1. Introduction and motivation
At U.S. tropical cyclone (TC) warning centers, advisories and forecasts provide estimates of the extent of a tropical cyclone’s wind field in terms of the maximum extent of wind exceeding the key threshold magnitudes of 34, 50, and 64 kt (1 kt ≈ 0.51 m s−1). These metrics are collectively referred to as “wind radii” and are provided in northeast, southeast, southwest, and northwest geographic quadrants in units of nautical miles (n mi, 1 n mi = 1.852 km) along with forecasts of the maximum sustained 1-min wind or intensity, storm motion, and storm location.
To help forecast and provide a skill baseline for forecasting wind radii, Knaff et al. (2007) developed a set of basin-specific 5-day statistical models in 2004. These models made use of the combination of climatology and persistence of initial conditions, commonly referred to as “CLIPER” models. CLIPER models have a long history of providing skill baselines for TC forecasting parameters (e.g., Neumann 1972; Jarvinen and Neumann 1979; Merrill 1980; Chu 1994; Aberson 1998; Knaff et al. 2003). Knaff et al. (2007) describe the wind radii CLIPER models for the North Atlantic, eastern North Pacific, and western North Pacific basins, which use the moniker “DRCL” in the Automated Tropical Cyclone Forecasting System (ATCF; Sampson and Schrader 2000) database at the National Hurricane Center (NHC), Central Pacific Hurricane Center (CPHC), and the Joint Typhoon Warning Center (JTWC). Henceforth, we will refer to the suite of wind radii CLIPER models as DRCL. The original DRCL basin-specific models (Knaff et al. 2007) were developed using wind radii contained in the advisories 1988–2003 and west of 55°W in the North Atlantic and 2001–03 in the eastern and western North Pacific. Knaff et al. (2018) updated the coefficients of the western North Pacific DRCL model using higher-quality postseason analyzed wind radii (2014–15) described in Sampson et al. (2017, 2018), which resulted in an improved and larger climatology in that basin.
One glaring exception is that the DRCL model struggles with TCs that are initially much smaller or much larger than their climatological counterparts. In nature, the majority of TCs remain in the same size category observed at formation (Lee et al. 2010) and tend to grow throughout their lifetimes (Kossin et al. 2007). Thus, the least operationally useful DRCL forecasts occur when the initial storm is much larger than climatology. In these cases, the DRCL forecast size shrinks to approximate climatology, whether that is expected in the operational forecast or not. Such poor forecasts are rare but result in loss of operational trust in the algorithm. One such case—a dramatic case—was that of Hurricane Sandy (2012) late in its development, when the storm was forecast to hit the east coast of the United States. Figure 1 shows the official and DRCL 34-kt wind radii forecasts for Sandy valid at 1800 UTC 26 October along with the resulting best track values of wind radii. Note that the initial conditions for DRCL are the same as for the official forecast, but that the DRCL 34-kt wind radii shrink as the forecast time increases. Meanwhile, the official forecast in Fig. 1a (and best track Fig. 1c) expands the wind radii through the forecast, which makes the DRCL shrinkage at odds with the official forecast and best track expansions. Figure 2 shows the operational 120-h cumulative 34-kt wind speed probabilities initialized at the same time, and the effects of the DRCL wind radii shrinkage can be seen in that product as the forecast wind radii from NHC are nearly as large as the entire cumulative 34-kt wind speed probability swath. One solution to the DRCL issues with extreme sizes and asymmetries is to construct the solution in short increments of 12 h (e.g., a Markov chain), which can possibly retain both the size and asymmetry of the current TC analysis.
The 34-kt wind radii forecasts for Hurricane Sandy valid at 1800 UTC 26 Oct 2012 from (a) the NHC official forecast and (b) the DRCL forecast along with the (c) best track intensities and wind radii.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
Cumulative 34-kt wind speed probabilities generated from the 1800 UTC 26 Oct 2012 initial conditions and 2100 UTC official forecasts of track and intensity for Hurricane Sandy. These are available from graphic archives at https://www.nhc.noaa.gov/.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
This work creates a statistical wind radii prediction model based on a second-order Markov process of TC size, which includes initial conditions that describes the current TC characteristics (storm latitude, storm intensity, 12-h intensity change, zonal translation speed, and meridional translation speed), which are chained together from the initial time to the end of the forecast period. The TC size information comes from a homogeneous record of satellite-based size estimates described in Knaff et al. (2014). This model, MMCL, has been developed for the North Atlantic (AL), eastern North Pacific (EP), western North Pacific (WP), north Indian Ocean (IO), and Southern Hemisphere (SH) basins. It has been in JTWC’s operations since late 2020. The IO model is not evaluated in this work due to sample size, but its development is discussed. In the following sections, we describe the data and methods, model development, and results based on independent forecasts from 2020 to 2022. Finally, we will provide a summary and discussion of future work on this topic.
