Add RMSE leaderboard

experiments/ClimaEarth/run_amip.jl

@@ -923,19 +923,49 @@ if ClimaComms.iamroot(comms_ctx)
 
             include("user_io/leaderboard.jl")
             compare_vars = ["pr"]
-            function plot_biases(dates, output_name)
+            function compute_biases(dates)
+                if isempty(dates)
+                    return map(x -> 0.0, compare_vars)
+                else
+                    return Leaderboard.compute_biases(atmos_sim.integrator.p.output_dir, compare_vars, dates)
+                end
+            end
+
+            function plot_biases(dates, biases, output_name)
+                isempty(dates) && return nothing
+
                 output_path = joinpath(dir_paths.artifacts, "bias_$(output_name).png")
-                Leaderboard.plot_biases(atmos_sim.integrator.p.output_dir, compare_vars, dates; output_path)
+                Leaderboard.plot_biases(biases; output_path)
             end
-            plot_biases(output_dates, "total")
+
+            ann_biases = compute_biases(output_dates)
+            plot_biases(output_dates, ann_biases, "total")
 
             ## collect all days between cs.dates.date0 and cs.dates.date
             MAM, JJA, SON, DJF = Leaderboard.split_by_season(output_dates)
 
-            !isempty(MAM) && plot_biases(MAM, "MAM")
-            !isempty(JJA) && plot_biases(JJA, "JJA")
-            !isempty(SON) && plot_biases(SON, "SON")
-            !isempty(DJF) && plot_biases(DJF, "DJF")
+            MAM_biases = compute_biases(MAM)
+            plot_biases(MAM, MAM_biases, "MAM")
+            JJA_biases = compute_biases(JJA)
+            plot_biases(JJA, JJA_biases, "JJA")
+            SON_biases = compute_biases(SON)
+            plot_biases(SON, SON_biases, "SON")
+            DJF_biases = compute_biases(DJF)
+            plot_biases(DJF, DJF_biases, "DJF")
+
+            rmses = map(
+                (index) -> Leaderboard.RMSEs(;
+                    model_name = "CliMA",
+                    ANN = ann_biases[index],
+                    DJF = DJF_biases[index],
+                    JJA = JJA_biases[index],
+                    MAM = MAM_biases[index],
+                    SON = SON_biases[index],
+                ),
+                1:length(compare_vars),
+            )
+
+            Leaderboard.plot_leaderboard(rmses; output_path = "bias_leaderboard.png")
         end
     end
 

experiments/ClimaEarth/user_io/leaderboard/compare_with_obs.jl

@@ -1,16 +1,36 @@
 const OBS_DS = Dict()
 const SIM_DS_KWARGS = Dict()
+const OTHER_MODELS_RMSEs = Dict()
 
 function preprocess_pr_fn(data)
     # -1 kg/m/s2 -> 1 mm/day
     return data .* Float32(-86400)
 end
 
+Base.@kwdef struct RMSEs
+    model_name::String
+    ANN::Union{<:Real, ClimaAnalysis.OutputVar} = 0.0
+    DJF::Union{<:Real, ClimaAnalysis.OutputVar} = 0.0
+    JJA::Union{<:Real, ClimaAnalysis.OutputVar} = 0.0
+    MAM::Union{<:Real, ClimaAnalysis.OutputVar} = 0.0
+    SON::Union{<:Real, ClimaAnalysis.OutputVar} = 0.0
+end
+
+function Base.values(r::RMSEs)
+    val_or_rmse(v::Real) = v
+    val_or_rmse(v::ClimaAnalysis.OutputVar) = v.attributes["rmse"]
+
+    return val_or_rmse.([r.ANN, r.DJF, r.JJA, r.MAM, r.SON])
+end
+
 OBS_DS["pr"] =
     ObsDataSource(; path = joinpath(pr_obs_data_path(), "gpcp.precip.mon.mean.197901-202305.nc"), var_name = "precip")
 
 SIM_DS_KWARGS["pr"] = (; preprocess_data_fn = preprocess_pr_fn, new_units = "mm / day")
 
+OTHER_MODELS_RMSEs["pr"] =
+    [RMSEs(; model_name = "AM4.0 (eyeballed)", ANN = 0.5, DJF = 1.0, JJA = 1.5, MAM = 0.5, SON = 1.0)]
+
 # OBS_DS["rsut"] = ObsDataSource(;
 #                              path = "OBS/CERES_EBAF-TOA_Ed4.2_Subset_200003-202303.g025.nc",
 #                              var_name = "toa_sw_all_mon",
@@ -27,13 +47,39 @@ function bias(output_dir::AbstractString, short_name::AbstractString, target_dat
     return bias(obs, sim, target_dates)
 end
 
-function plot_biases(output_dir, short_names, target_dates::AbstractArray{<:Dates.DateTime}; output_path)
-    fig = CairoMakie.Figure(; size = (600, 300 * length(short_names)))
+function compute_biases(output_dir, short_names, target_dates::AbstractArray{<:Dates.DateTime})
+    return map(name -> bias(output_dir, name, target_dates), short_names)
+end
+
+function plot_biases(biases; output_path)
+    fig = CairoMakie.Figure(; size = (600, 300 * length(biases)))
     loc = 1
-    for short_name in short_names
-        bias_var = bias(output_dir, short_name, target_dates)
+    for bias_var in biases
         ClimaAnalysis.Visualize.heatmap2D_on_globe!(fig, bias_var; p_loc = (1, loc))
         loc = loc + 1
     end
     CairoMakie.save(output_path, fig)
 end
+
+function plot_leaderboard(rmses; output_path)
+    fig = CairoMakie.Figure(; size = (600, 300 * length(rmses)))
+    loc = 1
+
+    for rmse in rmses
+        short_name = rmse.ANN.attributes["var_short_name"]
+        units = rmse.ANN.attributes["units"]
+        ax = CairoMakie.Axis(
+            fig[1, loc],
+            ylabel = "$short_name [$units]",
+            xticks = (1:5, ["Ann", "DJF", "JJA", "MAM", "SON"]),
+            title = "Global RMSE",
+        )
+        CairoMakie.scatter!(ax, 1:5, values(rmse), label = rmse.model_name)
+        for other_model_rmse in OTHER_MODELS_RMSEs[short_name]
+            CairoMakie.scatter!(ax, 1:5, values(other_model_rmse), label = other_model_rmse.model_name)
+        end
+        CairoMakie.axislegend()
+        loc = loc + 1
+    end
+    CairoMakie.save(output_path, fig)
+end

experiments/ClimaEarth/user_io/leaderboard/utils.jl

@@ -152,6 +152,7 @@ function bias(obs_ds::ObsDataSource, sim_ds::SimDataSource, target_dates::Abstra
 
     bias_attribs = Dict{String, Any}(
         "short_name" => "sim-obs_$short_name",
+        "var_short_name" => "$short_name",
         "long_name" => "SIM - OBS mean $short_name\n(RMSE: $rmse $units, Global bias: $global_bias $units)",
         "rmse" => rmse,
         "bias" => global_bias,