Transition from Shallow to Deep Convection and Its Connection to the Environmental Conditions across the Amazonia

Understanding and modeling atmospheric convection remains challenging because clouds form and organize at a wide range of spatial and temporal scales and depend on complex interactions between surface and atmospheric dynamics and thermodynamics. Moreover, long-term, high-resolution observations are scarce, particularly in the tropics, hindering our understanding of the shallow to deep (STD) convective transition. This explains, in part, why climate models still fail to represent the diurnal cycle of precipitation. One way to better understand the physical mechanisms responsible for the development and organization of convection is to use atmospheric models at cloud-resolving spatial resolutions. However, inferences from such models require validation with observations.

In this study, we perform high-resolution simulations of the STD transition over the central Amazon rainforest using the Advanced Research WRF (WRF-ARW) regional model at 9 km, 3 km, and 1 km resolutions. We used the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis data (ERA-5) to initialize the model, with lateral boundary conditions updated every hour. The model was run for December 2014, and observations during the GoAmazon 2014/15 experiment were used for model validation.

Our results show that the WRF model was able to reproduce the diurnal cycle of surface meteorological variables, such as specific humidity, temperature, wind speed, heat flux, and vertical profile of relative humidity, temperature, and wind speed and directions. However, some minor discrepancies in their magnitudes persist even at higher resolutions. For instance, the model overestimates the domain average precipitation by 0.263 mm/hr and exhibits an early peak (11h LT) compared to observation (13h LT). Overall, we found that enhancing the spatial resolution of the model leads to improved results. One illustrative example is surface heat flux (SH), a key factor in the formation of locally generated deep convective clouds. The bias in monthly average SH at T3 site during noon is improved from 44.32 W/m^2 to 39.47 W/m^2 when increasing the spatial resolution from 9 km to 1 km. The improvement can be attributed to the model's ability to identify distinct vegetation types accurately as we enhance the grid resolution.

Finally, we identified days with isolated deep convective cloud development in our simulation following the criteria proposed by Tian et al. (JGR, 2021): 1) excluding the influence of mesoscale convective systems within the analysis domain, and 2) considering clouds with a base height below 3 km, top height exceeding 8 km, and precipitation exceeding 0.015 mm/hr. Analysis of surface meteorology reveals that local deep convective clouds are generally associated with a lower surface temperature, higher surface humidity, and smaller surface wind speed. Moreover, precipitable water vapor has a strong correlation with deep convective clouds.