2. Data and methods
a. Developmental data
The TC size record used for MMCL development comes from an infrared (IR) satellite-based method described in Knaff et al. (2014). The IR imagery comes from an archive of TC-centered IR images maintained at the NOAA Center for Satellite Applications and Research (STAR) and the NOAA Cooperative Institute for Research in the Atmosphere (CIRA) at the Colorado State University. These IR images (with central wavelengths near 11 μm) are collected from operational geostationary satellites. The digital imagery is remapped to a Mercator projection with 4-km resolution. Most of the images in the archive are 640 × 480 pixels and centered on the TCs, and the temporal sampling depends on the satellite scan strategy, data availability/quality, and storm location but in general is typically hourly to half hourly. More information on the imagery archive can be found in Knaff et al. (2016, 2018).
The IR-based TC size estimates the radius of 5-kt tangential winds at 850 hPa (R5 in Knaff et al. 2014), which is an approximation of the theoretical radius of zero tangential wind based on storm latitude and the first two 1D principle components of the azimuthal profiles of brightness temperature. Three-hourly R5 estimates provide an extensive homogeneous record of TC size. Even though the intensity is not explicitly used to calculate R5 in the database, R5 is found to increase with intensity (Knaff et al. 2014) and that dependence is removed by creating “fraction TC size” or fR5 as described in Knaff et al. (2017). This fR5 size metric from the IR records used here varies from about 30% to 200% with near Gaussian distribution so the smallest storms and the largest storms range from roughly a third to twice the intensity-based climatology. Later, we will also exploit relationships that provide azimuthally averaged wind radii from R5 and storm intensity described in Knaff et al. (2016).
We also use the TC best tracks for information on TC motion, location, and intensity. The ATCF-formatted best tracks used are constructed by the NHC, CPHC, and JTWC. This same information is converted and disseminated in other formats: The North Atlantic and eastern North Pacific best tracks are in the Hurricane database (HURDAT2; Landsea and Franklin 2013) and both JTWC and NHC best track databases end up in the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010).
Best track intensity, motion, and latitude are interpolated to the 3-hourly records of fR5 at 0000, 0300, 0600, …, 2100 UTC using cubic splines. Then, the 12-h differences of fR5, position, and intensity from t = −12 are calculated at the time periods from the 3-hourly record. This forms the dataset for developing a 12-h statistical forecast model based on a second-order Markov process. The dependent variable is 12-h change of fR5 valid at t = 12 h, and the independent variables are sine of latitude [sin(|φ|)], intensity Vm, the past 12-h change in intensity dVm, fR5, the past 12-h change in fR5 dfR5, zonal translation speed u, and meridional translation speed υ.
b. Statistical model formulation
Using the developmental data discussed above, multiple linear regression is used to form a prediction of the fR5 at t = 12 h. As with most skill baselines, the cases are separated into individual basins. The number of cases found in the IO, WP, EP, AL, and SH basins are 2563, 18 046, 24 352, 21 676, and 14 659, respectively. Because of the large number of cases used, the variables are mostly significant at the 95th percentile using the Student’s t test. The notable exceptions are associated with variables Vm and dVm in the IO, Vm in the SH, and u in the EP and SH. These coefficients are very small but are left in the equations to standardize the 12-h forecast routines. Table 1 provides regression coefficients for the five basins. The dominant terms in all basins are the sin(|φ|), fR5, and dfR5 terms, while the other terms have minimal impact on the 12-h forecasts. The coefficients for sin(|φ|), fR5, and dfR5 are generally on the order of 10−1, 1, and 10−1, respectively, and suggest that persistence and poleward motion act to expand the radii while observed changes in TC size (dfR5) act to contract the radii.
