• The DTC conducted an extensive testing and evaluation activity to compare the error characteristics of two sets of runs: a control configuration very similar to the HWRF 2014 operational configuration (HDGF), and a variant (HDRF) in which the RRTMG radiation parameterization was used in lieu of the GFDL radiation parameterization and a partial cloudiness scheme was added.

• The alteration in radiation parametrization was motivated by findings by Bu et al. (JAS 2014; doi: http://dx.doi.org/10.1175/JAS-D-13-0265.1) regarding deficiencies in the representation of longwave cloud radiative forcing when the GFDL radiation parameterization is employed.

• A partial cloudiness scheme (see DTC report to HFIP) was added because HWRF underforecasts the presence of clouds, especially stratus clouds off the western coast of North and South America. This deficiency is not unique to HWRF – it is well known that many forecast models have trouble representing stratus clouds, in part because of their relatively coarse horizontal and vertical grid spacing.

• The partial cloudiness parameterization used in HDRF also remedied a lack of cloud radiative forcing in the RRTMG parameterization associated with clouds created in the SAS cumulus parameterization. It should be noted that the GFDL radiation parameterization employed in HDGF provides cloud radiative forcing associated with SAS clouds.

• Another difference between the HDGF and HDRF is the frequency of calls to the radiation parameterization: 60 minutes in HDGF and 15 minutes in HDRF. The higher frequency employed in HDRF is justifiable to represent meteorological processes with higher spatial and temporal time scales present in the HWRF 3-km inner nest.

• For each of the two HWRF configurations, 198 retrospective forecasts for the 2011-2014 seasons of the eastern North Pacific and North Atlantic basins were conducted in order to produce a large sample from which robust conclusions could be derived.

• Track errors grow steadily with forecast lead time for both configurations, reaching 200 nm in the North Atlantic and 175 nm in the eastern North Pacific. The average track errors are indistinguishable between the HDGF and HDRF configurations.

• In the North Atlantic both configurations have near zero intensity bias throughout the forecast period, with HDRF on average producing stronger storms. In the eastern North Pacific, both configurations have near zero bias at the initial time, with the bias becoming increasingly negative over time, stabilizing at approximately -10 kt at 72 h. Intensity bias is similar for both configurations up until 72 h, with the HDRF configuration producing stronger storms later in the forecast, which slightly improves the already negative bias.

• The intensity mean absolute error grows in the first 48 h of the forecast, and becomes stable thereafter. Similarly to the intensity bias, the intensity mean absolute error is larger in the eastern North Pacific, where it reaches 16 kt, than in the North Atlantic, where it is capped at 13 kt. For the most part, there are no statistically significant differences in intensity mean absolute error between the two configurations. In the North Atlantic, the intensity mean absolute error is smaller for the HDRF configuration than for the HDGF configuration in days 1 and 2, but larger at day 3. In the eastern North Pacific basin, the intensity mean absolute error is statistically significantly smaller for the HDRF configuration at the 84, 108, and 120 h forecast lead times. The results are virtually identical when all forecasts are aggregated versus when only the forecasts over water are aggregated, indicating that the intensity errors are not explained by erroneous landfall.

• In the North Atlantic, the 34-kt radii are only statistically significantly different from zero in the SW quadrant of the storm, where they are too large for both configurations, but larger on average for HDRF. In the East Pacific the storms tend to start with too large 34-kt radii (in some quadrants as large as 30 nm on average) and shrink during the forecast, ending with bias near zero or negative (up to -20 nm in the SE quadrant). No systematic differences are noted between the two configurations in the East Pacific.

• The results for the 50- and 64-kt radii need to be interpreted with caution due to limited sample size. In the North Atlantic, the storms are statistically significantly too large in all quadrants, with the HDRF configuration producing larger storms at most lead times and quadrants. In the eastern North Pacific the storms are also generally too large, except that the shrinking of storm size with forecast lead time is pronounced, resulting in smaller positive biases at days 4 and 5. In the eastern North Pacific there is less difference in storm size between the two configurations, but in a few quadrants and lead times the HDRF configuration produces larger storms (as seen in the North Atlantic).

• As far as structure absolute error, in both basins the distributions of the two configurations are statistically indistinguishable, even though the sample mean is slightly higher for the HDRF configuration for most quadrants and lead times. It should be noted structure absolute mean errors tends to decrease in the first few hours of the forecast, especially in the eastern North Pacific, indicating an area that should be targeted for model improvement.

• The temperature biases from both configurations over the continents are noticeably positive at 2-m and negative at lower isobaric levels. These biases increase in magnitude and spread over the ocean at later forecast lead times. This weakness of HWRF has been documented in the past, and may be related to the slab land parameterization used in HWRF, which may be too simplistic, and to insufficient vertical mixing in the planetary boundary layer over land.

• A positive temperature bias and a negative relative humidity bias are noted at 1000 hPa on the Southern and Northern Pacific, off the cost of the Americas, regions where a stratus deck is climatologically present. Further west, over the Central Pacific Ocean, the situation reverses, with negative temperature bias and positive RH bias. Interestingly, at 850 hPa the temperature bias is flipped with respect to its sign at 1000 hPa, with values too cold and with high RH near the coast, and too warm and with low RH over the Central Pacific. These biases are present in both the HDGF and HDRF configurations, but are alleviated in the HDRF configuration, in which the biases have smaller areal extent and magnitude. This suggests that the combination of partial cloudiness and RRTMG radiation has a positive impact on the representation of clouds over the Pacific Ocean.

• Throughout most of the domain, the HDRF configuration has lower geopotential height than the HDGF configuration in the lower troposphere, but higher geopotential height in the upper troposphere. This is consistent with HDRF having higher temperatures, and therefore higher thickness, throughout most of the troposphere. The HDRF higher temperatures are consistent with lower relative humidity in the mid to lower troposphere. These differences become larger at later forecast lead times.

• Temperature and relative humidity RMSE differences between the two configurations are inconclusive, with the exception of 1000 hPa over large areas of the Pacific Ocean, which show a meaningful improvement in both variables when the HDRF configuration is used. Geopotential height RMSE differences favor the control (HDGF) configuration at most levels, except for 250 hPa, where the HDRF configuration leads to improvements over the CONUS, the Pacific Ocean and the tropical belt.

• There were no systematic differences in wind speed RMSE between the two configurations.

• Model output files have been archived and are available to the community for future studies. They can be obtained at the NOAA HPSS at /HFIP/dtc-hurr/HGDF and /HFIP/dtc-hurr/HDRF.