using the Psurge v2.9 regression method for filling in Rmax in OFCL a…

…dvisories (#96) * adding complete method for rmax forecast regression without smoothing or limiting * reformatting back to black * adding the upper and lower bounds for RMW forecast * changing upper bound to be larger of 120.0 nmi or rmw0 * changing sorting of advisories by index to avoid out of order isotach_radius * adding 3-pt rolling mean for the rmax forecasts * finalizing the smoothing which interpolates to 12-hr intervals where required and then computes a 24-hr moving mean * preserving 34-kt isotach for lead times > 72-hrs if Vmax still above 34-kt * adding changes to the forecasts where the isotachs are correctly ordered and filling in RMW using regression technique from NHC * adding changes to the OFCL deck for RMW regression * ensuring presence of rads * initialize rads * Update test configuration * Remove python 3.12 from test matrix * Quick-test on lowser supported python * Try test without multiworker * Fixing python version for quicktest * Use py3.10 instead of min for coverage test * adding RMW regression coefficients into const.py and making a test for the retrieval function * switching to keep 50-kt radius as well as 34-kt * modified OFCL RMW forecast tests preserving the 50-kt isotach as well as 34-kt --------- Co-authored-by: SorooshMani-NOAA <[email protected]>
oceanmodeling · Apr 24, 2024 · ee51dd1 · ee51dd1
1 parent e9e1ae2
commit ee51dd1
Show file tree

Hide file tree

Showing 10 changed files with 6,021 additions and 5,897 deletions.
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
@@ -22,7 +22,7 @@ jobs:
       - uses: actions/checkout@main
       - uses: actions/setup-python@main
         with:
-          python-version: '3.x'
+          python-version: '3.9'
       - uses: actions/cache@main
         id: cache
         with:
@@ -41,7 +41,7 @@ jobs:
     strategy:
       matrix:
         os: [ ubuntu-latest, macos-latest ]
-        python: [ '3.9', '3.10' ]
+        python: [ '3.9', '3.10', '3.11' ]
     steps:
       - uses: actions/checkout@main
       - uses: actions/setup-python@main
@@ -62,7 +62,7 @@ jobs:
       - uses: actions/checkout@main
       - uses: actions/setup-python@main
         with:
-          python-version: '3.x'
+          python-version: '3.10'
       - uses: actions/cache@main
         id: cache
         with:

diff --git a/.github/workflows/test_quick.yml b/.github/workflows/test_quick.yml
@@ -17,7 +17,7 @@ jobs:
       - uses: actions/checkout@main
       - uses: actions/setup-python@main
         with:
-          python-version: '3.x'
+          python-version: '3.9'
       - uses: actions/cache@main
         id: cache
         with:
@@ -37,7 +37,7 @@ jobs:
       - uses: actions/checkout@main
       - uses: actions/setup-python@main
         with:
-          python-version: '3.x'
+          python-version: '3.9'
       - uses: actions/cache@main
         id: cache
         with:

