The Intensification of Hurricane Maria 2017 in the Antilles The Intensification of Hurricane Maria 2017 in the Antilles

: Environmental influences on Hurricane Maria in the Antilles Islands are analyzed at the large-scale (1–25 September) and at the meso-scale (17–20 September 2017). The storm intensified rapidly prior to landfall in Dominica, going from category 1 to 5 in 15 h. As the storm progressed toward Puerto Rico (PR), its NE flank entrained air from seas cooled by the earlier passage of two hurricanes, and strengthened on its SW flank. Operational model forecasts tended to delay intensification until west of the Antilles Islands, thus motivating two independent weather research and forecasting (WRF) simulations. These gave minimal track errors at 1-to 3-day lead time. The simulation for landfall at Dominica on 19 September 2017 showed that a static nest with 0.8 km resolution using a Holland-type synthetic vortex and Yonsei University (YSU) / Kain-Fritsch schemes performed better; with a track error of 8 km and intensity error of 10 m / s. Our PR-area simulation of central pressure lagged 30 hPa behind observation; and caught up with reality by landfall in PR. The simulated rainband structure corresponded with Cloudsat observations over PR. Maria’s intensification occurred in an area of thermodynamic gradients included cooler SST in the right side of the track, so operational models with right-track bias were late in predicting intensification. Category-2 forecasts prior to 18 September 2017 left many Antilles islanders unprepared for the disaster that ensued.


Introduction
More than 15% of all Atlantic hurricanes make landfall in the Caribbean Antilles Islands during August-September. The most probable track is between Dominica (15 N, 61W) and Puerto Rico (18 N, 66 W). Destructive impacts are unavoidable on small islands exposed to high winds, flash-floods, landslides, and nowhere to evacuate. Anomalous warm sea surface temperatures (SST) and weak upper easterly winds promote intensification, amongst other factors [1][2][3]. Hurricane Maria (2017) was a prime example of rapid intensification-going from a category 1 to 5 in just 15 h.
Operational research has found increasing skill in hurricane forecasts at multi-day lead times when evaluated against persistence and climatology [4,5]. Track forecast errors in the Atlantic basin have decreased 1% a year since 1980 [6,7], while hurricane intensity errors declined 0.5% a year [8] except in the formative stage. It is well known that track is influenced by the large-scale environment whose forecasts benefit from steady progress in weather data assimilation. In contrast, hurricane intensity depends on how the inner-core thermodynamics and vorticity interact with the surrounding environment [9]. Although mesoscale structure can be resolved in numerical models with a horizontal resolution <4 km [10], the inner-core convective heating and vortex Rossby waves are seldom adequate. affect the marine environment and consequently the circulation of Maria? In Section 2 the data and methods are outlined. Section 3 depicts the results, and a concluding discussion is given in Section 4.

