Convection

Convection is the atmospheric process of rising motion because of positively buoyant air parcels compared to its surroundings. In hydrostatic models like the primitive equation model in SpeedyWeather.jl convection has to be parameterized as the vertical velocity is not a prognostic variable that depends on vertical stability but rather diagnosed to satisfy horizontal divergence. Convection can be shallow and non-precipitating denoting that buoyant air masses can rise but do not reach saturation until they reach a level of zero buoyancy. But convection can also be deep denoting that saturation has been reached during ascent whereby the latent heat release from condensation provides additional energy for further ascent. Deep convection is therefore usually also precipitating as the condensed humidity forms cloud droplets that eventually fall down as convective precipitation. See also Large-scale condensation in comparison.

Convection implementations

Currently implemented are

using SpeedyWeather
subtypes(SpeedyWeather.AbstractConvection)

3-element Vector{Any}:
 BettsMillerConvection
 BettsMillerDryConvection
 ConvectiveHeating

which are described in the following. See Parameterizations for the general parameterization interface these implement.

Simplified Betts-Miller convection

We follow the simplification of the Betts-Miller convection scheme (Betts, 1986; Betts and Miller, 1986) as studied by Frierson (2007). The central idea of this scheme is to represent the effect of convection as an adjustment towards a (pseudo-) moist adiabat reference profile and its associated humidity profile. Meaning that conceptually for every vertical column in the atmosphere we

1. Diagnose the vertical temperature and humidity profile of the environment relative to the adiabat up to the level of zero buoyancy.

2. Decide whether convection should take place and whether it is deep (precipitating) or shallow (non-precipitating).

3. Relax temperature and humidity towards (corrected) profiles from 1.

Reference profiles

The dry adiabat is

The temperature of an air parcel at pressure is determined by the temperature it had at pressure (this can be surface but it does not have to be) and the gas constant for dry air   and the heat capacity  . The pseudo adiabat follows the dry adiabat until saturation is reached (the lifting condensation level, often abbreviated to LCL),  . Then it follows the pseudoadiabatic lapse rate

with gravity , heat capacity , the saturation specific humidity of the parcel (which is its specific humidity given that it has already reached saturation), latent heat of vaporization , dry gas constant , water vapor gas constant , and Virtual temperature . Starting with a temperature and humidity   at the lifting condensation level temperature aloft changes with    between two layers separated in geopotential apart. On that new layer, is recalculated as well as the virtual temperature   . is derived from the ratio of dry to vapor gas constants see Virtual temperature. Note that the pseudoadiabatic ascent is independent of the environmental temperature and humidity and function of temperature and humidity of the parcel only (although that one starts with surface temperature and humidity from the environment). Solely the level of zero buoyancy is determined by comparing the parcel's virtual temperature to the virtual temperature of the environment at that level. Level of zero buoyancy is reached when   but continues for   which means that the parcel is still buoyant. Note that the virtual temperature includes the effect that humidity has on its density.

The (absolute) temperature a lifted parcel has during ascent (following its pseudoadiabat, dry and/ or moist, until reaching the level of zero buoyancy) is then taken as the reference temperature profile that the Betts-Miller convective parameterization relaxes towards as a first guess (with a following adjustment as discussed below). The humidity profile is taken as    with a parameter (default  ) of the scheme (Simplified Betts-Miller, SBM) that determines a constant relative humidity of the reference profile, applied to the saturation humidity evaluated at the reference temperature of that layer.

An illustration of adiabat reference line is below

Entrainment

A real convective plume mixes with the environmental air around it as it rises, which dilutes the parcel's temperature and moisture excess, reducing its buoyancy and moisture content and generally shallowing convection compared to the undiluted ascent described above. The scheme optionally applies this dilution at every level during the ascent, after the dry or moist adiabatic step:

with parcel temperature/humidity , environmental temperature/humidity at that level , and an entrainment rate   that can vary by level. Because is a mixing fraction per model level rather than a rate per unit height, the resulting profile depends on the vertical resolution (nlayers): the same AbstractEntrainment profile mixes in more environmental air over a coarser column, since each level spans a larger part of the atmosphere. In the moist branch of the ascent, mixing happens after condensation at that level, so the saturation adjustment (and thus the LZB and the associated at that level) already reflects the diluted parcel; the latent heat implied by the change in from mixing itself is not separately accounted for.

Available profiles, selected via the entrainment option of BettsMillerConvection or BettsMillerDryConvection:

subtypes(SpeedyWeather.AbstractEntrainment)

3-element Vector{Any}:
 ConstantEntrainment
 LinearEntrainment
 NoEntrainment

• NoEntrainment (the default for BettsMillerDryConvection):   everywhere, reproducing the scheme exactly as described above with no dilution.

