Merge #105

105: Sa/diagnostics debug and minor improvements r=sajjadazimi a=sajjadazimi



Co-authored-by: Sajjad Azimi <[email protected]>

Project.toml

@@ -1,7 +1,7 @@
 name = "Kinematic1D"
 uuid = "a8a3bb05-6cc5-4342-abf6-5cc636bd2b35"
 authors = ["Climate Modeling Alliance"]
-version = "0.3.0"
+version = "0.3.1"
 
 [deps]
 ClimaCore = "d414da3d-4745-48bb-8d80-42e94e092884"
@@ -30,7 +30,7 @@ ClimaCorePlots = "0.2"
 CloudMicrophysics = "0.10"
 DiffEqBase = "6.75"
 Distributions = "0.25"
-EnsembleKalmanProcesses = "0.13"
+EnsembleKalmanProcesses = "1.1"
 Interpolations = "0.14"
 JLD2 = "0.4"
 NCDatasets = "0.12"

src/calibrateKiD/Diagnostics.jl

@@ -10,21 +10,21 @@ Returns liquid water path [kg/m^2]
 """
 function lwp(vec::Vector{FT}, config) where {FT <: Real}
 
-    @assert issubset(Set(["rho", "rl"]), Set(config["observations"]["data_names"]))
+    @assert issubset(Set(["rho", "ql"]), Set(config["observations"]["data_names"]))
 
     (n_c, n_v, n_z, n_t) = get_numbers_from_config(config)
     n_vht = n_v * n_z * n_t
     dz = (config["model"]["z_max"] - config["model"]["z_min"]) / n_z
     lwp = zeros(n_t, n_c)
     ind_ρ = findall(x -> x == "rho", config["observations"]["data_names"])[1]
-    ind_rl = findall(x -> x == "rl", config["observations"]["data_names"])[1]
+    ind_ql = findall(x -> x == "ql", config["observations"]["data_names"])[1]
 
     for i in 1:n_c
         v_ = vec[((i - 1) * n_vht + 1):(i * n_vht)]
         m_ = reshape(v_, n_v * n_z, n_t)
         ρ = m_[((ind_ρ - 1) * n_z + 1):(ind_ρ * n_z), :]
-        rl = m_[((ind_rl - 1) * n_z + 1):(ind_rl * n_z), :]
-        lwp[:, i] = sum(rl .* ρ, dims = 1) .* dz
+        ql = m_[((ind_ql - 1) * n_z + 1):(ind_ql * n_z), :]
+        lwp[:, i] = sum(ql .* ρ, dims = 1) .* dz
     end
     return lwp
 end
@@ -42,21 +42,21 @@ Returns rain water path [kg/m^2]
 """
 function rwp(vec::Vector{FT}, config) where {FT <: Real}
 
-    @assert issubset(Set(["rho", "rr"]), Set(config["observations"]["data_names"]))
+    @assert issubset(Set(["rho", "qr"]), Set(config["observations"]["data_names"]))
 
     (n_c, n_v, n_z, n_t) = get_numbers_from_config(config)
     n_vht = n_v * n_z * n_t
     dz = (config["model"]["z_max"] - config["model"]["z_min"]) / n_z
     rwp = zeros(n_t, n_c)
     ind_ρ = findall(x -> x == "rho", config["observations"]["data_names"])[1]
-    ind_rr = findall(x -> x == "rr", config["observations"]["data_names"])[1]
+    ind_qr = findall(x -> x == "qr", config["observations"]["data_names"])[1]
 
     for i in 1:n_c
         v_ = vec[((i - 1) * n_vht + 1):(i * n_vht)]
         m_ = reshape(v_, n_v * n_z, n_t)
         ρ = m_[((ind_ρ - 1) * n_z + 1):(ind_ρ * n_z), :]
-        rr = m_[((ind_rr - 1) * n_z + 1):(ind_rr * n_z), :]
-        rwp[:, i] = sum(rr .* ρ, dims = 1) .* dz
+        qr = m_[((ind_qr - 1) * n_z + 1):(ind_qr * n_z), :]
+        rwp[:, i] = sum(qr .* ρ, dims = 1) .* dz
     end
     return rwp
 end
@@ -77,37 +77,44 @@ Returns rain rate [mm/hr]
 """
 function rainrate(vec::Vector{FT}, config; height::FT = FT(0), isref = false) where {FT <: Real}
 
-    @assert issubset(Set(["rho", "rr", "rain averaged terminal velocity"]), Set(config["observations"]["data_names"]))
+    @assert issubset(Set(["rho", "qr", "rain averaged terminal velocity"]), Set(config["observations"]["data_names"]))
 
     (n_c, n_v, n_z, n_t) = get_numbers_from_config(config)
     n_vht = n_v * n_z * n_t
     dz = (config["model"]["z_max"] - config["model"]["z_min"]) / n_z
     rainrate = zeros(n_t, n_c)
     ind_ρ = findall(x -> x == "rho", config["observations"]["data_names"])[1]
-    ind_rr = findall(x -> x == "rr", config["observations"]["data_names"])[1]
+    ind_qr = findall(x -> x == "qr", config["observations"]["data_names"])[1]
     ind_vt = findall(x -> x == "rain averaged terminal velocity", config["observations"]["data_names"])[1]
-    ind_z = min(n_z, max(1, ceil(Int, (height - config["model"]["z_min"]) / dz)))
+    ind_z1 = min(n_z - 1, max(1, ceil(Int, (height - config["model"]["z_min"]) / dz - FT(0.5))))
+    ind_z2 = ind_z1 + 1
+
+    a1 = ((ind_z2 - 0.5) * dz - height) / dz
+    a2 = (height - (ind_z1 - 0.5) * dz) / dz
 
     for i in 1:n_c
         v_ = vec[((i - 1) * n_vht + 1):(i * n_vht)]
         m_ = reshape(v_, n_v * n_z, n_t)
         ρ = m_[((ind_ρ - 1) * n_z + 1):(ind_ρ * n_z), :]
-        rr = m_[((ind_rr - 1) * n_z + 1):(ind_rr * n_z), :]
+        qr = m_[((ind_qr - 1) * n_z + 1):(ind_qr * n_z), :]
 
-        ρ_z = ρ[ind_z, :]
-        rr_z = rr[ind_z, :]
+        ρ_z = a1 .* ρ[ind_z1, :] + a2 .* ρ[ind_z2, :]
+        ρ_z[findall(x -> x < 0, ρ_z)] .= FT(0)
+        qr_z = a1 .* qr[ind_z1, :] + a2 .* qr[ind_z2, :]
+        qr_z[findall(x -> x < 0, qr_z)] .= FT(0)
         if isref
             vt = m_[((ind_vt - 1) * n_z + 1):(ind_vt * n_z), :]
-            vt_z = vt[ind_z, :]
+            vt_z = a1 .* vt[ind_z1, :] + a2 .* vt[ind_z2, :]
+            vt_z[findall(x -> x < 0, vt_z)] .= FT(0)
         else
             ϕ_names = collect(keys(config["prior"]["parameters"]))
             ϕ_values = collect([v.mean for v in values(config["prior"]["parameters"])])
             params_calib = create_param_dict(ϕ_values, ϕ_names)
             params = create_parameter_set(Float64, config["model"], params_calib)
             microphys_params = KP.microphysics_params(params)
-            vt_z = CM1.terminal_velocity.(microphys_params, CMT.RainType(), ρ_z, rr_z)
+            vt_z = CM1.terminal_velocity.(microphys_params, CMT.RainType(), ρ_z, qr_z)
         end
-        rainrate[:, i] = rr_z .* ρ_z .* vt_z .* 3600
+        rainrate[:, i] = qr_z .* ρ_z .* vt_z .* 3600
     end
     return rainrate
 end

