Time Series Forecasting
PyTorch
English
pluvium
weather
weather-forecasting
climate
meteorology
earth-science
precipitation
rainfall
extreme-rainfall
nowcasting
probabilistic-forecasting
tropical-cyclones
hurricane
multi-model-ensemble
post-processing
online-learning
geospatial
caribbean
naturev1
ihelix
graphcast
aifs
File size: 7,070 Bytes
fbb9a05 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 | # Copyright 2026 Nathan. Apache-2.0.
"""Two charts for the frontier benchmark, from its result files.
python scripts/chart_frontier.py RESULTS FIRST docs/img
``frontier_heavy_rain.png``: Brier skill for 50 mm or more on days 1-3, place by place: the first Blend, the new
Blend and the best other system. ``frontier_handover.png``: what the record-storm hand-over did in the test year in
each place, and what the place's staged storms had said about it from the training year.
"""
import json
import sys
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from pluvium.benchmarks import fullyear as F
from pluvium.frontier import report as R
SURFACE, INK, INK2, MUTED, GRID, AXIS = "#fcfcfb", "#0b0b0b", "#52514e", "#898781", "#e1e0d9", "#c3c2b7"
BLUE, ORANGE, GREEN = "#2a78d6", "#eb6834", "#1baf7a" # the validated order: Pluvium is always blue
ORDER = ("jamaica", "caribbean", "maritime", "alps", "pacific_nw", "great_plains", "andes_south")
SHORT = {"jamaica": "Jamaica", "caribbean": "Caribbean", "maritime": "Maritime Continent", "alps": "Alps",
"pacific_nw": "US Pacific Northwest", "great_plains": "US Great Plains", "andes_south": "Southern Andes"}
GROUP = "days 1-3"
results_dir, first_dir, out_dir = (Path(p) for p in sys.argv[1:4])
new = {k: json.loads((results_dir / f"{k}.json").read_text()) for k in ORDER if (results_dir / f"{k}.json").exists()}
old = {k: json.loads((first_dir / f"{k}.json").read_text()) for k in new if (first_dir / f"{k}.json").exists()}
places = [p for p in ORDER if p in new][::-1] # first place on top
plt.rcParams.update({"font.family": "DejaVu Sans", "font.size": 9, "axes.edgecolor": AXIS, "axes.labelcolor": INK2,
"xtick.color": MUTED, "ytick.color": MUTED, "text.color": INK})
def frame(ax) -> None:
ax.set_facecolor(SURFACE)
ax.grid(axis="x", color=GRID, lw=0.8)
ax.set_axisbelow(True)
for side in ("top", "right", "left"):
ax.spines[side].set_visible(False)
ax.tick_params(length=0)
def skill(res: dict, model: str) -> float:
return F._value(res["groups"][GROUP].get(model), "bss", "50")
# ---------------------------------------------------------------------------------------- heavy rain --
fig, ax = plt.subplots(figsize=(8.2, 0.52 * len(places) + 1.9))
fig.patch.set_facecolor(SURFACE)
frame(ax)
notes = []
for y, place in enumerate(places):
res = new[place]
mine = skill(res, "stack")
rival, who = R.best_rival(res["groups"][GROUP], "bss", "50", True)
first = skill(old[place], F._top(old[place])) if place in old else float("nan")
if first == first:
ax.plot([first, mine], [y, y], color=AXIS, lw=2, solid_capstyle="round", zorder=2)
ax.plot(first, y, "o", ms=8, color=GREEN, mec=SURFACE, mew=2, zorder=3)
ax.plot(rival, y, "D", ms=7, color=ORANGE, mec=SURFACE, mew=2, zorder=3)
ax.plot(mine, y, "o", ms=9, color=BLUE, mec=SURFACE, mew=2, zorder=4)
label = F.LABELS.get(str(who).replace("_raw", ""), str(who)) + (" raw" if str(who).endswith("_raw") else "")
right = max(v for v in (mine, rival, first) if v == v)
was = f" (first {first:+.3f})" if first == first else ""
notes.append((right, y, f"{mine:+.3f}{was} best other: {label} {rival:+.3f}"))
