Exploring the Feasibility of Using Data Mining Approach for Estimation and Outlook of Evapotranspiration

Energy exchange associated with evapotranspiration (ET) between land surface and the atmosphere is among the largest energy sources for weather and climate formation. With the massive ET related data sets becoming available from Earth observing satellite in the past years, many approaches to estimating ET from the satellite sensors have been developed with various accuracy performances. ET observations are needed not only for NOAA numerical weather and climate prediction models, but also for the monitoring and outlook of hydrological events (e.g. agricultural droughts). In this study we attempt to explore the usage of the big data sets from satellite remote sensing and numerical model simulations, and a data mining algorithm for estimating daily ET. Regression trees for the data mining are obtained using near real time retrievals of land surface temperature, vegetation dynamics, solar insolation from satellites, relevant simulations or forecasts from numerical models, and in situ measurements from ground stations in order to estimate or outlook ET values. Preliminary results for the selected ground stations will be presented. How the data mining algorithm is compared with a physical approach based on Atmosphere-Land Exchange Inversion (ALEXI) model and application potential of the data mining approach will be discussed.