Mechanisms of diurnal precipitation over the US Great Plains: a cloud resolving model perspective

Abstract

The mechanisms of summertime diurnal precipitation in the US Great Plains were examined with the two-dimensional (2D) Goddard Cumulus Ensemble (GCE) cloud-resolving model (CRM). The model was constrained by the observed large-scale background state and surface flux derived from the Department of Energy (DOE) Atmospheric Radiation Measurement (ARM) Program’s Intensive Observing Period (IOP) data at the Southern Great Plains (SGP). The model, when continuously-forced by realistic surface flux and large-scale advection, simulates reasonably well the temporal evolution of the observed rainfall episodes, particularly for the strongly forced precipitation events. However, the model exhibits a deficiency for the weakly forced events driven by diurnal convection. Additional tests were run with the GCE model in order to discriminate between the mechanisms that determine daytime and nighttime convection. In these tests, the model was constrained with the same repeating diurnal variation in the large-scale advection and/or surface flux. The results indicate that it is primarily the surface heat and moisture flux that is responsible for the development of deep convection in the afternoon, whereas the large-scale upward motion and associated moisture advection play an important role in preconditioning nocturnal convection. In the nighttime, high clouds are continuously built up through their interaction and feedback with long-wave radiation, eventually initiating deep convection from the boundary layer. Without these upper-level destabilization processes, the model tends to produce only daytime convection in response to boundary layer heating. This study suggests that the correct simulation of the diurnal variation in precipitation requires that the free-atmospheric destabilization mechanisms resolved in the CRM simulation must be adequately parameterized in current general circulation models (GCMs) many of which are overly sensitive to the parameterized boundary layer heating.

Appendix: Sensitivity tests

Sensitivity experiments were conducted to examine the impact of the domain size and the horizontal resolution changes on the simulated diurnal precipitation variability. For simplicity, we selected the idealized case of EXP1 where the model tends to generate both the daytime and nighttime precipitation from the imposed diurnally varying large-scale advection and surface heat flux. We tested the GCE model in four different settings with different domain sizes and horizontal resolutions, which are summarized in Table 2. Each experiment was run for 20 days for a quick evaluation and we use the last 15 days to examine the diurnal variation. Due to relative short-time integration period, the precipitation frequency exhibits clearer diurnal variation than the precipitation amount, which we compare in Fig. 11. These frequency statistics are stable when we differ the averaging period. The results show that increasing the horizontal resolution from 1 km to 250 m (EXP1a) does not modify the simulated diurnal cycle of precipitation significantly (compare Fig. 11a with EXP1 in Fig. 4). The domain size seems to have a bigger impact on the phase (timing of the peak) of the diurnal convection. At a domain size of 256 km (EXP1b, Fig. 11b) the results are quite similar to those from the 128-km domain (EXP1). When, however, the GCE domain is extended to 512 km (EXP1c, Fig. 11c), the diurnal peaks tend to be delayed in time both in the daytime and the nighttime precipitation. The delay is largest for the daytime precipitation with the peak shifted into late evening. When we both extend the domain to 512 km and reduce the grid spacing to 250 m, the sensitivity is largest with only a late evening peak (EXP1d, Fig. 11d).

Table 2 A description of the sensitivity experiments to the domain size and the horizontal resolution

Fig. 11

Diurnal variations of the frequency for the precipitation events larger than 1 mm day⁻¹ in the four sensitivity experiments a EXP1a b EXP1b c EXP1c, and d EXP1d. The results are calculated for the last 15 days from the total 20-day integrations

A late evening peak in a bigger domain is intriguing whether it is driven by boundary layer heating or free-atmospheric large-scale advection. To address this issue, we again select a single storm in this case (EXP1d) and examine its temporal evolution following the storm center (Fig. 12). The result is quite consistent with the daytime convection in EXP1 (as shown in Fig. 6). In this case, the simulation has slower transition from shallow to deep convection (Fig. 12c). Surface precipitation (Fig. 12b) reaches its maximum intensity after maximum development of cloud (Fig. 12a) associated with deep convection penetrating through the PBL. We found a similar high-cloud destabilization process in the developing early morning convection in the largest domain run of EXP1c (not shown). Therefore, we conclude that the fundamental mechanisms for diurnal convection that we identified in the control simulations are qualitatively consistent with the experiments with larger domains and finer resolution. This is consistent with the sensitivity tests done with the 2D GCE model by Johnson et al. (2002, see their Fig. 12).

Fig. 12

Life cycle of the storm developed in EXP1d. a The total hydrometeors (g kg⁻¹) b the surface precipitation (mm day⁻¹), and c the vertical motion (cm s⁻¹). Others are same as in Fig. 6a–c

The impact of the domain size and the resolution on the individual storm can be understood better by comparing the Hovmuller plots of precipitation (Fig. 13). The increase of domain and resolution in general tends to increase the lifetime of individual storms that propagate more slowly. This explains the delay in the peak of the diurnal precipitation (Lang et al. 2007). Both in the small and large domain cases, precipitation develops regularly on a diurnal basis, presumably due to the imposed, regular diurnal forcing. We note that, even for the small domain, the cyclic lateral boundary condition in the GCE does not give any unrealistic influence on the diurnal cycle due to the relatively short durations of the simulated storms.

Fig. 13

Hovmuller plots of surface precipitation (mm h⁻¹) from day 5–15 in a EXP1d and b EXP1

Although we examine the sensitivity for the idealized case, we do not expect drastic changes for the case of more realistic forcing, or for the case of 3D experiments. Khairoutdinov and Randall (2003) tested their CRM with the ARM IOP 1997 single-column model forcing and found that the simulations are rather insensitive to changes in the domain size and the horizontal resolution when these are varied over a wide range. They further indicated that the overall effects of increasing the model dimension from 2D into 3D are minor in terms of the evolution of the simulated domain-mean fields such as precipitation and total precipitable water (see their Fig. 2). They did, however, find more rapid temporal fluctuations in the 2D model compared with the 3D counterpart (Grabowski et al. 1998; Tompkins 2000). They suggested that continuous, strong large-scale forcing in the CRM simulation might constrain the model too much to reveal any model differences.