ConvectiveIndices.jl

A Julia package for calculating convective indices (e.g. CAPE) from atmospheric sounding data.

The core functions are calc_dilute_CAPE and calc_CAPE_thetae. The latter outputs parameters such as CAPE, Lifted Index and CIN from input columns of pressure, temperature, specific humidity and geopotential height, for a user-defined parcel (surface/most unstable, with/without vertical mixing).

help?> calc_CAPE_thetae
search: calc_CAPE_thetae

calc_CAPE_thetae (ps[hPa], tks[K], qs[kg/kg], zs[m], parcel, dp_mix[hPa], dp_intp[hPa], kiss)

Calculate convective indices such as CAPE and CIN for a parcel of choice (surface/most unstable, 
with/without vertical mixing). 

  Examples
  ≡≡≡≡≡≡≡≡≡≡

julia> # most unstable parcel, mixed over 50 hPa (default)
LI, CAPE, CIN = calc_CAPE_theta(ps,tks,qs,zs) 
  (-8.94582333506736, 1613.7159227760612, 327.257167221434))
julia> # surface parcel, not mixed
LI, CAPE, CIN = calc_CAPE_theta(ps,tks,qs,zs, parcel = 2, dp_mix = 0)
  (-12.416924139871522, 2428.182537242374, 85.85516940477973) 
julia>  # full calculations
LI, CAPE, CIN, pLCL, zBCL, CAPECIN_ALCL, CIN_LCL, MRH_ALCL, MRH1, MRH2 = calc_CAPE_thetae(ps,tks,qs,zs, FULL = 1)
  (-8.94582333506736, 1613.7159227760612, 327.257167221434, 936.6429885118564, 1230.0, -189.68905798724995, 128.5705360872618, 69.90722164805184, 56.290565968008316, 30.494525283693054)


OUTPUT for FULL=1:

Lifted Index [°C], CAPE [J/kg], CIN [J/kg], pLCL [hPa], Buoyant Condensation Level [m], CAPE-CIN above the LCL [J/kg], CIN below LCL [J/kg], MRH (mean RH%) above the LCL [%], MRH 600-800 hPa, MRH 300-600 hPa

INPUT: (N-element ARRAYs) ps,tks,qs,zs = vertical profiles of pressure, temperature, specific humidity and geopotential height

  OPTIONAL keyword arguments:

  parcel = 1 : the most unstable parcel in the lowest 350 hPa (default)

  parcel = 2 : surface parcel, or parcel from the lowest level

  dp_mix = 0...100 : pressure layer depth [hPa] for mixing the source parcel (default 50, use 0 for no mixing)

  dp_intp = 5 linearly interpolate to a uniform pressure grid with resolution dp (default 5). Use 0 to skip. This is a lot faster, but disables mixing and FULL option, and is not recommended for low-resolution input.

  FULL = 1: Full calculations to include also less known convective predictors, which were found useful in [1].

This routine uses a equivalent potential temperature formulation for all indices (similarly to ECMWF CAPE), avoiding vertical loops altogether. This results in larger absolute values (e.g. for CAPE, 30% larger) than classic computations, but in no worse correlation with observed convection [1].

NOTE: Default option parcel=1 and dp_mix=50 corresponds to a hybrid mixed-layer most-unstable parcel similar to the one used by ECMWF. The MLMU-Lifted Index was the overall thunderstorm index for Europe in [1].

  [1] Ukkonen and Mäkelä (2019): Evaluation of machine learning classifiers for predicting deep convection, JAMES.

The code is fast yet easy to read and modify thanks to Julias language design. Processing 6.2 million reanalysis pseudosoundings on an 8-core CPU using an expensive interpolation option:

@time @sync @distributed for i = 1:nlon
               for j = 1:nlat
                       for k = 1:ntime
                               tks = tk[i,j,:,k]; qs = q[i,j,:,k]; ps = p[i,j,:,k]; zs = z[i,j,:,k];
                                LI[i,j,k],CAPE[i,j,k],CIN[i,j,k],PLCL[i,j,k],ZBCL[i,j,k],CAPECIN_ALCL[i,j,k], CIN_LCL[i,j,k],
                                MRH_ALCL[i,j,k],MRH1[i,j,k],MRH2[i,j,k]  
                                  = calc_CAPE_thetae(ps,tks,qs,zs,dp_interp=2,FULL=1);
                       end
               end
       end
208.778669 seconds (3.02 M allocations: 148.865 MiB, 0.03% gc time)