NetCDF file: Reconstruction of TCA-based Time-variant Errors using LGBM and DNN Models

[Paper title] A Stand-Alone Framework for Predicting Spatiotemporal Errors in Satellite-Based Soil Moisture Using Tree-Based Models and Deep Neural Networks

Uncertainty Estimation via Triple Collocation Analysis:The dataset includes uncertainties derived from triple collocation analysis (TCA) of SMAP soil moisture (SM) measurements.

Machine Learning and Deep Learning Reconstructions:Masked uncertainty information due to four masking flags are reconstructed using ML/DL models:

Light Gradient Boosting Machine (LGBM)

Deep Neural Networks (DNN)

Variables:

latitude / longitude

Equal-Area Scalable Earth version 2.0 (EASE-2) Grid system (36 km)

flags_corr

[Flag1] correlation flag

flags_n_valid

[Flag2] valid number flag

flags_fMSE

[Flag3] fMSE flag

flags_negative_vars_err 

[Flag4] negative error variance flag 

SMAP_SM

SMAP soil moisture

SMAP_TCA_err

SMAP soil moisture TCA error (before reconstruction)

SMAP_TCA_err_recons_LGBM

SMAP soil moisture TCA error (after reconstruction by LGBM)

SMAP_TCA_err_recons_DNN

SMAP soil moisture TCA error (after reconstruction by DNN)

Note: The current dataset includes only the 2022 data, although the full data period spans from 2015 to 2023. Please be aware that if the data is updated, the Zenodo link will change. In that case, please visit https://www.hydroai.net or contact the data manager directly ([e-mail]) to download the dataset.