diff --git a/stormevents/nhc/const.py b/stormevents/nhc/const.py
@@ -0,0 +1,46 @@
+from numpy import isnan, array, argwhere
+
+# Regression coefficients for the Rmax forecast
+# ref: Penny et al. (2023). https://doi.org/10.1175/WAF-D-22-0209.1
+fhrs = [12, 24, 36, 48, 72, 96, 120]
+RMW_regression_coefs = {
+    3: [  # a0    #a1      #a2     #a3      #a4     #a5     #a6
+        [3.1894, 0.3524, 0.1208, -0.1091, 0.5862, -0.8070, 0.0057],
+        [4.4373, 0.1473, 0.1045, -0.1112, 0.7566, -1.0689, 0.0061],
+        [4.9447, 0.0784, 0.1168, -0.1448, 0.8246, -1.1709, 0.0059],
+        [5.1818, 0.0549, 0.1335, -0.2345, 0.8972, -1.2038, 0.0063],
+    ],
+    2: [  # a0    #a1      #a2     #a3      #a5     #a6
+        [3.1131, 0.3680, 0.1589, 0.4710, -0.9111, 0.0068],
+        [4.1567, 0.1834, 0.2085, 0.5873, -1.1841, 0.0073],
+        [4.6694, 0.1062, 0.2330, 0.6295, -1.3122, 0.0074],
+        [4.9434, 0.0459, 0.3027, 0.5828, -1.3675, 0.0079],
+        [4.7906, 0.0157, 0.3953, 0.5321, -1.3617, 0.0067],
+    ],
+    1: [  # a0    #a1      #a2     #a5      #a6
+        [2.6272, 0.4230, 0.6320, -0.9117, 0.0064],
+        [3.6525, 0.2142, 0.8222, -1.2158, 0.0082],
+        [4.2822, 0.0884, 0.9059, -1.3656, 0.0091],
+        [4.7700, -0.0042, 0.9225, -1.4349, 0.0102],
+        [4.7307, -0.0365, 0.9153, -1.3882, 0.0086],
+    ],
+    0: [  # a0    #a1      #a5     #a6
+        [2.1633, 0.6360, -0.3314, 0.0154],
+        [3.7884, 0.3953, -0.5738, 0.0219],
+        [5.0213, 0.1999, -0.7481, 0.0276],
+        [5.8092, 0.0615, -0.8508, 0.0318],
+        [6.3321, -0.0362, -0.9079, 0.0343],
+        [6.6181, 0.0041, -0.9599, 0.0295],
+        [6.7073, -0.0028, -0.9478, 0.0257],
+    ],
+}
+
+
+def get_RMW_regression_coefs(fcst_hr, radii_values):
+    num_radii_available = (~isnan(radii_values)).sum()
+    coefs_by_radii_available = array(RMW_regression_coefs[num_radii_available])
+    fcst_index = argwhere(fhrs == fcst_hr)
+    if fcst_index.size == 0 or fcst_index > coefs_by_radii_available.shape[0] - 1:
+        return coefs_by_radii_available[-1].flatten()
+    else:
+        return coefs_by_radii_available[fcst_index].flatten()
diff --git a/stormevents/nhc/track.py b/stormevents/nhc/track.py
@@ -33,6 +33,7 @@
 from stormevents.nhc.atcf import get_atcf_entry
 from stormevents.nhc.atcf import read_atcf
 from stormevents.nhc.storms import nhc_storms
+from stormevents.nhc.const import get_RMW_regression_coefs
 from stormevents.utilities import subset_time_interval
 
 
@@ -1204,7 +1205,7 @@ def separate_tracks(data: DataFrame) -> Dict[str, Dict[str, DataFrame]]:
                 track_data = advisory_data[
                     advisory_data["track_start_time"]
                     == pandas.to_datetime(track_start_time)
-                ].sort_values("forecast_hours")
+                ].sort_index()
 
             tracks[advisory][
                 f"{pandas.to_datetime(track_start_time):%Y%m%dT%H%M%S}"
@@ -1235,6 +1236,9 @@ def correct_ofcl_based_on_carq_n_hollandb(
     :return: dictionary of forecasts for each advisory (aside from best track ``BEST``, which only has one hindcast) with corrected OFCL
     """
 
+    def clamp(n, minn, maxn):
+        return max(min(maxn, n), minn)
+
     ofcl_tracks = tracks["OFCL"]
     carq_tracks = tracks["CARQ"]
 
@@ -1269,10 +1273,75 @@ def correct_ofcl_based_on_carq_n_hollandb(
         mslp_missing = missing.iloc[:, 1]
         radp_missing = missing.iloc[:, 2]
 