• LinearEntrainment (the default for BettsMillerConvection, with surface_entrainment = 0.5 and σ_entrainment = 0.5): ramps linearly from surface_entrainment at the surface ( ) to zero at σ_entrainment, and is zero above that level.

• ConstantEntrainment: a single entrainment_rate applied at every level.

The wet scheme therefore entrains by default; pass entrainment = NoEntrainment() to BettsMillerConvection to recover the undiluted scheme. The dry scheme defaults to NoEntrainment and can opt in the same way, e.g. BettsMillerDryConvection(spectral_grid; entrainment = LinearEntrainment(spectral_grid)).

First-guess relaxation

With the Reference profiles obtained, we relax the actual environmental temperature and specific humidity in the column

with the second parameter of the parameterization, the time scale . Note that because this is a first-guess relaxation, these tendencies are not actually the resulting tendencies from this scheme. Those will be calculated in Corrected relaxation.

Note that above the level of zero buoyancy no relaxation takes place   , or, equivalently  ,   there. Vertically integration from the level of zero buoyancy to the surface (so that  , as  ) yields

is the precipitation in units of due to drying (as a consequence of the humidity tendency) and is the precipitation in the same units due to warming (as resulting from temperature tendencies). Note that they are the vertically difference between current profiles and the references profiles, so if   this would mean that a convective adjustment to would release humidity from the column through condensation, but can also be negative. Consequently similar for .

Convective criteria

We now distinguish three cases

1. Deep convection when   and  

2. Shallow convection when   and  

3. No convection for  .

Note that to evaluate these cases it is not necessary to divide by in the first-guess relaxation, neither are the and necessary to multiply during the vertical integration as all are positive constants. While this changes the units of they can be reused in the following.

This is illustrated in the following

Deep convection

Following Frierson (2007), in order to conserve enthalpy we correct the reference profile for temperature   so that  .

is the pressure difference  . The terms inside the integral are rearranged compared to Frierson (2007) to show that the vertical integral in First-guess relaxation really only has to be computed once.

Shallow convection

In the following we describe the "qref" scheme from Frierson (2007), which corrects reference profiles for both temperature and humidity to guarantee that  , i.e. no precipitation during convection. In that sense, shallow convection is non-precipitating. Although shallow convection is supposed to be shallow we do not change the height of the convection and keep using the determined during the calculation of the Reference profiles.

This scheme is non-precipitating ( ) as derived below. follows from the vertical integral of the moisture tendency , which in this scheme is a relaxation towards (see Corrected relaxation).

Inserting from above yields

The integral becomes

So this scheme is indeed non-precipitating, i.e.  .

Corrected relaxation

After the reference profiles have been corrected in Deep convection and Shallow convection we actually calculate tendencies from

with   as default.

Convective precipitation

The convective precipitation results then from the vertical integration of the tendencies, similar to Large-scale precipitation.

The integral is taken with sign over all layers: layers that moisten are fed by the water removed from layers that dry, so is the net column drying and the column water budget (precipitation equals the loss of column water vapour) is closed exactly. Clamping each layer's contribution to be non-negative instead would count the water that moistens a layer twice, once as vapour and once as precipitation (see #1275). The result is additionally multiplied by a deep-convection indicator so that   for shallow convection, where the qref correction already makes the column-integrated vanish up to rounding.

In the shallow convection case   due to the correction (and the indicator above) even though in the first guess relaxation   was possible, but for deep convection   by definition.

Convective snow

If the temperature of the lowermost model layer is below a freezing_threshold (default  ), all of falls as snow instead of rain for that column and time step, controlled by the snow option (true by default). This is a single check on the layer the parcel starts its ascent from, unlike Large-scale condensation's falling-flux melt cascade through intermediate layers: convective precipitation is generated and deposited within the same time step rather than fluxed layer by layer, so there is no intermediate layer for it to fall through and potentially melt in. The phase swap conserves total precipitation mass (rain + snow is unchanged, only the partitioning between the two is affected), and shallow convection produces neither since   there regardless of temperature. With snow = false all convective precipitation is always rain, recovering the previous behaviour.

Dry convection

In the primitive equation model with humidity the Betts-Miller convection scheme as described above is defined. Without humidity, a dry version reduces to the Shallow convection case. The two different shallow convection schemes in Frierson (2007), the "shallower" shallow convection scheme and the "qref" (as implemented here in Shallow convection) in that case also reduce to the same formulation. The dry Betts-Miller convection scheme is the default in the primitive equation model without humidity.