tools/download_forcing.py

#!/usr/bin/env python3
"""
download_forcing.py — download 2-min MRMS PrecipRate over an event window from
the Iowa State MT archive, clip to a watershed polygon, and write the basin-mean
precipitation-rate time series (same format as the released
mrms_2min_precip_basin_mean.csv: datetime_utc, precip_rate_mm_hr).

Requires: rasterio (GDAL GRIB driver), shapely, numpy, pandas, requests.

Example (light-package testbed -> add its forcing)
--------------------------------------------------
  python download_forcing.py \
      --watershed ../../data/L3/testbeds/FF_2024_09_VA_ep011/03165500/watershed.geojson \
      --begin 2024-09-27T06:30 --end 2024-09-28T04:00 \
      --out mrms_2min_precip_basin_mean.csv --cache ./mrms_cache

Window convention: the released forcing covers episode begin - 24 h to
episode end + 6 h; pass --pre-hours/--post-hours to change it.
"""
import argparse
import gzip
import io
import tempfile
from datetime import timedelta
from pathlib import Path

import numpy as np
import pandas as pd
import requests

from torrent_tools import UA, mrms_precip_urls, parse_utc


def load_watershed(path):
    import json
    from shapely.geometry import shape
    from shapely.ops import unary_union
    gj = json.load(open(path))
    geoms = [shape(f["geometry"]) for f in gj.get("features", [gj])]
    poly = unary_union(geoms)
    return poly, poly.bounds  # (west, south, east, north)


def clip_mean(grib_bytes, bounds, mask_cache, poly):
    import rasterio
    import rasterio.windows
    from rasterio.features import geometry_mask
    w, s, e, n = bounds
    with tempfile.NamedTemporaryFile(suffix=".grib2") as tf:
        tf.write(grib_bytes)
        tf.flush()
        with rasterio.open(tf.name) as src:
            win = rasterio.windows.from_bounds(w + 360 if src.bounds.left > 180 else w,
                                               s, e + 360 if src.bounds.left > 180 else e,
                                               n, transform=src.transform)
            win = win.round_offsets().round_lengths()
            arr = src.read(1, window=win).astype(float)
            tfm = src.window_transform(win)
    arr[arr < 0] = np.nan
    if mask_cache.get("mask") is None or mask_cache.get("shape") != arr.shape:
        from shapely.affinity import translate
        p = poly
        if tfm.c > 180:                       # grid uses 0-360 longitudes
            p = translate(poly, xoff=360.0)
        mask_cache["mask"] = geometry_mask([p.__geo_interface__], out_shape=arr.shape,
                                           transform=tfm, invert=True)  # True inside
        mask_cache["shape"] = arr.shape
        if not mask_cache["mask"].any():      # degenerate: fall back to bbox mean
            mask_cache["mask"] = np.ones(arr.shape, bool)
    vals = arr[mask_cache["mask"]]
    return float(np.nanmean(vals)) if np.isfinite(vals).any() else np.nan


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--watershed", required=True, help="watershed.geojson")
    ap.add_argument("--begin", required=True, help="episode begin (UTC)")
    ap.add_argument("--end", required=True, help="episode end (UTC)")
    ap.add_argument("--pre-hours", type=float, default=24.0)
    ap.add_argument("--post-hours", type=float, default=6.0)
    ap.add_argument("--out", default="mrms_2min_precip_basin_mean.csv")
    ap.add_argument("--cache", default="./mrms_cache", help="GRIB download cache dir")
    args = ap.parse_args()

    poly, bounds = load_watershed(args.watershed)
    b = parse_utc(args.begin) - timedelta(hours=args.pre_hours)
    e = parse_utc(args.end) + timedelta(hours=args.post_hours)
    steps = mrms_precip_urls(b, e)
    cache = Path(args.cache)
    cache.mkdir(parents=True, exist_ok=True)
    print(f"{len(steps)} 2-min timesteps  {b} -> {e}")

    mask_cache = {}
    rows, missing = [], 0
    with requests.Session() as ses:
        ses.headers.update(UA)
        for i, (t, url) in enumerate(steps):
            gz = cache / url.rsplit("/", 1)[-1]
            if not gz.exists():
                r = ses.get(url, timeout=60)
                if r.status_code != 200:
                    missing += 1
                    rows.append((t, np.nan))
                    continue
                gz.write_bytes(r.content)
            try:
                raw = gzip.decompress(gz.read_bytes())
                rows.append((t, clip_mean(raw, bounds, mask_cache, poly)))
            except Exception:
                missing += 1
                rows.append((t, np.nan))
            if (i + 1) % 100 == 0:
                print(f"  {i+1}/{len(steps)} done ({missing} missing)")

    df = pd.DataFrame(rows, columns=["datetime_utc", "precip_rate_mm_hr"])
    df.to_csv(args.out, index=False)
    tot = np.nansum(df.precip_rate_mm_hr.to_numpy()) * (2 / 60)
    print(f"wrote {args.out}  ({len(df)} steps, {missing} missing, "
          f"event total ~{tot:.1f} mm)")


if __name__ == "__main__":
    main()