-        # fill OFCL maximum wind radius with the first entry from 0-hr CARQ
-        forecast.loc[mrd_missing, "radius_of_maximum_winds"] = carq_ref[
-            "radius_of_maximum_winds"
+        # fill OFCL maximum wind radius based on regression method from
+        # Penny et al. (2023). https://doi.org/10.1175/WAF-D-22-0209.1
+        isotach_radii = forecast[
+            [
+                "isotach_radius_for_NEQ",
+                "isotach_radius_for_SEQ",
+                "isotach_radius_for_NWQ",
+                "isotach_radius_for_SWQ",
+            ]
         ]
+        isotach_radii[isotach_radii == 0] = pandas.NA
+        rmw0 = carq_ref["radius_of_maximum_winds"]
+        fcst_hrs = (forecast.loc[mrd_missing, "forecast_hours"]).unique()
+        rads = numpy.array([numpy.nan])  # initializing to make sure available
+        for fcst_hr in fcst_hrs:
+            fcst_index = forecast["forecast_hours"] == fcst_hr
+            if fcst_hr < 12:
+                rmw_ = rmw0
+            else:
+                vmax = forecast.loc[fcst_index, "max_sustained_wind_speed"].iloc[0]
+                if numpy.isnan(isotach_radii.loc[fcst_index].to_numpy()).all():
+                    # if no isotach's are found, preserve the isotach(s) if Vmax is greater
+                    if vmax > 50:
+                        rads = rads[0 : min(2, len(rads))]
+                    elif vmax > 34:
+                        rads = rads[[0]]
+                    else:
+                        rads = numpy.array([numpy.nan])
+                else:
+                    rads = numpy.nanmean(
+                        isotach_radii.loc[fcst_index].to_numpy(), axis=1
+                    )
+                coefs = get_RMW_regression_coefs(fcst_hr, rads)
+                lat = forecast.loc[fcst_index, "latitude"].iloc[0]
+                bases = numpy.hstack((1.0, rmw0, rads[~numpy.isnan(rads)], vmax, lat))
+                rmw_ = (bases[1:-1] ** coefs[1:-1]).prod() * numpy.exp(
+                    (coefs[[0, -1]] * bases[[0, -1]]).sum()
+                )  # bound RMW as per Penny et al. (2023)
+            forecast.loc[fcst_index, "radius_of_maximum_winds"] = clamp(
+                rmw_, 5.0, max(120.0, rmw0)
+            )
+        # apply 24-HR moving mean to unique datetimes
+        fcsthr_index = forecast["forecast_hours"].drop_duplicates().index
+        df_temp = forecast.loc[fcsthr_index].copy()
+        # make sure 60, 84, and 108 are added
+        fcsthrs_12hr = numpy.unique(
+            numpy.append(df_temp["forecast_hours"].values, [60, 84, 108])
+        )
+        rmw_12hr = numpy.interp(
+            fcsthrs_12hr, df_temp["forecast_hours"], df_temp["radius_of_maximum_winds"]
+        )
+        dt_12hr = pandas.to_datetime(
+            fcsthrs_12hr, unit="h", origin=df_temp["datetime"].iloc[0]
+        )
+        df_temp = DataFrame(
+            data={"forecast_hours": fcsthrs_12hr, "radius_of_maximum_winds": rmw_12hr},
+            index=dt_12hr,
+        )
+        rmw_rolling = df_temp.rolling(window="24.01 H", center=True, min_periods=1)[
+            "radius_of_maximum_winds"
+        ].mean()
+        for valid_time, rmw in rmw_rolling.items():
+            valid_index = forecast["datetime"] == valid_time
+            if (
+                valid_index.sum() == 0
+                or forecast.loc[valid_index, "forecast_hours"].iloc[0] == 0
+            ):
+                continue
+            forecast.loc[valid_index, "radius_of_maximum_winds"] = rmw
 
         # fill OFCL background pressure with the first entry from 0-hr CARQ background pressure (at sea level)
         forecast.loc[radp_missing, "background_pressure"] = carq_ref[