Data and Methods
The National Hurricane Center (NHC) records on hurricane track and intensity formed the initial basis for our research [25] and [20]. Hourly to daily rainfall estimates derived from: CMORPH (CPC MORPHing technique) and GPM-merged multi-satellite [26,27], and ECMWF-5 reanalysis [28]. Large-scale features were described by MERRA-2 atmospheric reanalysis [29] latent heat flux, convective available potential energy (CAPE), 925 hPa potential vorticity, 500 hPa specific humidity, 300 and 150 hPa winds, 150 hPa ozone and 100 hPa air temperature. Sea surface temperature was described using the NOAA satellite IR and MW measurements blended with in situ data [30]. Upper ocean conditions were described by HYCOM reanalysis [31]: Temperature, salinity, and currents that characterize upper ocean heating. Detailed wind analyses derived from the Ascat scatterometer [32] and in situ sensors on the National Data Buoy Center (NDBC) [33] moored buoy west of Guadeloupe (#42060) and at a coastal site (Arecibo harbor, PR). Sea surface height observations were extracted at the NOAA gauge [34] in Yabucoa harbor, PR. Estimated wave heights during hurricane transit were obtained from Wave-watch-3 reanalysis [35]. Weather radar reflectivity maps at Martinique, Guadeloupe and PR were obtained from MeteoFrance-Antilles and the US National Weather Service. A CloudSat reflectivity slice [36] was obtained during landfall over PR. USGS MODIS satellite vegetation fraction data [37] were analyzed for changes before and after the hurricane. USGS lake level data for Guayo, PR were analyzed and a back-trajectory analysis was performed using HYSPLIT in NOAA-Ready-ARL. Hurricane data from COAMPS-TC [21] were employed to inter-compare track and intensity forecasts, initialization data and observations. Similar comparisons using the FV-GFS model are available in [38]. The prior effect of hurricanes Irma and Jose on regional ocean conditions was analyzed over 5-15 September 2017 using latent heat flux anomalies and 925 hPa potential vorticity. Hourly time series 17-19 September 2017 from CFSr2 reanalysis [39,40] averaged over the intensification area east of Dominica, are given in the Appendix A.
For regional-scale PR-area simulations, the advanced research weather research and forecasting (WRF-ARW) v3.9 was configured with two nested grids covering the southeastern Caribbean at 3 km resolution (Appendix B), and PR at 1 km resolution with 39 vertical sigma levels from surface to 50 hPa, as suggested in [41]. The simulation was initialized using operational ECMWF system-5 fields at 9 km horizontal resolution, starting at 0Z 17 September 2017. The simulation utilized the WRF Single-Moment 6-class (WSM6) cloud microphysics scheme [42]. Other sub-model components include the Yonsei University (YSU) scheme for the planetary boundary layer [43], the Noah scheme for land surface physics, and the rapid radiative transfer model (RRTMG) scheme for radiation physics [44]. Forecast outputs included: Sea level air pressure, surface wind, and convection reflectivity, at lead times from 36 to 96 h. Hurricane resolving forecasts were compared with CIRA-NHC wind maps based on multi-satellite and aircraft reconnaissance data [45] in the period 17-20 September 2017.
A second independent modeling effort was conducted to simulate the intensification of Maria near Dominica in the period 18 00Z to at 19 00Z. These 'Dominica-area' experiments used two-way downscaling involving both static and moving nesting. The static nests (FIX) had three domains with respective resolutions of 7.5, 2.5, and 0.8 km (Appendix B). The moving nests (MOV) were similar, but followed the vortex with 15 min update. Two cumulus schemes were tested: KF [46] and NSAS [47]; and two boundary layer schemes were tested: YSU and MYNN [48]. After evaluation of 7.5 and 2.5 km resolution results, the optimal schemes were used in the 0.8 km resolution static nest experiment (FIX-YSU-KF). In all Dominica simulations the hurricane was initialized at 00Z 18 September with hybrid ETKF-3DVAR data assimilation [49] and a parametric Holland vortex [50,51]. As in the PR-area simulations described above, the outer domain of the Dominica simulation was initialized using operational 9 km ECMWF system-5 atmospheric fields and SSTs derived from NCEP-RTG. The WRF-ARW v4 model was configured from surface to 30 hPa, with 99 vertical levels in a logarithmic resolution that is finer in lower levels. The other physics parameterizations used here included the Monin-Obukhov similarity scheme with a strong wind Donelan-Garratt surface flux option [52], the RRTMG radiation scheme, and the WSM6 microphysics scheme.
Results were provided in sequence as large-scale precursor atmosphere and ocean conditions (2-30 N, 80-35 W), local conditions during intensification of hurricane Maria (10-22 N, 69-57 W), and observed and simulated hurricane structure on landfall in Dominica and PR and some impacts that followed. Our work followed a chronological format similar to the NHC report of [20], and provided additional insights on upper air and ocean influences, and model forecasts of hurricane intensity.

