Monkeypox Study

BACKGROUND AND MOTIVATION FOR STUDY

Monkeypox is a viral zoonotic (animal-to-human) disease that can also spread human-to-human, though at a much lesser and unsustainable rate. It normally occurs in parts of central and west Africa near tropical rainforests, and recent outbreaks in Sudan and the United States have fueled new research that focuses on environmental factors that likely contribute to the expanded geographical spread of this disease. Climate change is believed to be a significant driver to the geographical shift in Monkeypox prevalence, either by direct effects on the pathogen resulting from environmental changes of near-surface and soil conditions, or indirectly from the migration of carriers such as rodents seeking more favorable conditions. To further study the environmental influences on the spread of Monkeypox, CDC asked RAL to produce and verify five years (2012-2016) of WRF model simulations over equatorial Africa to obtain near-surface meteorological fields, with which to support their predictive disease modeling efforts.

WRF Simulation Setup

Figure 1 shows the WRF spatial domains, with 12-km grid spacing on the outer domain of 637 x 509 grid points, and a nested 4-km inner domain (used exclusively in all analysis and production) of 910 x 781 grid points. There are 74 vertical levels up to the model top at 20 hPa. The model timestep is 72 s for the 12-km domain and 24 s for the 4-km domain.

Figure 1. WRF domains. “d02” is the 4-km domain.
Figure 1. WRF domains. “d02” is the 4-km domain.

Initial and lateral boundary conditions for the WRF simulations come from the ERA5 Reanalysis, the fifth-generation climate reanalysis from the European Centre for Medium-Range Weather Forecasts (ECMWF). Output is obtained from the Research Data Archive (RDA) at the National Center for Atmospheric Research (NCAR). ERA5 output is at 31-km grid spacing on a 1280 longitude x 640 latitude N320 Gaussian grid, on 37 vertical pressure levels. ERA5 data is input to WRF every three hours.

WRF simulations are run for 7.5 days, with the first 12 hours discarded to give adequate time for the model to “spin-up.” Output is saved at hourly intervals. To help prevent the model from “drifting” from the observed state, spectral nudging is used. A small correction term is applied to the model solution in the top 12 model levels for geopotential, winds, temperature, and moisture every 6 hours, to “nudge” the model towards the ERA5 solution. Only upper tropospheric and stratospheric levels are nudged, to keep large-scale weather features in line with ERA5, while allowing WRF to generate its own solutions in the lower troposphere.

Model Verification at Boende Site

Initial and lateral boundary conditions for the WRF simulations come from the ERA5 Reanalysis, the fifth-generation climate reanalysis from the European Centre for Medium-Range Weather Forecasts (ECMWF). Output is obtained from the Research Data Archive (RDA) at NCAR. ERA5 output is at 31-km grid spacing on a 1280 longitude x 640 latitude N320 Gaussian grid, on 37 vertical pressure levels. ERA5 data is input to WRF every three hours.

WRF simulations are run for 7.5 days, with the first 12 hours discarded to give adequate time for the model to spin up. Output is saved at hourly intervals. To help prevent the model from drifting from the observed state, spectral nudging is used. A small correction term is applied to the model solution in the top 12 model levels for geopotential, winds, temperature, and moisture every 6 hours, to nudge the model toward the ERA5 solution. Only upper tropospheric and stratospheric levels are nudged, to keep large-scale weather features in line with ERA5, while allowing WRF to generate its own solutions in the lower troposphere.

Model Verification at Boende Site

A comparison was done between weather observations obtained from CDC staff at Boende and output from the nearest WRF grid point. The comparison period used is hourly from 0000 LST 6 September 2015 to 0000 LST 5 September 2016. The observation location is at latitude -0.282206° and longitude 20.883112°, and the nearest WRF grid point is centered at latitude -0.3095 and longitude 20.8604°. The distance between these points is approximately 4 km. Variables used in the comparison include near-surface air temperature, near-surface relative humidity, and rainfall. Wind speed was not used due to quality control issues.

Figure 2 shows daily average plots of temperature and relative humidity, and daily sum plots of precipitation, for the observations and for WRF at Boende throughout the verification period. The bias, RMSE, and correlation for the three variables, calculated across the verification period from hourly values, is shown in Table 1. WRF reproduces the seasonal march of temperature, increasing from the boreal winter dry season through the long-rains season before decreasing again into boreal summer. WRF is overestimating most nighttime low temperatures, typically by 1°C-2°C, but sometimes by more. Some daytime highs are underestimated by about 1°C as well. WRF underestimates daily average relative humidity, although this may reflect the nighttime positive temperature bias, as highest relative humidity values would be expected at night. The source of the larger negative model bias in relative humidity from May-August 2016 is not clear. Climatologically, this is typically a drier period for much of the DRC (e.g., Dezfuli 2017). WRF underestimates the magnitude of rainfall at Boende, more so than underestimating the number of rain days, as there are 147 rain days in the observations and 126 rain days in the WRF output. WRF was missing rain events in May and June that largely accounts for the negative bias.

Sources of the temperature and relative humidity biases may stem from differences in the physical environment around the observation site from that represented by the closest gridpoint in WRF. From Google Earth, the observation location appears to be inside Boende, a town of about 30,000 people. This location has built-up structures around the observation site, and is largely cleared of foliage. In contrast, the grid point from WRF is primarily representing deciduous broadleaf forest, which has a shade fraction of 80%. Effects from not only shading differences, but also differences in emission of longwave radiation from surrounding surfaces and other anthropogenic effects may be in play. Precipitation, and atmospheric moisture in general, is the most difficult atmospheric process to model and observe accurately. The negative bias in WRF may result from inaccurate input conditions, model deficiencies in representing the precipitation generating processes in this area, or local effects. Observational errors do not appear to be an issue, as CMORPH precipitation for the closest grid point over this time period is larger than the observation (2716.86 mm vs. 2238.49 mm). Local effects, and the lack of representation of such effects in the model, may be in play. Both observed and WRF-simulated rainfall can vary substantially over small areas, and a few large events can greatly alter the annual rainfall total. Further validation could be done with comparison between WRF precipitation and those from gridded datasets like CMORPH, however these datasets have their own uncertainties as well.

 

Figure 2. Plots of daily average near-surface temperature (top left), daily average near- surface relative humidity (top right), and daily sum precipitation (bottom left) for the September 2015-September 2016 verification period at Boende. Red line represents observations and blue line represents WRF.
Figure 2. Plots of daily average near-surface temperature (top left), daily average near-surface relative humidity (top right), and daily sum precipitation (bottom left) for the September 2015-September 2016 verification period at Boende. The red line represents observations and the blue line represents WRF.
 
Table 1. WRF bias, RMSE, and correlation with Boende observations for the 2015-2016 verification period.
Table 1. WRF bias, RMSE, and correlation with Boende observations for the 2015-2016 verification period.