈping. ISBN 91-85748-11-0 (70 pp.), -13-7 (53 pp.), -14-5 (121 pp.) (in Swedish). Hagner, S., 1958. On the production of cones and seed in Swedish coniferous forests. Meddelanden fraÊn statens skogsforskningsinstitut, Vol. 47 (8), 120 pp. (in Swedish). Keefe, P.D., Moore, K.G., 1982. Frost damage during stratification: Mechanism and protection in Pinus Sylvestris seeds. Seed Sci. Technol. 10, 485±494. Leikola, M., Raulo, J., Pukkala, T., 1982. Prediction of the variation of the seed crop of Scots pine and Norway spruce (in Finnish, with English abstract). Folia For. 537, 1±43. Linder, S., Flower-Ellis, J., 1992. Environmental and physiological constraints to forest yield. In: Teller, A., Mathy, P., Jeffers, J.N.R. (Eds.), Responses of Forest Ecosystems to Environmental Changes. Elsevier Appl. Sci., Oxford, pp. 149±164. MoreÂn, A.-S., Perttu, K., 1994. Regional temperature and radiation indices and their adjustment to horizontal and inclined forest land. Stud. For. Suec., No 194, 19 pp. ISBN 91-576-4915-4. Olsson, B., 1995. Soil and vegetation changes after clearfelling coniferous forests: effects of varying removal of logging residues. Dissertation, Swedish University of Agricultural Science, Uppsala, Report 80, 25 pp. Persson, B., 1994. Effects of provenance transfer on survival in eight experimental series with Scots pine (Pinus sylvestris L.) in northern Sweden. Scand. J. For. Res. 9, 275±287. Persson, B., StaÊhl, E.G., 1993. Effects of provenance transfer and spacing in an experimental series of Scots pine (Pinus sylvestris L.) in northern Sweden. Swedish University of Agricultural Science, Report 35, 92 pp. ISBN 0048-7636 (in Swedish, with English summary). Perttu, K., HuszaÂr, A., 1976. Vegetationsperioder, temperatursummor och frostfrekvenser beraÈknade ur SMHI-DATA (Growing seasons, day degrees and frost frequences computed from SMHI-data). Department of Reforestation, Royal College of Forestry, Stockholm, Research Notes 72 (in Swedish). Perttu, K., Odin, H., EngsjoÈ, T., 1978. Computed climatic data from the network stations in Sweden. Part 2. Growing seasons, day degrees and growth units as mean values with standard deviations during 1961±1976. Department of Reforestation, Royal College of Forestry, Stockholm. Research Notes 101 (in Swedish).
239
Saarenmaa, L., 1990. Choice of reforestation method based on an expert system in Finnish Lapland (in Finnish, with English summary). Folia For. 762, 30. Sabin, T.E., Stafford, S.G., 1990. Assessing the need for transformation of response variables. Forest Research Laboratory, Oregon State University, Corvallis, special publication 20, 31 pp. SahleÂn, K., 1992. Anatomical and physiological ripening of Pinus Sylvestris L. seeds in northern Fennoscandia. Dissertation, Swedish University of Agricultural Science, UmeaÊ, 18 pp. StaÊhl, E.G., Persson, B., Prescher, F., 1990. Effect of provenance and spacing on stem straightness and number of stems with spike knots in Pinus sylvestris L.: Northern Sweden and countrywide models. Stud. For. Suec. 184, 16 pp. ISBN 91-5764345-8. Steijlen, I., Nilsson, M-C., Zackrisson, O., 1994. Seed regeneration of Scots pine in boreal forest stands dominated by lichen and feather moss. Can. J. For. Res. 25, 713±723. Stone, M., Brooks, R.J., 1990. Continuum regression: Crossvalidated and sequentially constructed prediction embracing ordinary least squares, partial least squares and principal components regression. J. Roy. Statist. Soc. Ser. B 52, 237±269. TireÂn, L., 1945. Erfarenheter av naturlig foÈryngring, Inledare II: JaÈgmaÈstare Lars TireÂn. (Experiences of natural regeneration) Sveriges skogsvaÊrdsfoÈrb (in Swedish). Tidskrift 43, 117±135. Troedsson, T., Nykvist, N., 1973. MarklaÈra och markvaÊrd (Soil Science and Soil Tending). Almqvist and Wiksell LaÈromedel AB, Stockholm, 403 pp. (in Swedish). ISBN 91-21-04114-8. Voukila, Y., 1982. The improvement of the technical quality of forests. Folia For. 523, 1±55. Wold, S., Albano, C., Dunn III, W.J., Esbensen, K., Hellberg, S., Johansson, E., SjoÈstroÈm, M., 1982. Pattern recognition: finding and using regularities in multivariate data. Food Research and Data Analysis. In: Martens, H., Russwurm Jr, H. (Eds.), Proc. IUFoST Symp. 20±23 September 1982, Oslo, Norway, Applied Science Publishers, London, pp. 147±188. ISBN 0-85334-2067. Wold, S., Albano, C., Dunn III, W.J., Esbensen, K., Hellberg, S., Johansson, E., Lindberg, W., SjoÈstroÈm, M., 1984. Modelling data tables by principal components and PLS: class patterns and quantitative predictive relations. Analysis 12(10), 477±485.