Thursday, 14 January 2016: 8:45 AM
Room 226/227 ( New Orleans Ernest N. Morial Convention Center)
In tropical areas the surface air temperature is affected by the interaction with atmospheric and surfaces processes including solar radiation, cloud cover, rainfall, relative humidity, wind, elevation, vegetation, soil texture, etc. Visible and infrared GOES data were used to up scale the hourly observations of air temperature measured at each ground station. Twenty one stations located in Puerto Rico with hourly data were used to derive a time series model. The air temperature exhibits three major components, a trend, a seasonal, and a stochastic component. The trend is a deterministic model that represents the long-term variations associated with climate change. The deterministic model also includes the local intrinsic physical properties such elevation, soil texture, and vegetation index. The seasonal behavior has two periodic patterns associated to daily and annual variations. Sinusoidal models were used to represent the daily and annual seasonal variations of air temperature. The stochastic behavior was represented by a transfer function model, which includes the impulse response function and an autoregressive moving average (ARMA) model. The impulse response function modulates the visible and infrared radiation effects at the top-clouds and at the surface level. The presence of clouds during the daytime produce an abrupt reduction of surface air temperature, and the intervention of clouds and rainfall were also modeled by a second impulse response function using infrared brightness temperature, the visible reflectance (0.65 ìm), and albedo from near infrared (3.9 ìm) data. During the nighttime only the brightness temperature from the water vapor channel (WV; 6.7 µm) and from the thermal infrared channels (10.7 µm and 12 µm) were used to correlate with air temperature. Brightness temperature differences were also added to explain best the air temperature variability. The autocorrelated component of the transfer function was represented by an ARMA (1,1). The nearest pixel in time and space were selected to match with ground stations and identify the time series model. The initial point for parameter estimation was obtained by using a model decomposition, and a nonlinear optimization technique was used to search for the best structures of the impulse respond functions, and parameter estimations for the entire model. GOES data were collected at every 15 minutes and resampling was applied to derive estimates at 1 km spatial resolution, and the hourly average was computed for each pixel. The introduced model was implemented for the Puerto Rico climate conditions. Two months (January and February 2013) of hourly data were used for training the model and one week for model validation. It was found the model explains 63 % of the total air temperature variability, and shows a standard error of 2.5 C.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner