A carregar...
Publicação
Soil moisture estimation of eucalyptus forests in Portugal with L-band SAR using polarimetric - Decompositions and machine learning
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 10.07 MB | Adobe PDF |
Autores
Resumo(s)
Soil moisture is a critical ecological parameter because it is a primary input for all processes
that involve the complex interaction between land surface and the atmosphere. Remote
sensing, especially using microwaves, has shown great promise in measuring soil moisturewith
several operating satellites focused on its continuous estimation and monitoring on a
global scale. Portugal is predominantly characterized by Mediterranean and semi-arid
climates that feature low and sporadic precipitation. Over 10% of Portugal’s land area has
been planted with Eucalyptus globulus- a non-native, fast-growing tree primarily planted for
industrial use. Some studies have demonstrated that eucalyptus plantations adversely affect
water availability, but overall results have been inconclusive as there are numerous other
confounding variables. The goals of this study were to determine, using fully polarimetric
L-band SAR and machine learning, if soil moisture could be accurately predicted in
eucalyptus forests, and if there is a significant difference in soil moisture inside eucalyptus
forests relative to other forests. Vegetated surfaces complicate the estimation of soil moisture
because their structure and water content contribute significantly to backscatter of the radar
signal. Thus, four polarimetric decompositions were compared to separate vegetative versus
surface backscatter. The inputs from those decompositions, as well as several additional radar
indices and polarizations from the microwave images, were used as feature inputs into two
different machine learning models. After a feature selection process, the soil moisture
estimations were retrieved and compared using cross-validation. The best overall soil
moisture retrieval for Eucalyptus forests came from Random Forest with a RMSE of 0.021, a
MAE of 0.017, and a MBE of 0.001. Through a statistical t-test, predicted soil moisture
values in eucalyptus forests did not differ significantly as compared to other forest types in the
study area.
Descrição
Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies
Palavras-chave
Soil Moisture Polarimetry Eucalyptus L-Band SAR Feature Selection Machine Learning