src/calibrateKiD/DistributionUtils.jl

@@ -1,6 +1,6 @@
 function initial_parameter_ensemble(priors, N_ensemble; rng_seed = 10)
     Random.seed!(rng_seed)
-    initial_ensemble = construct_initial_ensemble(priors, N_ensemble; rng_seed = rng_seed)
+    initial_ensemble = construct_initial_ensemble(priors, N_ensemble)
     return initial_ensemble
 end
 

src/calibrateKiD/KiDUtils.jl

@@ -279,17 +279,42 @@ function create_parameter_set(FT, model_settings::Dict, params_cal::Dict)
         CM.Parameters.CloudMicrophysicsParameters{FT, TP}(; fixed_microphys_pairs..., pairs..., thermo_params)
     MP = typeof(microphys_params)
 
+    precip_sources = if ("precip_sources" in keys(model_settings))
+        model_settings["precip_sources"]
+    else
+        1
+    end
+    precip_sinks = if ("precip_sinks" in keys(model_settings))
+        model_settings["precip_sinks"]
+    else
+        1
+    end
+    r_dry = if ("r_dry" in keys(model_settings))
+        model_settings["r_dry"]
+    else
+        0.04 * 1e-6
+    end
+    std_dry = if ("std_dry" in keys(model_settings))
+        model_settings["std_dry"]
+    else
+        1.4
+    end
+    κ = if ("κ" in keys(model_settings))
+        model_settings["κ"]
+    else
+        0.9
+    end
     param_set = KP.KinematicParameters{FT, MP}(;
         w1 = model_settings["w1"],
         t1 = model_settings["t1"],
         p0 = model_settings["p0"],
-        precip_sources = 1,
-        precip_sinks = 1,
+        precip_sources = precip_sources,
+        precip_sinks = precip_sinks,
         prescribed_Nd = model_settings["Nd"],
         qtot_flux_correction = Int(model_settings["qtot_flux_correction"]),
-        r_dry = model_settings["r_dry"],
-        std_dry = model_settings["std_dry"],
-        κ = model_settings["κ"],
+        r_dry = r_dry,
+        std_dry = std_dry,
+        κ = κ,
         microphys_params,
     )
     return param_set

src/calibrateKiD/OptimizationUtils.jl

@@ -105,15 +105,21 @@ end
 
 function generate_ekp(param_ensemble::Matrix{FT}, RS::ReferenceStatistics, process_settings) where {FT <: Real}
     if process_settings["EKP_method"] == "EKI"
-        ekpobj = EnsembleKalmanProcess(param_ensemble, RS.y, RS.Γ, Inversion(), Δt = process_settings["Δt"])
+        ekpobj = EnsembleKalmanProcess(
+            param_ensemble,
+            RS.y,
+            RS.Γ,
+            Inversion(),
+            scheduler = DefaultScheduler(process_settings["Δt"]),
+        )
     elseif process_settings["EKP_method"] == "UKI"
         process = Unscented(
             mean(param_ensemble, dims = 2)[:],
             cov(param_ensemble, dims = 2);
             α_reg = process_settings["α_reg"],
             update_freq = process_settings["update_freq"],
         )
-        ekpobj = EnsembleKalmanProcess(RS.y, RS.Γ, process, Δt = process_settings["Δt"])
+        ekpobj = EnsembleKalmanProcess(RS.y, RS.Γ, process, scheduler = DefaultScheduler(process_settings["Δt"]))
     end
     return ekpobj
 end

test/unit_tests/calibration_pipeline/config.jl

@@ -26,11 +26,11 @@ end
 function get_process_config()
     config = Dict()
     config["batch_size"] = 1
-    config["n_iter"] = 10
+    config["n_iter"] = 5
     config["n_ens"] = 10
     config["Δt"] = 1.0
     config["EKP_method"] = "EKI"
-    config["α_reg"] = 0.5
+    config["α_reg"] = 1.0
     config["update_freq"] = 1
     config["tol"] = 1e-8
     config["maxiter"] = 1000

