Integrating Land Surface Temperature (LST) Images from the Moderate Resolution Imaging Spectroradiometer (MODIS) Sensor for Agricultural and Human Health Studies

Land surface temperature (LST) images from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor have been widely utilized across scientific disciplines for a variety of purposes. The goal of this dissertation was to utilize MODIS LST for three spatial modeling applications within the conterminous United States (CONUS). These topics broadly encompassed agriculture and human health. The first manuscript compared the performance of all methods previously used to interpolate missing values in 8-day MODIS LST images. At low cloud cover (\u3c30%), the Spline spatial method outperformed all of the temporal and spatiotemporal methods by a wide margin, with median absolute errors (MAEs) ranging from 0.2°C-0.6°C. However, the Weiss spatiotemporal method generally performed best at greater cloud cover, with MAEs ranging from 0.3°C-1.2°C. Considering the distribution of cloud contamination and difficulty of implementing Weiss, using Spline under all conditions for simplicity would be sufficient. The second manuscript compared the corn yield predictive capability across the US Corn Belt of a novel killing degree day metric (LST KDD), computed with daily MODIS LST, and a traditional air temperature-based metric (Tair KDD). LST KDD was capable of predicting annual corn yield with considerably less error than Tair KDD (R2 /RMSE of 0.65/15.3 Bu/Acre vs. 0.56/17.2 Bu/Acre). The superior performance can be attributed to LST’s ability to better reflect evaporative cooling and water stress. Moreover, these findings suggest that long-term yield projections based on Tair and precipitation alone will contain error, especially for years of extreme drought. Finally, the third manuscript assessed the extent to which daily maximum heat index (HI) across the CONUS can be estimated by MODIS multispectral imagery in conjunction with land cover, topographic, and locational factors. The derived model was capable of estimating HI in 2012 with an acceptable level of error (R 2 = 0.83, RMSE = 4.4°F). LST and water vapor (WV) were, by far, the most important variables for estimation. Expanding this analytical framework to a more extensive study area (both temporally and spatially) would further validate these findings. Moreover, identifying an appropriate interpolation and downscaling approach for daily MODIS imagery would substantially increase the utility of the corn yield and HI models