Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Pays Bas → 2020 → Spatial downscaling of satellite soil moisture utilising high-resolution UAS data over Alento catchment in Italy.

University of Twente - International Institute for Geo-Information Science and Earth Observation (ITC) 2020

Spatial downscaling of satellite soil moisture utilising high-resolution UAS data over Alento catchment in Italy.

Byukusenge, Elie

Titre : Spatial downscaling of satellite soil moisture utilising high-resolution UAS data over Alento catchment in Italy.

Auteur : Byukusenge, Elie

Etablissement de soutenance : University of Twente - International Institute for Geo-Information Science and Earth Observation (ITC)

Grade : Master of Science in Geo-Information Science and Earth Observation 2020

Résumé
Soil moisture plays a vital role in water resources management related applications. Nevertheless, the coarse resolution of satellite-based surface soil moisture products has limited applications at field-scale, for example, precision agriculture. The current Sentinel-1 satellite mission provides soil moisture products at 1km resolution, which is still not matching the need at field scales. Therefore, the spatial downscaling approach was applied to downscale coarse resolution (1km) satellite surface soil moisture (SSM) products to high resolution (15 cm) utilising UAS measurements using the random forest (RF) machine learning-based model. In this study, the RF model was trained using various configurations of input data prepared with remotely sensed SSM and ancillary land surface parameters of LST, DEM and NDVI. The performance of different trained RF models was evaluated to find out which RF model could represent the best relationship of SSM and surface parameters with the best capability for the prediction of SSM. The results indicated that all trained RF models have good performances. However, the trained RF model using 2018 - 2019 dataset on 78 km by 78km spatial extent outperformed the others with the highest correlation coefficient (R ) of 0.83 and RMSE of 12.13 %. Therefore, this trained RF model was considered for further process and was applied with the land surface features derived from UAS imageries to predict the SSM at 15cm resolution at noon and sunrise time. The trained RF model can also identify the relative importance of land surface parameters /features in predicting SSM. It was found that the LST has a higher impact than other features, while DEM being the least influential. The downscaled SSM can capture the spatial pattern of SSM at noon and sunrise time, when compared with the in situ measurements from the study area in Monte Cilento Sub catchment in Alento Catchment, Italy. The averaged ubRMSE, RMSE and R are reported 0.07 cm3/cm3, 0.21 cm3/cm3, and 0.60 respectively. Notably, all statistical metrics showed acceptable results even though the average of ubRMSE does not reach the SMAP and Global Climate Observing System (GCOS) mission accuracy target of 0.04 cm3/cm3 for soil moisture due to the downscaled SSM products were generated at 5 cm while the in situ measurements were taken at 15 cm. In summary, this study successfully generates high spatial resolution SSM data from coarse-scale satellite products by integrating UAS measurements and RF model as a downscaling approach. The generated soil moisture products could provide useful information for better agricultural management in Monteforte Cilento sub-catchment in Alento River catchment.

Présentation

Page publiée le 20 avril 2021