An Algorithm for Estimating Heat Index, Based on GOES and MODIS Satellite Data

A real time algorithm is proposed to estimate hourly heat index over Mesoamerica and Caribbean (MAC) countries. The algorithm was developed with Geostationary Operational Environmental Satellite (GOES)-13 and the Moderate-Resolution Imaging Spectroradiometer (MODIS) data. The methodology was designed to operate with the newer GOES-16 satellite data. The heat index is the combination of air temperature and relative humidity and is an attempt to estimate what humans feel as an apparent temperature. The variation of heat index is linked to both human health and usually the extreme values of heat index are used for advising heat warning events.

The algorithm estimates first the relative humidity and the air temperature, and then estimates the heat index. The algorithm of relative humidity was developed in homogeneous climatic regions that were identified in the MAC countries by using a self-organizing features maps neural network. A regression model was used to estimate relative humidity. The introduced regression model exhibit four major components, a trend, a seasonal, a short-term variability, and a noise component. The trend is represented by a deterministic model that includes the time and the intrinsic surface characteristics such, soil texture, and elevation. The short-term variability was modeled by the water vapor channel and the albedo from the near infrared channel, and also by precipitable water, land surface temperature, and normalize differences of vegetation index. The seasonal behavior was developed by the relative position of the Earth with respect to the Sun and includes the monthly and the daily variations for each grid. The seasonal component was modelled by a set of binary variables and sinusoidal functions, whose amplitudes and phases were varying according to the sunrise and sunset observations. The noise component was modeled by using a stochastic autoregressive model. The proposed retrieve algorithm was validated with GOES-16 data during the first week of February 2018. The validation data includes 13,383 measurements from stations that exhibited more than 80% of data. It was concluded that the average of mean absolute error was 7.39 % and root mean squared error was 9.32 %. The air temperature model was developed in a similar form to the relative humidity model. The air temperature also exhibits four major components, a trend, a seasonal, a short-term variation and a noise stochastic component. The trend is represented by a deterministic model and exhibits the intrinsic surface properties such orography, soil, and vegetation. The seasonal behavior was driven by the relative position of the Earth with respect to Sun and has two periodic components associated to daily and annual variations. The short term variation was modeled by using the land surface temperature, the brightness temperature from the water vapor, and the thermal infrared channels. The noise stochastic behavior was represented by an autoregressive model, and the estimation of regression with stochastic model was developed by using the generalized linear models.

Steadman (1979) derived a method to estimate the heat index based on the amount of heat lost via exhaling and the skin’s resistance to heat and moisture transfer. Rothfusz (1990) used Steadman’s results to develop a regression equation, which expresses the relationship between temperature at different relative humidity and the skin's resistance to heat and moisture transfer. The United States National Weather Service (NWS) developed its own algorithm for estimating heat index based on Steadman's and Rothfusz's work, and we adopted this method, which is available online (NOAA/NWS, 2016) with a detailed description of the algorithm given by Anderson et al. (2013).