Regression coefficients for 12-h forecasts of dfR5 in IO, WP, EP, AL, and SH basins. Boldface indicates statistical significance at the 95th percentile level.
c. Chaining the Markov processes
Up to this point, we have discussed how 12-h forecasts of fR5 are initialized and created, but to create a longer forecast, we must chain successive 12-h forecasts together. To do so, we use the forecast track and intensity to provide initial conditions every 12 h. The procedure is as follows: At t = 0, a 12-h forecast of fR5 is made. Then, the 12-h forecast values of sin(|φ|), Vm, dVm, u, and υ are used along with the newly computed fR5 (at 12 h) and dfR5 (12-h change in fR5 from t = 0 to t = 12 h) from the Markov model’s first forecast. This process continues until fR5 is forecast all the forecast times, which is typically from 12 to 120 h. Note that extending the forecast time only requires additional forecast track and intensity estimates at these longer forecast lengths. At each forecast time, Vm, latitude, storm speed, and storm direction are retained to allow estimation of R5 and wind radii based on R5 as described in the following subsections.
d. Treatment of initial conditions
Initial wind radii are used in two ways in this model. The first is to estimate the TC size parameters, and the second is to apply the initial wind radii asymmetries as they were observed. In this way, the model always starts with the observed conditions, if they exist.
Initial wind radii asymmetries are also assigned from the observations at t = 0 h. Beyond the initial time, the asymmetries are also blended with the model’s wind radii. This process will be described in detail below.
e. The vortex model and wind radii estimation
f. Growth correction
The 12-h TC size changes in MMCL predictions can be thought of as the residuals of a naïve or persistence forecast (Elsner et al. 2008). Thus, each TC size time series becomes stationary. Or stated differently, the time series exhibits no trend with constant variance and autocorrelation structures over time (Wilks 2011). That is, by using 12-h differencing, the tendency of TCs to grow during their lifetime is removed. This growth tendency of TCs, as observed, is somewhat ubiquitous and has been related to warm air advection and vertical wind shear, extratropical transition and eyewall replacement cycles, importation of angular momentum, and prior interaction with land (Maclay et al. 2008; Chan and Chan 2013; Knaff et al. 2014, 2017), but complete understanding of TC growth over time is still in need of investigation. Growth in R5 has also been observed to be different in different basins (Knaff et al. 2014), and those trends are removed by our Markov process assumption.
To mitigate the removal of TC growth in our algorithm, we add the growth back in using a constant that can be added to each 12-h forecast. This constant is sometimes referred to as a “drift” in the case of random walk models. The years 2017–19 were used to determine the drift in each basin primarily because of the availability of higher quality and reanalyzed wind radii from JTWC (Sampson et al. 2017, 2018; Knaff et al. 2021). Table 2 shows the model drift values for each basin, which are constants added to the predicted value of fR5 at each 12-h step. These constants imply that the growth of TCs measured by fR5 is at a rate of 3.4%, 9.8%, 4.9%, 6.6%, and 8.8% every 12 h in the IO, WP, EP, AL, and SH TC basins, respectively.
Drift values for each basin determined from the 2017 to 2019 best track values of gale force winds. Values are applied every 12 h to the predicted value of fR5.
3. Independent validation and example forecasts
In this section, we show the validation of MMCL and DRCL for independent forecasts made during 2020–22 in WP, SH, EP, and AL. The validation is based on finalized wind radii best tracks from JTWC and NHC. The verification strategy follows Knaff and Sampson (2015) so that if any of the quadrants in the best track have nonzero wind radii, all quadrants for that case are verified. This strategy allows the individual quadrant statistics to be combined to form a single measurement of mean absolute error (MAE) and bias (i.e., the mean error) for each forecast lead time. Table 3 provides the number of quadrants verified for 34-, 50-, and 64-kt wind radii (R34, R50, and R64, respectively) at each forecast lead time for each basin. Even with a 3-yr sample, the number of cases available at longer leads and higher wind radii thresholds is relatively small. Short-duration TCs are most evident in the SH and EP, and this fact should be considered when interpreting MAE and bias. Since the verification uses four quadrants per verifying forecast point, 1200 points is the equivalent of 300 individual cases, and many researchers consider 300 cases as a minimum for stable statistics in TC verification. However, many R50 and R64 cases often have fewer than four nonzero quadrants. 900 and 600 may be closer to the number of points that indicate stable statistics. These breakpoints are indicated by boldface in Table 3. These guidelines are likely conservative given the autoregressive nature of MMCL where later forecasts depend on earlier ones. By this measure, the WP and AL R34, R50, and R64 statistics are reasonably stable to 96 h. This is not true for R50 and R64 in the EP and SH where the sample sizes are smaller.
