first of all you should have to arrange your data according to the number of environments you can get from this series experiments. as you mentioned you have two seasons and three locations which can compose six different environments ( 2x 3). then you can run normal analysis of variance ANOVA for regression to estimate Eberhart and Russel stability parameters, or you can run also AMMI analysis (based on principal component analysis) using appropriate software programs as Genstat or SAS. from these analyses you can determine different stability parameters depending on the objectives of your experiment, whether you need just to show the stability of genotypes across environments or also you need to show the adaptations of genotypes to the different environments (clustering), and to quantify the interaction of genotypes with environments, etc..
you have two years and three locations. So you have 6 environments. If every environment owns the same genotypes you can regress yield of each genotype against field average yield (calculated as the mean on the genotypes). The linear regression coefficient (b) is the (in)stability coefficient (Finlay and Wilkinson). Plotting b's against yield, for all varieties, identifies genotypes suitable for best environment (high yield and moderate stability) or for poor environment (good yield and hight stability).
If you also consider CV associated with the estimates of b's it is possible to separate reliable varieties (with low CV) from not reliable ones (those with high CV of stability coefficient b).
First, I think what type of experiment it is needs to be clear, e.g. breeding or agronomic trial, single or factorial and the experimental design used. Were the two seasons similar in terms of rainfall frequency and distribution and temperature? Are the locations (environments), agroecology and soil type different? If someone is clear with these variables, first they may conduct homogeneity or heterogeneity test for error variances.
Based on the result of the test for homogeneity of variance you can apply the appropriate analysis of variance. If heterogeneity of variance is indicated, choose an appropriate data transformation that can stabilize the error variances and compute the pooled analysis of variance based on the transformed data. If heterogeneity of error variance is not observed, you can run a pooled analysis of variance based on original data using appropriate statistical software.
The main objective of a combined analysis of data over years is to identify technologies whose average effect over years or seasons is stable and high. The interaction between treatment and year has no agronomic meaning, and is much less relevant than the interaction between treatment and season.
Data from a series of experiments at different locations can be analyzed together at the end of each crop season to evaluate the treatment by location interaction effect and the average effects of the treatments over homogeneous locations. These effects are the primary basis for identifying the best performers among the different technologies tested. Overall, the type and objectives of the experiment, and the data generated may guide you how to arrange your data and which type of statistical analysis to run.