Observations of the Regional Environment
Hurricane Maria data on track and intensity in Table 1 Hovmoller plots in the 12-18 • N band 1-25 September 2017 (Figure 1a-d) provided useful context on the regional environment. The rainfall and 850 V wind plots identified prior hurricanes Irma and Jose passing steadily westward in the first half of the month. Maria on the other hand, developed further west and had declining speed. The marine environment was dominated by warm SST and low salinity to the west of 55 • W. The zonal gradient of SST was 27-30 • C from 40 to 70 • W, and the area of low salinity shifted slowly westward from 1 to 25 September 2017. It is apparent in Figure 1c that the (right) flank of each hurricane induced (cooler) SST due to evaporation and the vertical motion induced by surface wind stress. The 'pre-Maria' marine environment was analyzed with maps and sections of sea temperature, salinity, and currents averaged 5-15 September (Figure 2a-d). The SST field exhibited a >30 °C warm NW-SE axis in the Caribbean, coincident with lower salinity from the river plumes. SST north of the Antilles were <28 °C as a result of prior cooling by hurricanes Irma and Jose. The Antilles sections revealed a warm fresh layer from 0 to 40 m deep in latitudes 14-16 N. The ocean circulation was near climatology except for retroflection of the North Brazil Current. The Amazon discharge tended to disperse eastward, so fresh water in the Antilles was mainly sourced from the Orinoco plume. Considerable thermal energy was available for the rapid intensification of Maria, mainly in the SW quadrant as seen in Figure 3. The 'pre-Maria' marine environment was analyzed with maps and sections of sea temperature, salinity, and currents averaged 5-15 September (Figure 2a-d). The SST field exhibited a >30 • C warm NW-SE axis in the Caribbean, coincident with lower salinity from the river plumes. SST north of the Antilles were <28 • C as a result of prior cooling by hurricanes Irma and Jose. The Antilles sections revealed a warm fresh layer from 0 to 40 m deep in latitudes 14-16 N. The ocean circulation was near climatology except for retroflection of the North Brazil Current. The Amazon discharge tended to disperse eastward, so fresh water in the Antilles was mainly sourced from the Orinoco plume. Considerable thermal energy was available for the rapid intensification of Maria, mainly in the SW quadrant as seen in Figure 3.   Conditions prior to Maria formation were altered by the latent heat flux and wind vorticity from earlier hurricanes Irma and Jose (Figure 3a, b). Large swaths of thermal energy were lost along the NE edge of the Antilles in the period 5-15 September 2017. The anomalous evaporation swath was wider than the cyclonic vorticity swath. In contrast, there was negative evaporation anomalies and weak anticyclonic vorticity to the west of the Antilles that contributed to above normal SST and CAPE >3000 J/kg (Figure 3c,d). During the intensification of Hurricane Maria on 18 September 2017, backtrajectories spiraled inward toward the system from the SE. Yet many trajectories arrived from the NE quadrant characterized by negative SST anomalies and CAPE <1500 J/kg. As will be seen later, operational models with right-track bias entrained air from the low CAPE zone previously cooled by Irma and Jose.
The regional atmospheric environment prior to hurricane formation is illustrated in Figure 4a-c using 300 hPa wind and 500 hPa humidity maps on 12, 15, and 18 September 2017. On the 12th, upper level winds show a cold front had penetrated into the tropical Atlantic with a divergent northerly flow that was particularly dry. By 15 September, the frontal air mass had dissipated eastward, leaving remnants of low humidity in the mid-troposphere. Yet on 18 September the upper westerly trough re-established in the sub-tropical North Atlantic, so the zone of low wind shear was quite narrow in the vicinity of hurricane Maria. Conditions prior to Maria formation were altered by the latent heat flux and wind vorticity from earlier hurricanes Irma and Jose (Figure 3a,b). Large swaths of thermal energy were lost along the NE edge of the Antilles in the period 5-15 September 2017. The anomalous evaporation swath was wider than the cyclonic vorticity swath. In contrast, there was negative evaporation anomalies and weak anticyclonic vorticity to the west of the Antilles that contributed to above normal SST and CAPE >3000 J/kg (Figure 3c,d). During the intensification of Hurricane Maria on 18 September 2017, back-trajectories spiraled inward toward the system from the SE. Yet many trajectories arrived from the NE quadrant characterized by negative SST anomalies and CAPE <1500 J/kg. As will be seen later, operational models with right-track bias entrained air from the low CAPE zone previously cooled by Irma and Jose.
The regional atmospheric environment prior to hurricane formation is illustrated in Figure 4a-c using 300 hPa wind and 500 hPa humidity maps on 12, 15, and 18 September 2017. On the 12th, upper level winds show a cold front had penetrated into the tropical Atlantic with a divergent northerly flow that was particularly dry. By 15 September, the frontal air mass had dissipated eastward, leaving remnants of low humidity in the mid-troposphere. Yet on 18 September the upper westerly trough re-established in the sub-tropical North Atlantic, so the zone of low wind shear was quite narrow in the vicinity of hurricane Maria.