The number of homogeneous quadrants used in the independent validation of wind radii forecasts made in 2020–22 for DRCL and MMCL stratified by basin and lead time. The number of MMCL quadrants is shown in italics. Boldface indicates where we feel the statistics are most stable.
The MAEs of the WP, SH, EP, and AL validation samples are shown in Fig. 3. The MAEs in the WP shown in Fig. 3a suggest that the MAEs associated with MMCL forecasts for all wind radii are nearly always smaller than those of DRCL. This is also true for the SH (Fig. 3b), EP (Fig. 3c), and AL (Fig. 3d). The largest MMCL versus DRCL MAE differences occur in the EP R34 forecasts and the R64 forecasts in all basins. The former is likely due to the better prediction of smaller storms that occur in this basin (Chavas et al. 2016, and references therein) and their smaller observed growth (Table 2 above). The latter is the result of an improved satellite-based R64 determination [(5)]. Similarly, smaller improvements are also noted for R50 due to an improved R50 formulation. In terms of errors, MMCL produces generally smaller errors for all wind radii.
The 2020–22 homogeneous wind radii MAEs from independent forecasts from DRCL (dashed) and MMCL (solid) for R34 (black), R50 (blue), and R64 (red) in (a) WP, (b) SH, (c) EP, and (d) AL. The number of cases is provided in Table 3.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
MAEs, however, are only half the story, a wind radii forecast system could produce smaller MAE by being generally smaller and more symmetric given our verification strategy. To further examine the MMCL versus DRCL comparison, biases from these same samples are shown in Fig. 4. One of the primary goals of this work was to reduce biases caused by the return to climatology that is a fundamental behavior of DRCL. Figure 4 shows that in general this goal has been accomplished. Large improvements in R34 biases are seen in MMCL relative to DRCL in the SH (Fig. 4b), EP (Fig. 4c), and AL (Fig. 4d). In both the SH and AL, large DRCL negative biases are replaced by smaller MMCL high biases. In the EP, MMCL biases are generally smaller for all times and radii. In the WP, R34 biases for MMCL are slightly higher than for DRCL, but the errors (Fig. 3a) are comparable or slightly smaller. The WP DRCL was recently updated with more data, so this is expected. High biases in R50 in DRCL are noticeably reduced in WP (Fig. 4a) and EP (Fig. 4c), whereas initial biases are reduced at short forecast times and increase slightly in SH (Fig. 4b), again noting that the numbers of cases at long leads in the SH are less representative of the initial sample. And R50 biases in the AL (Fig. 4d) are largely unchanged. With respect to R64, in all basins, MMCL reduces biases closer to zero.
As in Fig. 3, but for biases.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
These results are encouraging, but does MMCL produce climatological average circles like DRCL at the longer forecast periods? To examine this, we present specific 120-h forecasts that are representative of TC forecasts for each basin.
The first case examined is Hurricane Larry (AL122021) initialized at 1200 UTC 6 September 2021. At this time, Larry was a 105-kt hurricane located in a favorable environment and was forecast to increase 5 kt in intensity in the next 12 h following an eyewall replacement cycle, remained a major hurricane through 48 h, and weakened thereafter as it was forecast to move steadily northwest and recurve after 72 h. The storm was expected to approach, but not strike, eastern Canada and had a typical size with gales extending to 150 n mi in the northeast and southeast quadrants. In the end, Larry made landfall in Newfoundland with maximum winds of 70 kt and a forward speed of 40 kt (Brown 2021). Figure 5a shows the 6-hourly best track from 1200 UTC 6 September to 1200 UTC 11 September. The DRCL forecast is shown in Fig. 5b, and the forecast shows only very slight growth of the wind radii except at 96 and 120 h, reflecting a scenario very close to climatology. On the other hand, the MMCL forecast (Fig. 5c) shows pronounced growth after initialization and ultimately a larger storm beyond 48 h. Also note that MMCL forecasts have smaller R64 than DRCL through 4 days, and that these are closer to the best track values. So, in this case, MMCL, by not returning to a climatological solution, made a better forecast.
