I will suggest you to use Design Expert. http://www.statease.com/dx9.html. Using this tool, you can regress your parameters. At the mean time you can perform anova with multiple variables.

Thank you very much for your advice Shaukat.

probably there are a lot of way to test for the GxE interaction. however, an easy way to look at this (i am assuming that the locations are random) is to find the population average of each genotype within each location, rank the genotype within each location. if the ranking of genotypr differ from location to location, we will say that there is a significant GxE interaction. this approach is easy because in most solfware that you will, since GxE is random, they will not provide the test for GxE or you can do by hand.

i hope this will help. 

here, i will add two more way that can help you test for the GxE interaction.

• the first one a formal Likelihood Ratio Test (LRT)

You can also perform a likelihood ratio test as follows

H0: the model without GxE is better (reduce model)

H1: model with GxE better fits our data (the full model is better)

this will be a chi square test where the degree of freedom will be the difference in the number of parameter estimated in both models.

p-value = 1- pchisquare(< difference betseen likelihoods), and the p-value will be compared to the alpha level.

• This is not a formal

Here, you just have to see if the variance component of GxE is different from zero. if its different from zero, you can conclude that there is a GxE interaction.

I hope this idea can help you.

Use 'statistical significance' as just one piece of information, not the entire basis to make inferences. More exciting, and practically more useful, will be the information on the nature (crossover or non-crossover) and magnitude of GxE interaction. Once you have your GxE BLUPs, one way to quantify these may be to use of Pattern Analysis. The link below provides some notes that may be helpful

https://www.researchgate.net/publication/270157426_Classification_and_Ordination_in_Pattern_Analysis_of_Genotype_x_Environment_Interactions

Chapter Classification and Ordination in Pattern Analysis of Genotyp...

I suggest, to go to the roots of variation statistics, to sound farmers logic and have in mind, that significance is just a convention. I would proceed in two steps. 1. estimate in each environment independently the genotype effects and the error variance of  genotypes (excluding the checks). 2. Calculate from the obtained genotype effects across the 8 environments a one-way ANOVA and estimate the residual variance, which by simple logic should be the sum of GxE variance and error variance. Subtract the pooled error variance, obtained in step 1 from the 8 environments, from the residual ANOVA variance, to obtain a oractical estimate of the interaction variance. Keep in mind, that the plausibility depends on the appropriateness of the additive yield model. If the yield in the environments differs very much, a multilicative yield model might be better.