Observations of Conditions in the Eastern Antilles
Rainfall estimates from one model and two satellite products averaged 17-20 September (Figure  6a-c) indicate fluctuations in intensity related to moisture convergence and spiral bands joining the hurricane. CMORPH shows pulsing along the track, while most ECMWF5 rainfall lies on the northern flank. The GPM product exhibits a large area of intense rainfall between Dominica and PR. In general, the rainfall estimates exceed 50 mm/day over a width of ~200 km, except for CMORPH, which was narrower. Wind scatterometer maps are given in Figure 7a,b during Maria passage over the eastern Antilles on 19 September. Storm winds had a circular shape with steady intensity and slightly greater

Observations of Conditions in the Eastern Antilles
Rainfall estimates from one model and two satellite products averaged 17-20 September (Figure 6a-c) indicate fluctuations in intensity related to moisture convergence and spiral bands joining the hurricane. CMORPH shows pulsing along the track, while most ECMWF5 rainfall lies on the northern flank. The GPM product exhibits a large area of intense rainfall between Dominica and PR. In general, the rainfall estimates exceed 50 mm/day over a width of~200 km, except for CMORPH, which was narrower.

Observations of Conditions in the Eastern Antilles
Rainfall estimates from one model and two satellite products averaged 17-20 September (Figure  6a-c) indicate fluctuations in intensity related to moisture convergence and spiral bands joining the hurricane. CMORPH shows pulsing along the track, while most ECMWF5 rainfall lies on the northern flank. The GPM product exhibits a large area of intense rainfall between Dominica and PR. In general, the rainfall estimates exceed 50 mm/day over a width of ~200 km, except for CMORPH, which was narrower. Wind scatterometer maps are given in Figure 7a,b during Maria passage over the eastern Antilles on 19 September. Storm winds had a circular shape with steady intensity and slightly greater Wind scatterometer maps are given in Figure 7a,b during Maria passage over the eastern Antilles on 19 September. Storm winds had a circular shape with steady intensity and slightly greater radius in the NE quadrant as expected. Wave heights reached 10 m (Figure 7c) and pounded the southern shore of Guadeloupe during passage. Winds at the NOAA buoy west of Guadeloupe (Figure 7d) reached 45 m/s rather symmetrically. The WRF simulated and MeteoFrance observed radar reflectivity (Figure 7e,f) show intense spiral bands on the hurricane's northern flank hitting Dominica. The simulated central pressure was 976 hPa regional (3 km resolution without synthetic vortex assimilation)/962 hPa local (0.8 km resolution with synthetic vortex assimilation) in versus 924 hPa observed (at 00Z 19 September 2017).

Dominica Simulation and Forecasts
The Dominica area WRF model experiments show good agreement with NOAA NHC best track during the 24 h of simulation from 00Z 18 to 19 September 2017. The simulated hurricane wandered north then south of track (Figure 8a, Table 2) and under-estimated the intensification of Maria. The static nest with YSU-KF schemes gave strongest winds at the Dominica landfall (Figure 8b, 63 m/s).