The black polygons show R34, R50, and R64 moving inward from the outermost polygon. The best track information is shown for every 6 h, and the DRCL and MMCL forecasts are shown at 0, 12, 24, 36, 48, 60, 72, 96, and 120 h. DRCL and MMCL forecasts are independent of their developmental data. Depictions of the 5-day (a) best track, (b) DRCL forecasts, and (c) MMCL forecasts for Hurricane Larry (AL122021) initialized at 1200 UTC 6 Sep 2021 and (d) 5-day best track, (e) DRCL forecasts, and (f) MMCL forecasts for Tropical Storm Maysak (WP102020) initialized at 1200 UTC 28 Aug 2020.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
The second case shown is Tropical Storm Maysak (WP102020) initialized at 1200 UTC 28 August 2020 with a 35-kt intensity; Maysak was then forecast to intensify to 70 kt in 24 h and to 110 kt by 72 h. It was also expected to move nearly due north, affecting smaller Japanese islands, Kyushu, and South Korea—a very important forecast for JTWC. At initialization, Maysak was large and asymmetric with gales extending 90, 30, 120, and 40 n mi in the northeast, southeast, southwest, and northwest quadrants, respectively. The 5-day best track depicts the storm symmetrizing over the next few days, with little indication of an increase in size, but eventually growing larger off the coast of South Korea (Fig. 5d). The JTWC intensity forecast (input into both DRCL and MMCL) indicated a more rapid attainment of 110 kt than was observed, then a weaker storm at days 3–5. In this scenario, DRCL (Fig. 5e) predicts a large typhoon, one that is typical of the new climatology (Knaff et al. 2018) and continued growth of the gales through 5 days. MMCL (Fig. 5f), on the other hand, forecasts a smaller and more realistic evolution as Maysak intensified rapidly by indicating slower growth during intensification. The MMCL forecast also produced much smaller R50 and R64 forecasts, which were much closer to those depicted in the best track. Here again, MMCL produced a smaller and more realistic forecast when the TC was intensifying. This was expected as the dVm term is negative in Table 1 and acts to slow growth during intensification.
The final case shown is a forecast for Hurricane Darby (EP052022) initialized at 1200 UTC 12 July 2022 with a 115-kt intensity and very small wind radii (R34 ≤ 40 n mi, R64 = 10 n mi). The TC was expected to move slightly north of due west with steady and then more rapid weakening with the possibility of a very weak TC affecting the Hawaiian Islands. The best track (Fig. 6a), however, indicates that the storm moved slightly south of west weakened for 18 h, then intensified to 100 kt at 1800 UTC 13 July 2022 where it remained for 6 h, and then began to weaken (Bucci 2021). During this time, the wind radii became slightly smaller. The differences between the best track and forecasts affected both DRCL (Fig. 6b) and MMCL (Fig. 6c), but in different ways. As DRCL relaxed to a larger climatology, MMCL grew much slower. The higher anticipated latitude resulted in larger wind radii forecasts in both, but the forecast of weakening, also resulted in larger wind radii in MMCL. Here, however, the largest errors are associated with DRCL’s tendency to approach a climatological mean at longer forecast lengths. MMCL, free of this constraint, produced more realistic wind radii.
As in Fig. 5, except that depictions are of the 5-day (a) best track, (b) DRCL forecasts, and (c) MMCL forecasts for Hurricane Darby (EP052022) initialized at 1200 UTC 12 Jul 2022.
Citation: Weather and Forecasting 40, 2; 10.1175/WAF-D-24-0078.1
R50 and R64 forecasts remained much smaller in Darby until the vortex weakened to a tropical storm. MMCL produces more realistic and better-performing forecasts than DRCL for Darby because MMCL uses separate R5-based estimates of R34, R50, and R64 in (5), which is unlike DRCL, and creates more realistic and smaller R50 and R64 forecasts. Figure 5 also shows similar improvements in MMCL R50 and R64 forecasts in Hurricane Larry and Tropical Storm Maysak. These results are consistent with the improved biases shown in Fig. 4.
