Development of Rapid Prototyping Capability to Evaluate Potential Uses of NASA Research Products and Technologies to Estimate Distribution of Mold Spore Levels over Space and Time

University of MississippiMedicalCenter, Fazlay S. Faruque,

For estimating the mold spore levels using NASA and NOAA data, we are collecting two types of ground data, meteorological and mold spore, to validate our estimation model. Currently, we have 6 ground-monitoring stations collecting these data. Base units collect temperature, humidity, dew point, rainfall, wind direction and speed (average and maximum gust) which are logged every 15 minutes. We encountered several problems with field data loggers, which were resolved to ensure quality data collection.

We have been collecting mold spore data from 6 sites in the Jackson metropolitan and surrounding areas since Saturday, November 3, 2007;with the exception of the Harrisville site which began operating on Thursday, November 22, 2007. As of April 25, 2008, we have collected 1055 daily samples from these sites. Because the tape attached to the rotating drum inside the Burkard air sampler has to be cut vertically for removal during the weekly collection, 1 of the days in a week is not a complete 24 hours. For consistency, the splicing occurs on Wednesday, and we have elected not to count the spores on these incomplete days. Each 24-hour section of tape is cut using a standardized grid and mounted on a glass slide for coverslip staining as specified in a standard protocol by the National Allergy Bureau (NAB), a section of the American Academy of Allergy, Asthma and Immunology's (AAAAI) Aeroallergen Network, that is responsible for reporting current pollen and mold spore levels to the public. With the exclusion of Wednesday, we have collected samples for mold spore enumeration and identification of weekly predominant species, including quantification of Alternaria species and qualitative assessment of the other frequently observed species. To date, we have counted spores from188 samples (days). Additionally, we have conducted 12 weeks of species identification including at least 1 week per site.

We have already found all 7 of the clinically relevant molds identified by the AAAAI Immunotherapy Committee. Cladosporium cladosporioides is the most frequently identified followed by Cladosporium herbarum, Alternaria alternata, Epicoccum nigrum, Helminthosporium, Aspergillus and Penicillium. Of the known naturally occurring mycotoxin producers, we have most frequently identified the tricothecene-producers, Fusarium and Stachybotrys as well as the aflatoxin producer Aspergillus.

Although weather station and remote sensing data are collected automatically, the mold spore counts involve considerable manual effort. Preliminary analysis was performed using the data for the timeframe from 11/3/2007–12/4/2007 from 4 of the 6 sites. The steps of the analysis were as follows:

  1. Weather and mold data were gathered, and variables available for all dates and locations were identified
  2. MODIS acquired data weredownloaded, processed and extracted. This MODIS product was atmospherically corrected reflectanceat 500 meter resolution for near infrared (MODIS Band 2, ~850 nm) and shortwave infrared (MODIS Band 6, ~1630 nm). The specific product designation was MOD09A1, indicating an 8-day compositing frame which was judged to be sufficient, since these reflectances are slow-varying. The input bands were used to calculate the Normalized Difference Moisture Index (NDMI): NDMI=(Band2Band6)/(Band2+Band6).
    Time series processing was applied to remove cloudy data and other low-quality observations and replace them with interpolated values. The NDMI values corresponding to specific places and times were extracted using nearest neighbor.
  3. Since there are over 2 dozen explanatory variables available, the field was narrowed by computing simple 1-variable relationships across all sites and dates and choosing 5 of the most significant for further study. These are: 1) hours of relative humidity 80%, 2)NDMI, 3)minimum dew point, 4)7-day cumulative rainfall, and 5) minimum temperature. When variables were related to the same phenomena, only 1 variable per distinct phenomenon was carried forward.
  4. Following the narrowing of the explanatory variables, a multivariate linear regression analysis was performed including all 4 sites (Table 1). All 5 variables together accounted for approximately 36% of the variation in spore counts. Reducing the number of variables in tested models systematically from 5 to 2, we found that all best models included both relative humidity and NDMI. Using those 2 variables alone still explained 31% of the variation in spore counts with both variables clearly significant (α = 0.05). By contrast, a model with the 4 weather variables alone accounted for only 21% of the spore count variance, with only relative humidity being clearly significant. Analyses were also carried out on a site-by-site basis, and the role of NDMI was diminished. This seems to indicate that NDMI helped account for spatial variation among sites.

Table 1: Multivariate linear regression analysis for mold spore estimation
p-value / p-value / p-value / p-value / p-value / p-value
R2 / F / Signif. F / Inter-cept / NDMI / RH / RNF / min-TMP / min-DEW
NDMI-RH-RNF-T-D / 0.362 / 10.87 / 2.6E-08 / 0.88 / 6.7E-06 / 0.0010 / 0.025 / 0.065 / 0.23
NDMI-RH-RNF-T / 0.352 / 13.16 / 1.3E-08 / 0.16 / 1.13E-05 / 0.0018 / 0.036 / 0.10
NDMI-RH-RNF / 0.334 / 16.38 / 1.0E-08 / 0.018 / 2.97E-05 / 5.9E-05 / 0.082
NDMI-RH / 0.313 / 22.55 / 8.5E-09 / 0.012 / 9.1E-05 / 1.62E-06
4wx / 0.211 / 6.46 / 0.00012 / 0.53 / 0.00085 / 0.29 / 0.41 / 0.61
NDMI = Normalized Difference Moisture Index
RH = hours of relative humidity >= 80%
RNF = 7-day cumulative rainfall / T = minimum air temperature
D = minimum dew point

The model as estimated using all 5 explanatory variables was estimated as follows:

Spore Count = -775.4 + (-83179.3*NDMI) + (984.7*RH) + (251.0*RNF) + (1040.5*T) + (-500.3*D)

The scatterplot of how this model performed is shown in Figure 1. While a trend is evident, it also appears that the uncertainty of the prediction cannot be assumed constant but is likely related to the magnitude of the prediction.

Given the short period sampled in this exploratory analysis, the results cannot be considered conclusive. The analyzed period has not captured the seasonal effects which are known to be important. However, the analysis does indicate that relationships between explanatory variables and mold spore counts appear to be present, and the outlook for deriving useful models is promising. Further work is indicated which would consider additional remote sensing input such as precipitation from the Tropical Rainfall Monitoring Mission and aerosols from MODIS. More advanced analysis that can take into account temporal lags and spatial effects would be useful as well.