Chen Z, Helmersson A, Westin J, et al (2018) Efficiency of using spatial analysis for Norway spruce progeny tests in Sweden. Ann For Sci 75:2. doi: 10.1007/s13595-017-0680-8
Sweden and Norway spruce progeny space efficiency analysis
Summary and key messages
Spatial analysis can improve the accuracy of genetic analysis to improve prediction accuracy Norway spruce test breeding value and genetic gain.
Environmental spatial analysis has been more and more applications in the genetic evaluation of tree species in field trials. However, the gene test in Sweden, with respect to the conventional experimental design or analysis before and after occlusion, the efficiency of the spatial analysis system has not been evaluated. Objective: To evaluate the effectiveness of spatial analysis in breeding value and to improve the accuracy of prediction of genetic gain. Methods: detachable line, column residual autoregressive process, for nine traits Swedish spruce 145 field trials of spatial analysis. Results: 96% of the variable (feature) is converged to space model. Large test block having a large variance from the model to the experimental design space model accuracy has improved greatly. Growth and Pi Luodeng traits measured on a logarithmic likelihood, the accuracy of genetic gain respect and show greater improvements. In Sweden - Norway spruce test by spatial analysis, variance traits block height and diameter reduced more than 80%, indicating that the pre-analysis block and the block, the block variance is more effective. Diameter and height values of the prediction accuracy of the breeding progeny were increased by 3.6% and 3.4%. Characters also improve the efficiency and growth of the test site location, age, number of tree survival and related test interval. Conclusions: Spatial Analysis in Sweden - the traditional model of analysis more effective experimental design method is based on tests than Norway spruce.
Keywords : Spatial. Norway spruce. Autocorrelation. Genetic gain
1 Introduction
As we all know, in tree breeding programs, because of the large-scale field trials and space needed for large populations of plant species (tens to hundreds of families, each family there are multiple individuals), descendants of the test usually covers a large area. Progeny test required large physical area is usually reflected in a considerable environmental conditions change (bian et al. 2017; dutkowski et al. 2006; chen et al. 2017). To reduce this environmental heterogeneity, usually the test is divided into several blocks experimental design (Williams et al., 2002). Randomized complete block (RCB) is by far the most common experimental design, in which a site is divided into several block (copy) to the environment (block) and genetic variations separately (White et al. 2007). In order to make more accurate comparison of the genetic entry, improve the accuracy of the predicted breeding and genetic parameters estimated values, the RCB is further proposed a new design, referred to as a balanced incomplete blocks (BIB), including an incomplete randomized block (RIB) . This is to reduce the overall residual error by further eliminating the associated environmental error between the row and / or column direction, and a small piece finish. Row / column design and other Latin square (LS) design is commonly used in forest genetic test BIB design (Williams et al. 2002). Although a variety of methods for blocking the Forest Genetic testing is possible to effectively reduce global (trend) variation, but initially for crops and trees for the recent introduction of spatial analysis to account for global and local micro-site variation, so as to further improve the breeding value forecast accuracy (Cullis et al. 1998; Ye-AnD Jayawickrama, 2008). Even traditional RCB design, implementation of spatial analysis model can still be improved by the micro-site variations between the detection block variation or change block body, including most tests and nutrients present in soil depth varying types (Dutkowski et al. 2006; Ye and Jayawickrama 2008).
Spatial analysis may be captured within a local gradient test piece (plaque) and the global trend line test block gradient (plaque) and along the columns, so it is a popular method for agriculture and forestry in field trials ( Anekonda and Libby 1996; Brownie and Gumpertz 1997; CulliS et al. 1998; Cullis and Gleeson 1989; Cullis and Gleeson 1991; Fox et al. 2007a, b; Gilmour et al. 1997; Qiao et al., 2000; Yang et al. 2004; Ye and Jayawickrama 2008; Chen et al. 2017). In the field of forestry, spatial analysis using several methods, such as post-occlusion method (Dutkowski, who in 2002; Ericsson 1997), nearest neighbor Adjustment Act (Anekonda and Libby in 1996; Joyce et al. In 2002; Wright 1978) and Kriging method (Hamann et al. in 2002; Zas 2006). However, crop and forestry trials the most common method seems to be a combination of experimental design and space components, using Gilmour et al. (1997) recommended a two-dimensional separable first order autoregression residual variation.