The summary statistics and case studies presented above indicate that the MMCL baseline is in most cases a superior performing and more realistic baseline forecast of wind radii. With that information in mind, we now provide a short summary and discussion.
4. Summary, discussion, and future work
This paper discusses the development of a skill baseline tropical cyclone wind radii method, MMCL, that addresses a known shortcoming of the more traditional climatology- and persistence-based DRCL. DRCL’s persistence component is rather short-lived, and the model returns to a climatological vortex following about 48 h. In nature, TC wind fields tend to grow, although at different rates in different basins (Table 2). Large storms rarely shrink to a climatological size for any time other than during their decay. In extremely large TCs, the DRCL forecast wind radii are generally too small, while in extremely small TCs, the DRCL forecast wind radii are generally too big. Examples of these effects are shown in the extremely large Hurricane Sandy (2012) case in Fig. 1, and more typical cases are provided in Figs. 5 and 6.
The solution developed here uses a satellite record of TC size to develop a 12-h forecast of TC size change based on current and routinely available information and developing a Markov chain of 12-h forecasts along any intensity and track forecast of any length. As part of the development, a climatological drift for each basin is also added to each 12-h forecast to account for tendency for TCs to grow as a function of time. Thus, MMCL provides a more realistic baseline forecast based on information provided from the initial conditions and the official forecast and that does not revert to a climatological size.
Results based on 3 years of independent forecasts suggest that this model produces forecasts with generally smaller MAEs and improved biases for all wind radii thresholds. The largest MAE improvements of MMCL compared with DRCL are found in the EP and AL basins; however, biases are most improved in the SH, EP, and AL basins, particularly for R34 and R64.
MMCL is in the operational forecast guidance suite at JTWC for TCs occurring in the WP, IO, and SH basins and can produce forecasts in the AL and EP. While its performance represents only a small improvement in wind radii forecast baseline capability, it addresses known shortcomings of DRCL with extremely large or extremely small TCs that persist beyond 2 days.
MMCL forecasts, while simple and representing a baseline, may aid in the operational wind radii consensus forecasts (RVCN; Sampson and Knaff 2015) or replace the DRCL member in the consensus. It is particularly useful for longer leads where DRCL reverts to climatology. It also has realistic forecasts for inclusion in RVCN, particularly since it provides stability in the forecast when NWP model forecast intensities decay and their radii drop out of the RVCN consensus. Both DRCL and MMCL can be easily extended to longer forecast lead times, but whereas in DRCL longer-lead forecasts would just reflect the static climatology of TC size, MMCL produces wind radii forecasts that are physically consistent with the official forecast information and with the initial conditions. Thus, as official forecasts improve, MMCL will also likely improve. With that in mind, the MMCL formulation and similar strategies could be applied to synthetic tracks created to assess wind risks for real-time forecasts (e.g., in wind radii realizations for the operational wind speed probabilities; DeMaria et al. 2013) or even insurance risks and climate change scenarios. Finally, MMCL provides a more realistic baseline for verification of wind radii forecasts, one that can be easily extended for forecasts beyond 5 days.
Acknowledgments.
This work was accomplished while the authors were employed by the NOAA Center for Satellite Applications and Research to support several tropical cyclone applications and at the Naval Research Laboratory, Monterey, California, supported by ONR Award N0001420WX00517 to NRL Monterey and the NOAA Cooperative Institute for Research in the Atmosphere at Colorado State University (NOAA Grant NA17RJ1228). We also thank Chris Landsea and the other anonymous reviewer for their careful and constructive review of the manuscript. The scientific results and conclusions, as well as any views or opinions expressed herein, are those of the authors and do not necessarily reflect those of the NOAA or the Department of Commerce.
Data availability statement.
This publication uses data from geostationary satellites contained in the STAR/CIRA tropical cyclone IR image archive. Those data, which are in a non-standard format and reduced bit precision, are available upon request. Alternatively, raw IR data are available from NOAA CLASS (https://www.avl.class.noaa.gov/saa/products/welcome), Multi-Format Client-Agnostic File Extraction Through Contextual HTTP (MCFETCH; https://mcfetch.ssec.wisc.edu/), and EUMETSAT (https://www.eumetsat.int/access-our-data). We also make use of tropical cyclone best tracks available that are openly available from JTWC (https://www.metoc.navy.mil/jtwc/jtwc.html?best-tracks), the National Hurricane Center, and the Central Pacific Hurricane Center (https://ftp.nhc.noaa.gov/atcf/).