ax.set_yticks(range(len(places)))
ax.set_yticklabels([SHORT[p] for p in places], color=INK)
ax.set_ylim(-0.6, len(places) - 0.4)
lo, hi = ax.get_xlim()
ax.set_xlim(min(lo, -0.005), hi + 0.75 * (hi - lo))
gap = 0.022 * (ax.get_xlim()[1] - ax.get_xlim()[0]) # clear of the marker, whatever the scale
for right, y, text in notes:
ax.text(right + gap, y, text, fontsize=8, color=INK2, va="center")
ax.axvline(0, color=AXIS, lw=1)
ax.set_xlabel("Brier skill score for 50 mm or more, days 1-3 (higher is better; 0 is the climatology)")
handles = [plt.Line2D([], [], color=BLUE, marker="o", ms=9, mec=SURFACE, mew=2, lw=0),
plt.Line2D([], [], color=GREEN, marker="o", ms=8, mec=SURFACE, mew=2, lw=0),
plt.Line2D([], [], color=ORANGE, marker="D", ms=7, mec=SURFACE, mew=2, lw=0)]
fig.legend(handles, ["Pluvium Blend 0.3", "First Pluvium Blend", "Best other system, raw or post-processed"],
loc="upper left", ncol=3, frameon=False, fontsize=9, bbox_to_anchor=(0.01, 1.0))
fig.text(0.01, -0.03, "Test year April 2025 to March 2026, truth CMORPH. The same year the first Blend was scored on.",
fontsize=8, color=MUTED)
fig.tight_layout(rect=(0, 0.01, 1, 0.93))
out_dir.mkdir(parents=True, exist_ok=True)
fig.savefig(out_dir / "frontier_heavy_rain.png", dpi=200, facecolor=SURFACE, bbox_inches="tight")
plt.close(fig)
print("saved", out_dir / "frontier_heavy_rain.png")
# ------------------------------------------------------------------------------------------ hand-over --
rows = []
for place in places:
res = new[place]
verdict = R.kept(res) # what the place's staged storms said
effect = skill(res, "declared") - skill(res, "pool_online")
if verdict is not None and effect == effect:
rows.append((place, effect, verdict, res["staged"].get("gain", float("nan"))))
if rows:
fig, ax = plt.subplots(figsize=(8.2, 0.5 * len(rows) + 1.9))
fig.patch.set_facecolor(SURFACE)
frame(ax)
span = max(abs(e) for _, e, _, _ in rows) or 1e-3
for y, (place, effect, on, gain) in enumerate(rows):
ax.barh(y, effect, height=0.5, color=BLUE if on else ORANGE, zorder=3)
said = ("kept" if on else "switched off") + (f" ({-100 * gain:+.1f}% on staged storms)" if gain == gain else "")
side = 1 if effect >= 0 else -1
ax.text(effect + side * 0.03 * span, y, f"{effect:+.4f} {said}", fontsize=8, color=INK2, va="center",
ha="left" if side > 0 else "right")
ax.axvline(0, color=INK2, lw=1, zorder=4)
ax.set_yticks(range(len(rows)))
ax.set_yticklabels([SHORT[p] for p, _, _, _ in rows], color=INK)
ax.set_ylim(-0.6, len(rows) - 0.4)
ax.set_xlim(-3.0 * span, 3.0 * span)
ax.set_xlabel("Change in Brier skill at 50 mm on days 1-3 from handing record storms to the raw systems, test year")
handles = [plt.Rectangle((0, 0), 1, 1, color=BLUE), plt.Rectangle((0, 0), 1, 1, color=ORANGE)]
fig.legend(handles, ["Staged storms in the training year said: keep the hand-over", "They said: switch it off"],
loc="upper left", ncol=2, frameon=False, fontsize=9, bbox_to_anchor=(0.01, 1.0))
fig.text(0.01, -0.03, "Bars to the right: the hand-over helped in the test year. The verdict saw the training year only: "
"its four wettest days,\nwithheld from training and then forecast. In brackets: the change in the Brier score "
"at 50 mm on those storms.", fontsize=8, color=MUTED)
fig.tight_layout(rect=(0, 0.01, 1, 0.93))
fig.savefig(out_dir / "frontier_handover.png", dpi=200, facecolor=SURFACE, bbox_inches="tight")
plt.close(fig)
print("saved", out_dir / "frontier_handover.png")
|