test/unit_tests/calibration_pipeline/test_diagnostics.jl

@@ -1,7 +1,7 @@
 @testset "lwp rwp and rainrate" begin
     #setup
     config = get_config()
-    config["observations"]["data_names"] = ["rho", "rl", "rr", "rain averaged terminal velocity"]
+    config["observations"]["data_names"] = ["rho", "ql", "qr", "rain averaged terminal velocity"]
     (n_c, n_v, n_z, n_t) = KID.get_numbers_from_config(config)
     n_vht = n_v * n_z * n_t
     vec = rand(n_c * n_vht)
@@ -14,14 +14,17 @@
         v_ = vec[((i - 1) * n_vht + 1):(i * n_vht)]
         m_ = reshape(v_, n_v * n_z, n_t)
         ρ = m_[1:n_z, :]
-        rl = m_[(n_z + 1):(2 * n_z), :]
-        rr = m_[(2 * n_z + 1):(3 * n_z), :]
-        _lwp[:, i] = sum(rl .* ρ, dims = 1) .* dz
-        _rwp[:, i] = sum(rr .* ρ, dims = 1) .* dz
-        ρ0 = ρ[1, :]
-        rr0 = rr[1, :]
-        vt0 = m_[3 * n_z + 1, :]
-        _rainrate[:, i] = rr0 .* ρ0 .* vt0 .* 3600
+        ql = m_[(n_z + 1):(2 * n_z), :]
+        qr = m_[(2 * n_z + 1):(3 * n_z), :]
+        _lwp[:, i] = sum(ql .* ρ, dims = 1) .* dz
+        _rwp[:, i] = sum(qr .* ρ, dims = 1) .* dz
+        ρ0 = 1.5 * ρ[1, :] - 0.5 * ρ[2, :]
+        qr0 = 1.5 * qr[1, :] - 0.5 * qr[2, :]
+        vt0 = 1.5 * m_[3 * n_z + 1, :] - 0.5 * m_[3 * n_z + 2, :]
+        ρ0[findall(x -> x < 0, ρ0)] .= Float64(0)
+        qr0[findall(x -> x < 0, qr0)] .= Float64(0)
+        vt0[findall(x -> x < 0, vt0)] .= Float64(0)
+        _rainrate[:, i] = qr0 .* ρ0 .* vt0 .* 3600
     end
 
     #action

test/unit_tests/calibration_pipeline/test_optimization_utils.jl

@@ -23,7 +23,7 @@
     @test size(res[1]) == (n_vars, n_ensemble)
     @test u.ϕ_optim isa Vector
     @test u.cov_optim isa Matrix
-    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.05
+    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.1
 
     #setup
     config["process"]["EKP_method"] = "UKI"
@@ -39,7 +39,7 @@
     @test size(res[1]) == (n_vars, 2 * n_vars + 1)
     @test u.ϕ_optim isa Vector
     @test u.cov_optim isa Matrix
-    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.05
+    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.1
 
     #setup
     config["process"]["EKP_method"] = "UKP_"
@@ -81,20 +81,18 @@ end
     u = KID.get_results(res, priors)
 
     #test
-    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.05
+    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.1
 
     #setup
     config["process"]["EKP_method"] = "UKI"
-    config["process"]["batch_size"] = 1
-    config["process"]["α_reg"] = 0.1
     config["process"]["Δt"] = 10.0
 
     #action
     res = KID.calibrate(KID.EKPStyle(), priors, config, ref_stats_list, verbose = false)
     u = KID.get_results(res, priors)
 
     #test
-    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.05
+    @test norm((u.ϕ_optim .- u_true) ./ u_true) < 0.1
 end
 
 @testset "Calibrate by Optim and Get results" begin