Two versions of the “reliability ensemble averaging” (REA and REA Xu) method and an unweighted mean were used to generate multi‐model ensembles of 14 general circulation models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5) for a baseline (1971–2000) period and future scenarios (RCP4.5 and RCP8.5) for the 21st century. To test the consistency of the REA ensembles at different scales, one ensemble was area‐averaged at regional scale and two were obtained at 50‐km gridpoints. Climatic metrics of temperature and precipitation during the baseline period served to evaluate the performance of the GCMs and the ensembles in the North American monsoon (NAM) and the mid‐summer drought (MSD) regions. The metrics of the three REA ensembles are very similar among them in each region and show a better performance than the unweighted multi‐model ensemble (U‐MME). The majority of the GCMs overestimate winter precipitation in the NAM and fail to retract the end of the monsoon in autumn; REA ensembles reduce this overestimation. All ensembles capture well the MSD's double peak of rainfall, but underestimate summer precipitation. According to all ensembles, temperature increases of 1.5–2 °C in the two regions may be reached between 2035 and 2055 relative to the baseline, and by 2070–2099 temperature (precipitation) in Mexico may increase (decrease) between 2° C (5%) and 5.8 °C (10%) in the RCP4.5 and RCP8.5 scenarios, respectively. Annual changes of precipitation show a north (positive) and south of 35°N (negative) pattern. The largest impacts are expected during summer with a possible decrease of ~13% (up to −1.5 mm/day), especially in southern Mexico, Central America, and the Caribbean, while autumn precipitation may slightly increase. Future projections from the REA Xu (by gridpoint) ensemble show spatial patterns similar to the U‐MME, but with more regional detail, which could be an added value for regional climate impact studies.

Trends of 15 extreme and non‐extreme daily rainfall indices for Mexico and the southern United States are analyzed for the extended boreal winter (November–March) and summer (June–October) during 1981–2010. The analysis includes gridded observations (CHIRPS and Livneh), reanalyses (ERA5, ERA‐Interim, MERRA2, and CFSR), the regional climate model RegCM4.7 (forced by ERA‐Interim), and four general circulation models (GCMs) from the CMIP5 (HadGEM2‐ES, CNRM‐CM5, MPI‐ESM‐MR, and GFDL‐ESM2M). The domain is divided into 13 regions with similar seasonal rainfall patterns according to the k‐means algorithm applied to CHIRPS monthly precipitation, as a reference. Gridded observations and reanalyses (O&R) show similar regional annual cycles and metrics of precipitation. The two observational data sets show the largest differences in the magnitude of some indices in the wettest region (southeastern Mexico) and in the monsoon. During summer, O&R show positive significant trends in extreme rainfall indices in regions commonly affected by tropical cyclones (southern Baja California Peninsula, northeastern Mexico, and the southeastern United States), while negative significant trends of seasonal accumulation of rainfall are seen in the Mediterranean Californias and the Sonoran Desert region. There is an agreement among most data sets that the frequency and intensity of extreme winter precipitation indices have declined in northwest Mexico and the southwestern United States, but trends are not statistically significant, while southeastern Mexico shows significant increases in 5‐days accumulation (rx5d). ERA5 followed by MERRA2 show the best performance. RegCM4.7 reproduces the annual cycle of precipitation in most of the regions and captures the main trend signals especially in winter, but it has difficulties in reproducing the summer interannual variability of few indices. The GCMs tend to overestimate (underestimate) the annual cycle of precipitation and the median values of extreme indices in the driest (wettest) regions. They are not able to capture the regional trends of the indices.