Are NASA Land Information System (LIS) Data Useful for Predicting Dust Storms?

Convectively driven dust storms, or haboobs, are complex meteorological phenomena that are difficult to predict. They require strong, organized outflows that sufficiently outrun the precipitation core to loft dust while sustaining repeated convection in their wake. The goal of this research is to develop a machine learning (ML) model to predict dry thunderstorms associated with haboobs. An analysis of atmospheric profiles related to dust-vs-wind storm events to identify atmospheric ML predictors was previously completed. This investigation focuses on surface ML predictors by examining data from the NASA Land Information System (LIS) processed by the US Air Force Weather Agency 557th Weather Wing. The Fleet Numerical Meteorology & Oceanography Center (FNMOC) uses LIS data in real-time to initialize the Noah Land Surface Model which predicts land surface parameters for the COAMPS atmospheric forecasts. This data was used to conduct land surface analyses and to check for correlations between ground cover, soil moisture, and dust-vs-wind storms. After initial examination of annual LIS data, only summer (May-September) data was used for analysis. Some of the steps taken to ensure data quality included removing outliers that were 2 standard deviations outside the mean. Upper and lower bounds were set on each LIS parameter. Time series indicated two nearly discontinuous transitions in many parameters indicating likely changes in the model configuration. These transitions were mitigated by separating the data into three periods and normalizing each period separately. The data was normalized by subtracting the mean from the data for each grid point for each date over the corresponding time periods. Min-max scaling was performed on the normalized data for each grid point and each date over the time period. Then, using the normalized data, all LIS data points within three hours and 50 km of each storm event were found. The average of the data points within range for each parameter were then taken. Upon inspection, it was observed that the moisture parameters soilmo, soilh2, and resmoi were all lower, or less moist, for dust storms as compared to wind storms. Other parameters, such as soilty and vegtyp, were not as useful. This investigation is part of an effort to develop a machine learning model to predict dry thunderstorms associated with haboobs. This investigation shows that NASA LIS variables will be strong predictors for the machine learning model.