Dominica Simulation and Forecasts
The Dominica area WRF model experiments show good agreement with NOAA NHC best track during the 24 h of simulation from 00Z 18 to 19 September 2017. The simulated hurricane wandered north then south of track (Figure 8a, Table 2) and under-estimated the intensification of Maria. The static nest with YSU-KF schemes gave strongest winds at the Dominica landfall (Figure 8b, 63 m/s). We also examined forecasts from three operational models (Figure 8c,d) initialized at 00Z 18 September 2017. The HMON intensity forecast valid for 00Z 19 September 2017 (+24 h) had category 4 winds (62 m/s) followed by HWRF (51 m/s). Differences in performance related to north vs. south track bias and the SST gradient (Figure 3c). Landfall was predicted by HWRF (HMON) for Guadeloupe (Martinique), so disaster preparations became uncertain.
Atmosphere 2019, 10, x FOR PEER REVIEW 11 of 22 We also examined forecasts from three operational models (Figure 8c,d) (Figure 3c). Landfall was predicted by HWRF (HMON) for Guadeloupe (Martinique), so disaster preparations became uncertain.       Table 3 lists key features of COAMPS initialization data from 17 to 20 September 2017. Storm positions and intensities followed NHC best track (Table 1) but the heading wandered~10 • more than observed. The zonal wind shear changed from +7.3 m/s on 12Z 17 September to −4.5 m/s by 00Z 19 September. The SST and 850 relative vorticity dipped to 28.6 • C and −2 × 10 −5 s −1 on 17 September, suggesting dissipation of storm energy near waters cooled by earlier hurricanes. The SST initialization values were~1 • C below observed (Figure 2a,c) and contributed to delayed intensification forecasts (Figure 8c). In contrast, precipitable water and 200 hPa divergence remained steady. Cross-correlations between the time series of initialization data (N = 14) reveal that maxWind, SLPmin and 850 vorticity are sensitive to small changes in SST (r = +0.83, −0.85, and +0.82, respectively).

Structure at PR Landfall
Conditions at landfall in PR (Figure 9a-d) indicate the hurricane had grown in size and intensity. Simulated and observed radar reflectivity show at least two sets of circular cloud bands, consistent with eyewall replacement. 10 m waves and a 3 m storm surge pounded the eastern shore of the island. Winds on the north coast of PR exceeded 45 m/s and indicated the storm remained intense after landfall. The simulated vs. observed central pressure at landfall in PR was 932 hPa vs. 913 hPa, hence the 3-day forecast intensity was still weak but the track was good. Hurricane Maria's structure was analyzed by cross sections on 66.5 W at 18Z 20 September (Figure 10a,b). The PR-area WRF forecast radar reflectivity had a deep convection associated with the expanded eyewall after the PR landfall. Cloudsat radar reflectivity measured the deepest convection (>12 km) on the Caribbean (south) side of the hurricane. A noteworthy feature was the discrepancy in the level of maximum reflectivity where WRF had it mostly in the rain areas while CloudSat had it in the upper clouds, probably near sub-freezing. It is suspected that the simulation may not produce enough graupel in the upper clouds, even though the WSM6 scheme included graupel. [53] used satellite microwave sounders to measure Hurricane Maria's warm core anomaly and found upper tropospheric values of 8-12 °C from 19 to 25 September 2017. Hurricane Maria's structure was analyzed by cross sections on 66.5 W at 18Z 20 September (Figure 10a,b). The PR-area WRF forecast radar reflectivity had a deep convection associated with the expanded eyewall after the PR landfall. Cloudsat radar reflectivity measured the deepest convection (>12 km) on the Caribbean (south) side of the hurricane. A noteworthy feature was the discrepancy in the level of maximum reflectivity where WRF had it mostly in the rain areas while CloudSat had it in the upper clouds, probably near sub-freezing. It is suspected that the simulation may not produce enough graupel in the upper clouds, even though the WSM6 scheme included graupel. [53] used satellite microwave sounders to measure Hurricane Maria's warm core anomaly and found upper tropospheric values of 8-12 • C from 19 to 25 September 2017.

