A Machine Learning Approach to Estimate Daily Air Temperature Globally
Abstract
Air temperature (Ta) is a key factor in energy balance near the land surface and plays an important role in driving many environmental processes. There have been extensive efforts focused on estimating air temperature (Ta), particularly daily minimum and maximum Ta. One globally available source for estimating air temperature is MODerate resolution Imaging Spectroradiometer (MODIS), which has provided daily global 1km land surface temperature (LST) since the year 2000. The challenge is how best to estimate min and max Ta from land surface temperature. In this study, a machine learning algorithm is used to estimate Ta from LST. The algorithm is trained on 18 years of daily min and max Ta measurements (responses) collected at more than 3300 meteorological stations distributed globally. The model uses a suite of predictors: day and night LST observations from both Aqua and Terra satellites, latitude, elevation, day of year, Solar Zenith Angle, and several land cover/vegetation indices to predict the response, which is either daily min or max Ta. Results show that MODIS LST can be used to estimate daily Ta statistics using machine learning reasonably well. Machine learning methods are also compared to traditional linear and non-linear regression approaches. To illustrate the applicability of the daily min or max Ta estimates, a case study using an existing evapotranspiration method is presented.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.H34B..07R
- Keywords:
-
- 0434 Data sets;
- BIOGEOSCIENCESDE: 1855 Remote sensing;
- HYDROLOGYDE: 1926 Geospatial;
- INFORMATICSDE: 1942 Machine learning;
- INFORMATICS