Logistic Regression of Surface-Based Observations with Large-Scale Atmospheric Moisture onto Occurrence of Convective Precipitation in the North American Monsoon

Abstract:

This research focuses on the North American Monsoon (NAM) climate region, a semi-arid to arid region encompassing northwestern Mexico and the southwestern USA. In the NAM, intense summertime insolation over the Sonoran Desert lowers surface pressure, thereby inducing a seasonal influx of moist air primarily from the Gulf of California, with smaller contributions from the eastern Pacific and the Gulf of Mexico, as well as mixing over the elevated regions of the Sierra Madre Occidental. The seasonal influx of moist air, coupled with orographic uplift, triggers intense convective storms typically initiating over higher elevation and propagating as organized mesoscale convective systems over the lower deserts given conducive flow conditions or upper-level disturbances. Although a few prior studies have pointed to local land-atmosphere coupling affecting convective rainfall in the NAM, the topic remains largely unexplored. Using site-level eddy covariance and meteorological station data coupled with MODIS satellite measurements of atmospheric water vapor, a logistic regression approach is applied to quantify the marginal contributions of antecedent local surface moisture and energy fluxes to the triggering of convective precipitation in the presence of large-scale forcing. Overall, the results indicate that measures of surface moisture conditions or fluxes and of atmospheric water vapor content are more predictive of subsequent convective rainfall when modeled jointly than individually. Furthermore, while surface moisture is positively predictive, suggesting a positive feedback of the surface moisture conditions on rainfall, its predictive strength is weaker than for atmospheric moisture.. The relative contributions of surface-derived and large-scale atmospheric moisture to subsequent rainfall are shown to vary seasonally and with location. Finally, the inclusion of additional surface-level predictors to the model does not substantially improve the fit.