PR-Area Simulation and Impacts
As mentioned above, 48 h forecasts issued from 16 to 18 September based on a variety of models [20], contained track errors~30 km to the right, and intensity errors two categories too low. The ECMWF model had the left-most track but lacked intensity. HWRF forecasts caught up with the observed intensity at St Croix VI, with right-of-track errors and strongest winds in the NE quadrant.
Our PR-area WRF simulation with ECMWF initialization at 00Z 17 September 2017 compared with NHC central pressure (Figure 10c) under-predicted the hurricane intensification. The rate of deepening followed observed but central pressure was +30 hPa high. Just before landfall in PR (03Z 20 September) the WRF simulated winds reached 50 m/s, but the diameter was 50% below the CIRA-NHC hindcast (Figure 10d).
The satellite measured vegetation color fraction before and after hurricane passage in Dominica and PR illustrates a 30-50% reduction (Figure 11a-c), due to wind defoliation and uprooting. The change in NDVI (normalized difference vegetation index) appears to quantify hurricane damage rather well. Rainfall accumulated over the central mountains, where USGS measured lake levels and streamflows rose to unprecedented heights (Figure 11d). Landslides exceeded 25 /km 2 across half of PR [20], causing a loss of communication, water supply and electricity to~1 million people for~3 months,~3000 deaths, and~300,000 homes rendered uninhabitable [54].

Conclusions
Environmental influences on hurricane Maria in the Antilles Islands have been analyzed at large-scale in the precursor period 1-25 September 2017 and at meso-scale during intensification and landfall 17-20 September 2017. The under-predicted deepening before landfall in Dominica was related to low SST and CAPE on the NE flank, and rapid decline in upper-tropospheric wind shear (Figures 2-5, Table 2, Appendix A). As the storm passed St Croix, there was a sudden jump in rain rate ( Figure 6c). As shown in SST anomalies, a considerable thermal energy was available near Dominica (~60 W), which was mainly in the SW quadrant of Maria during 18-19 September 2017. Apparently, it was a positive factor that contributed to the rapid intensification of Maria. Hurricane Maria had a deep convection on its southern flank (Figure 10a,b), as destructive winds cut a swath across PR (Figure. 11c).

Conclusions
Environmental influences on hurricane Maria in the Antilles Islands have been analyzed at large-scale in the precursor period 1-25 September 2017 and at meso-scale during intensification and landfall 17-20 September 2017. The under-predicted deepening before landfall in Dominica was related to low SST and CAPE on the NE flank, and rapid decline in upper-tropospheric wind shear (Figures 2-5, Table 2, Appendix A). As the storm passed St Croix, there was a sudden jump in rain rate ( Figure 6c). As shown in SST anomalies, a considerable thermal energy was available near Dominica (~60 W), which was mainly in the SW quadrant of Maria during 18-19 September 2017. Apparently, it was a positive factor that contributed to the rapid intensification of Maria. Hurricane Maria had a deep convection on its southern flank (Figure 10a,b), as destructive winds cut a swath across PR (Figure 11c).
PR-area WRF simulations and operational models contained minimal track errors at multi-day lead times, as expected for a 295 • heading close to climatology. Most intensity forecasts lagged 20-30 hPa behind reality (Figure 10c), reaching category 5 by landfall in PR instead of Dominica. Given the delay, multiple experiments were conducted using different initializations, local domains with higher resolution, and alternative model physics. By assimilating a synthetic vortex in a static nest at 0.8 km resolution with YSU/KF schemes, we reduced forecast errors of central pressure on landfall in Dominica by 14 hPa as compared to the 3 km simulation without synthetic vortex initialization.
Our key findings pertain to thermodynamic gradients on Maria's NE flank due to earlier hurricanes Irma and Jose, and rapid evolution of the upper circulation. HWRF (HMON) landfall was predicted for Guadeloupe (Martinique) at one-day lead time, whereas Dominica suffered the impacts. Our WRF simulations had good track but lacked intensity. Category-2 forecasts prior to 18 September 2017 left many Antilles islanders unprepared for the disaster that ensued.