REFERENCES
Aberson, S. D., 1998: Five-day tropical cyclone track forecasts in the North Atlantic basin. Wea. Forecasting, 13, 1005–1015, https://doi.org/10.1175/1520-0434(1998)013<1005:FDTCTF>2.0.CO;2.
Brown, D. P., 2021: Tropical cyclone report: Hurricane Larry (AL122021), 31 August–11 September 2021. NOAA NWS NHC Tech Rep., 34 pp., https://www.nhc.noaa.gov/data/tcr/AL122021_Larry.pdf.
Bucci, L., 2021: Tropical cyclone report: Hurricane Darby (EP052022), 9–17 July 2022. NOAA NWS NHC Tech Rep., 16 pp., https://www.nhc.noaa.gov/data/tcr/EP052022_Darby.pdf.
Chan, K. T. F., and J. C. L. Chan, 2013: Angular momentum transports and synoptic flow patterns associated with tropical cyclone size change. Mon. Wea. Rev., 141, 3985–4007, https://doi.org/10.1175/MWR-D-12-00204.1.
Chavas, D. R., N. Lin, W. Dong, and Y. Lin, 2016: Observed tropical cyclone size revisited. J. Climate, 29, 2923–2939, https://doi.org/10.1175/JCLI-D-15-0731.1.
Chu, J.-H., 1994: A regression model for the western North Pacific tropical cyclone intensity forecasts. NRL Memo. Rep. 7541-94-7215, 33 pp.
DeMaria, M., J. A. Knaff, R. D. Knabb, C. Lauer, C. R. Sampson, and R. T. DeMaria, 2009: A new method for estimating tropical cyclone wind speed probabilities. Wea. Forecasting, 24, 1573–1591, https://doi.org/10.1175/2009WAF2222286.1.
DeMaria, M., and Coauthors, 2013: Improvements to the operational tropical cyclone wind speed probability model. Wea. Forecasting, 28, 586–602, https://doi.org/10.1175/WAF-D-12-00116.1.
Elsner, J. B., T. H. Jagger, M. Dickinson, and D. Rowe, 2008: Improving multiseason forecasts of North Atlantic hurricane activity. J. Climate, 21, 1209–1219, https://doi.org/10.1175/2007JCLI1731.1.
Jarvinen, B. R., and C. J. Neumann, 1979: Statistical forecasts of tropical cyclone intensity for the North Atlantic basin. NOAA Tech. Memo. NWS NHC-10, 24 pp., https://repository.library.noaa.gov/view/noaa/6555.
Knaff, J. A., and C. R. Sampson, 2015: After a decade are Atlantic tropical cyclone gale force wind radii forecasts now skillful? Wea. Forecasting, 30, 702–709, https://doi.org/10.1175/WAF-D-14-00149.1.
Knaff, J. A., M. DeMaria, C. R. Sampson, and J. M. Gross, 2003: Statistical, 5-day tropical cyclone intensity forecasts derived from climatology and persistence. Wea. Forecasting, 18, 80–92, https://doi.org/10.1175/1520-0434(2003)018<0080:SDTCIF>2.0.CO;2.
Knaff, J. A., C. R. Sampson, M. DeMaria, T. P. Marchok, J. M. Gross, and C. J. McAdie, 2007: Statistical tropical cyclone wind radii prediction using climatology and persistence. Wea. Forecasting, 22, 781–791, https://doi.org/10.1175/WAF1026.1.
Knaff, J. A., S. P. Longmore, and D. A. Molenar, 2014: An objective satellite-based tropical cyclone size climatology. J. Climate, 27, 455–476, https://doi.org/10.1175/JCLI-D-13-00096.1.
Knaff, J. A., S. P. Longmore, R. T. DeMaria, and D. A. Molenar, 2015: Improved tropical-cyclone flight-level wind estimates using routine infrared satellite reconnaissance. J. Appl. Meteor. Climatol., 54, 463–478, https://doi.org/10.1175/JAMC-D-14-0112.1.
