|
Titel |
Different combination of MODIS land surface temperature data for daily air surface temperature estimation in North West Vietnam |
VerfasserIn |
Thanh Noi Phan, Martin Kappas, Jan Degener |
Konferenz |
EGU General Assembly 2017
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 19 (2017) |
Datensatznummer |
250141678
|
Publikation (Nr.) |
EGU/EGU2017-5213.pdf |
|
|
|
Zusammenfassung |
Land air temperature (Ta) with high spatial and temporal resolution plays an important role
in various applications, such as: crop growth monitoring and simulations, environmental risk
models, weather forecasting, land use cover change, urban heat islands, etc. Daily Ta
(including Ta−max, Ta−min, and Ta−mean) is usually measured by weather stations (often at
2 m above the ground); thus, Ta is limited in spatial coverage. Satellite data, especially
MODIS land surface temperature (LST) data at 1 kilometre and high temporal resolution (4
times per day, combining TERRA and AQUA) are free available and easily to access.
However, there is a difference between Ta and LST because of the complex surface energy
budget and multiple related variables between them. Several researches states that the Ta
could be estimated using MODIS LST data with accurate of 2-4oC. However, there are only a
handful of studies using dynamically combining of four MODIS LST data for Ta estimation.
In this study, we evaluated all 15 – possible – combinations of four MODIS LST
using support vector machine (SVM) and random forests (RFs) models. MODIS
LST and Ta data was extracted from 4 weather stations in rural area in North West
Vietnam from 2010 to 2012 (three years). Our results indicated that the accuracy of
Ta estimation was affected by the different combination and the combined data
(multiple variables) gave better results than those of single LST (solely variable),
the best result was achieved (coefficient of determination (R2) = 0.95, 0.97, 0.97;
root mean square error (RMSE) =1.7, 1.4, 1.2 oC for Ta−min, Ta−max, Ta−mean
respectively) when all four LSTs were combined and RFs performed better than SVM. |
|
|
|
|
|