Performance of alternative spatial models in empirical Douglas-fir and simulated datasets
Based on an empirical dataset originating from the French Douglas-fir breeding program, we showed that the bidimensional autoregressive and the two-dimensional P-spline regression spatial models clearly outperformed the classical block model, in terms of both goodness of fit and predicting ability. In contrast, the differences between both spatial models were relatively small. In general, results from simulated data were well in agreement with those from empirical data.
Environmental (and/or non-environmental) global and local spatial trends can lead to biases in the estimation of genetic parameters and the prediction of individual additive genetic effects.
The goal of the present research is to compare the performances of the classical a priori block design (block) and two different a posteriori spatial models: a bidimensional first-order autoregressive process (AR) and a bidimensional P-spline regression (splines).
Data from eight trials of the French Douglas-fir breeding program were analyzed using the block, AR, and splines models, and data from 8640 simulated datasets corresponding to 180 different scenarios were also analyzed using the two a posteriori spatial models. For each real and simulated dataset, we compared the fitted models using several performance metrics.
There is a substantial gain in accuracy and precision in switching from classical a priori blocks design to any of the two alternative a posteriori spatial methodologies. However, the differences between AR and splines were relatively small. Simulations, covering a larger though oversimplified hypothetical setting, seemed to support previous empirical findings. Both spatial approaches yielded unbiased estimations of the variance components when they match with the respective simulation data.
In practice, both spatial models (i.e., AR and splines) suitably capture spatial variation. It is usually safe to use any of them. The final choice could be driven solely by operational reasons.
KeywordsGlobal and local spatial trends Forest genetics trials Autoregressive residual Two-dimensional P-splines
The authors sincerely acknowledge Jean-Charles Bastien for his help in identifying trials and accessing data. Thanks go to the staff of INRA experimental units (UE GBFOR, INRA Val de Loire) who have established, maintained, and assessed the field trials.
Eduardo P Cappa, F. Muñoz, and L. Sánchez received funding from the European Union’s Seventh Framework Program for research, technological development, and demonstration under grant agreement no. 284181 (“Trees4Future”). F. Muñoz is partially funded by research grant MTM2016-77501-P from the Spanish Ministry of Economy and Competitiveness.
Compliance with ethical standards
Conflict of interest
The authors declare that they have no conflict of interest.
- Anekonda TS, Libby WJ (1996) Effectiveness of nearest neighbor data adjustment in a clonal test of redwood. Silvae Genet 45(1):46–51Google Scholar
- Cressie N (1993) Statistics for Spatial Data. Wiley series in probability and statistics. Wiley, New YorkGoogle Scholar
- Hamann A, Koshy M, Namkoong G (2002) Improving precision of breeding values by removing spatially autocorrelated variation in forestry field experiments. Silvae Genet 51:210–215Google Scholar
- Henderson CR (1984) Applications of linear models in animal breeding. University of Guelph, Guelph, Ont, CanadaGoogle Scholar
- Joyce D, Ford R, Fu YB (2002) Spatial patterns of tree height variations in a black spruce farm-field progeny test and neighbors-adjusted estimations of genetic parameters. Silvae Genet 51:13–18Google Scholar
- Lopez GA, Potts BM, Dutkowski GW, Apiolaza LA, Gelid P (2002) Genetic variation and inter-trait correlations in Eucalyptus globulus base population trials in Argentina. For Genet 9:223–237Google Scholar
- Misztal I (1999) Complex models, more data: simpler programming. Proc Inter Workshop Comput Cattle Breed ‘99, March 18-20, Tuusala, Finland. Interbull Bul. 20:33-42Google Scholar
- Muñoz F, Sanchez L (2015) breedR: statistical methods for forest genetic resources analysts. R package version 0.7–16. https://github.com/famuvie/breedR
- Saenz-Romero C, Nordheim EV, Guries RP, Crump PM (2001) A case study of a provenance/progeny test using trend analysis with correlated errors and SAS PROC MIXED. Silvae Genet 50:127–135Google Scholar
- Thomson AJ, El-Kassaby YA (1988) Trend surface analysis of provenance-progeny transfer data. Can J For Res 18: 515–520Google Scholar
- Velazco JG, Rodríguez-Álvarez MX, Boer MP, Jordan DR, Eilers PH, Malosetti M, van Eeuwijk FA (2017) Modelling spatial trends in sorghum breeding field trials using a two-dimensional P-spline mixed model. Theor Appl Genet 130:1375–1392. https://doi.org/10.1007/s00122-017-2894-4 CrossRefPubMedPubMedCentralGoogle Scholar
- Verbyla AP, Cullis BR, Kenward MG, Welham SJ (1999) The analysis of designed experiments and longitudinal data by using smoothing splines (with discussion). Appl Stat 48:269–311Google Scholar