In Sweden - Norway spruce (Picea) breeding programs, some of the descendants of a great pilot scale, testing more than 1,000 families, including the provenance (stand) and family structure (Chen et al. 2014). In addition, in northern Sweden, many trials have adopted a completely randomized design (CRD), with no pre-blocking design. Traditionally, this design is used after blocking adjustment (PBA) have different effects analysis (Dutkowski, who in 2002; Ericsson 1997). Spatial analysis by reducing the environmental variation CRD field trials, greatly increase the value of predicted breeding.
Recently, several papers describe the competitive effects between adjacent trees(Cappa and Canet 2008; Cappa et al., 2015,2016; Costa E Silva and Kerr 2013; Costa E Silva et al. 2013; Dutkowski et al. 2002; Dutkowski et al. 2006; Ye and Jayawickrama 2008). Dutkowski et al (2006) reported that 10% of the diameter of the variable display competitive effect on the level of residual first-order autocorrelation coefficient column and row direction are negative. Ye and Jayawickrama (2008) report, a total of 1135 observed variables eight negative autocorrelation, either in one direction, either both directions of rows and columns. Stringer and Ku Lisi (2002) found that in many field trials sugarcane production, the inter-district competition is very fierce. In Sweden, most are located in 56 field trials? North. Northern latitude and trees are usually measured before the age of 20. Competition may appear some tests. Therefore, the first order spatial autocorrelation coefficient model to study whether competition played a role in this age group is significant.
The purpose of this study was to: (1) study the extent and severity of Sweden and Norway spruce genetic spatial variability of field trials; (2) the basis for the analysis of genetic diversity, it is estimated the average change in the variation of several components, as well as parental and descendants of the breeding value prediction accuracy. Symbol space model; (3) of different pitches, age and geographical factors affecting variance component estimates; (4) from the regression coefficients, testing whether there is growth traits based on first-order row, column direction competitive effects .
method
For open field test and control pollinated pollinated, and the use of the family structure clones test, the model calculated only additive variance components. For cloning experiments without parental lineage, genotype predicted value. In the model space, missing values were fitted as fixed effects.
For ASREML3.0 the logarithm likelihood ratio test (LRT), using the converted data not all features are analyzed. For comparison shows that between categorical variables, before the unconverted and the converted data is typically ranked a significant change (Rosvall et al. 2011) does not happen , there is no attempt to convert these other variables.
Is recorded as the count variable is not converted, the conversion is observed as the square root of the count used without further normalized distribution (Dutkowski et al. 2006).
In the model fitting process, we found that some variables have no significant effect on the Nuggets space model , which is similar to experiment with a number of agricultural crops. For these variables, Gilmour et al (1997) suggested that the effect of foreign fit to
the formula, b is a vector of fixed effects, has a large average linear row, column, and a linear edge effect corresponding design matrix X and u are vector random effects, with blocks of row-like, column-like, rows, columns and genetic (genotype or additive effect) corresponding to the design matrix Z is dependent on the residual space. Cross lines and columns of cross not suitable for the spline model, and the use of polynomial functions across columns interbank be explained global trend. Using the variogram model space and residual plots to reflect this trend.
2.3 variance parameters and model comparison
Variance parameter limit (or reduce or residual) Maximum Likelihood (REML) ASREML 3.0 is estimated (Gilmour et al 2009). Taylor series expansion method using the standard error of the estimate. To study the validity of the model with respect to the use of space-space model no effect, is always included in the final model blockiness only additive effect, whereas in the extended model, all other non-significant variance parameters (such as random effects columns and rows effect) are deleted from the fitted model , in addition to blocking effect and additional effects. Value of zero the boundary parameters, using a one-tailed LRT importance parameter is determined to a given; otherwise, two-tailed test. When the space for component R, in some cases, it is not easy to converge. Then tried several strategies to achieve convergence of: (1) update using the update multiple ASREML-R function; and (2) attempting to start a lower autocorrelation; (3) into the spatial component of the random effect.
Genotype effect calculated prediction accuracy of each parent and progeny of breeding value (the correlation between the actual value and the predicted genetic) or clones
wherein, PEV prediction error variance is estimated additive genetic variances. If cloning experiments, is genotypic variation estimate.
Relative genetic gain is estimated 100 (gs-gb) / gb , gs and gb are, respectively, where the use of space and the genetic basis of the expected response to an individual model selection ratio (Costa E Silva et al 2001). The genotype of clones breeding value or estimated values, the ratio is 20% prior to selecting parent before progeny or genotypic value of 5%. Gs and Gb calculated as an average value or genotype breeding clones. Sipierman calculated correlation coefficients, comparing the value of the base model and breeding space model.
Reproduced in: https: //www.jianshu.com/p/109acbb0decf