Knaff, J. A., C. J. Slocum, K. D. Musgrave, C. R. Sampson, and B. R. Strahl, 2016: Using routinely available information to estimate tropical cyclone wind structure. Mon. Wea. Rev., 144, 1233–1247, https://doi.org/10.1175/MWR-D-15-0267.1.
Knaff, J. A., C. R. Sampson, and G. Chirokova, 2017: A global statistical–dynamical tropical cyclone wind radii forecast scheme. Wea. Forecasting, 32, 629–644, https://doi.org/10.1175/WAF-D-16-0168.1.
Knaff, J. A., C. R. Sampson, and K. D. Musgrave, 2018: Statistical tropical cyclone wind radii prediction using climatology and persistence: Updates for the western North Pacific. Wea. Forecasting, 33, 1093–1098, https://doi.org/10.1175/WAF-D-18-0027.1.
Knaff, J. A., and Coauthors, 2021: Estimating tropical cyclone surface winds: Current status, emerging technologies, historical evolution, and a look to the future. Trop. Cyclone Res. Rev., 10, 125–150, https://doi.org/10.1016/j.tcrr.2021.09.002.
Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS). Bull. Amer. Meteor. Soc., 91, 363–376, https://doi.org/10.1175/2009BAMS2755.1.
Kossin, J. P., J. A. Knaff, H. I. Berger, D. C. Herndon, T. A. Cram, C. S. Velden, R. J. Murnane, and J. D. Hawkins, 2007: Estimating hurricane wind structure in the absence of aircraft reconnaissance. Wea. Forecasting, 22, 89–101, https://doi.org/10.1175/WAF985.1.
Landsea, C. W., and J. L. Franklin, 2013: Atlantic hurricane database uncertainty and presentation of a new database format. Mon. Wea. Rev., 141, 3576–3592, https://doi.org/10.1175/MWR-D-12-00254.1.
Lee, C., K. K. W. Cheung, W. Fang, and R. L. Elsberry, 2010: Initial maintenance of tropical cyclone size in the western North Pacific. Mon. Wea. Rev., 138, 3207–3223, https://doi.org/10.1175/2010MWR3023.1.
Maclay, K. S., M. DeMaria, and T. H. Vonder Haar, 2008: Tropical cyclone inner-core kinetic energy evolution. Mon. Wea. Rev., 136, 4882–4898, https://doi.org/10.1175/2008MWR2268.1.
Merrill, R. T., 1980: A statistical tropical cyclone motion forecasting system for the Gulf of Mexico. NOAA Tech. Memo. NWS NHC 14, 23 pp., https://repository.library.noaa.gov/view/noaa/7028.
Neumann, C. J., 1972: An alternate to the HURRAN (hurricane analog) tropical cyclone forecast system. NOAA Tech. Memo. NWS SR-62, 28 pp., https://repository.library.noaa.gov/view/noaa/3605.
Sampson, C. R., and A. J. Schrader, 2000: The Automated Tropical Cyclone Forecasting System (Version 3.2). Bull. Amer. Meteor. Soc., 81, 1231–1240, https://doi.org/10.1175/1520-0477(2000)081<1231:TATCFS>2.3.CO;2.
Sampson, C. R., and J. A. Knaff, 2015: A consensus forecast for tropical cyclone gale wind radii. Wea. Forecasting, 30, 1397–1403, https://doi.org/10.1175/WAF-D-15-0009.1.
Sampson, C. R., E. M. Fukada, J. A. Knaff, B. R. Strahl, M. J. Brennan, and T. P. Marchok, 2017: Tropical cyclone gale wind radii estimates for the western North Pacific. Wea. Forecasting, 32, 1029–1040, https://doi.org/10.1175/WAF-D-16-0196.1.
Sampson, C. R., J. S. Goerss, J. A. Knaff, B. R. Strahl, E. M. Fukada, and E. A. Serra, 2018: Tropical cyclone gale wind radii estimates, forecasts, and error forecasts for the western North Pacific. Wea. Forecasting, 33, 1081–1092, https://doi.org/10.1175/WAF-D-17-0153.1.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. 3rd ed. Elsevier, 676 pp.
