Estimating surface soil moisture from SMAP observations using a neural network technique

Authors: KOLASSA, J. - Goddard Space Flight Center; REICHLE, R. - Goddard Space Flight Center; LUI, Q. - Goddard Space Flight Center; ALEMOHAMMAD, S.H. - Columbia University; GENTINE, P. - Columbia University; AIDA, K. - University Of Tsukuba; ASANUMA, J. - University Of Tsukuba; BIRCHER, S. - Center For The Study Of The Biosphère From Space(CESBIO); CALDWELL, T. - University Of Texas; COLLIANDER, A. - Jet Propulsion Laboratory; Cosh, Michael; Holifield Collins, Chandra; Jackson, Thomas; JENSEN, K.H. - Copenhagen University; MARTINEZ-FERNANDEZ, J. - University Of Salamanca; MCNAIRN, H. - Agriculture And Agri-Food Canada; PACHECO, A. - Agriculture And Agri-Food Canada; THIBEAULT, M. - University Of Buenos Aires; WALKER, J. - Monash University

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/1/2018
Publication Date: 1/31/2018
Citation: Kolassa, J., Reichle, R., Lui, Q., Alemohammad, S., Gentine, P., Aida, K., Asanuma, J., Bircher, S., Caldwell, T., Colliander, A., Cosh, M.H., Holifield Collins, C.D., Jackson, T.J., Jensen, K., Martinez-Fernandez, J., Mcnairn, H., Pacheco, A., Thibeault, M., Walker, J. 2018. Estimating surface soil moisture from SMAP observations using a neural network technique. Remote Sensing of Environment. 204:43-59. https://doi.org/10.1016/j.rse.2017.10.045.
DOI: https://doi.org/10.1016/j.rse.2017.10.045

Interpretive Summary: Neural network algorithms can aid in the modeling of soil moisture, but require substantial datasets with which to train. The Soil Moisture Active Passive (SMAP) satellite mission provides such data, and along with in situ resources, these data were assimilated into an existing NASA water cycle model. Improvement in the model’s skill was evaluated and determined to be significant for understanding areas where the model and the satellite product differ, pointing to areas of future research. This study is significant for hydrologic modelers and climate and weather forecasters who are encouraged to include soil moisture products into existing operational systems.

Technical Abstract: A Neural Network (NN) algorithm was developed to estimate global surface soil moisture for April 2015 to June 2016 with a 2-3 day repeat frequency using passive microwave observations from the Soil Moisture Active Passive (SMAP) satellite, surface soil temperatures from the NASA Goddard Earth Observing System Model version 5 (GEOS-5) land modeling system, and Moderate Resolution Imaging Spectroradiometer-based vegetation water content. The NN was trained on GEOS-5 soil moisture target data, thereby making the NN estimates consistent with the GEOS-5 climatology such that they may ultimately be assimilated into this model without further bias correction. Moreover, the non-local calibration of the NN algorithm reveals areas where the SMAP observations are inconsistent with the model estimates and thus where the assimilation of the NN retrieval product has the greatest potential to correct the model. The average unbiased RMSE, correlation and anomaly correlation of the NN retrievals were 0.038 m3/m3, 0.72, and 0.64, respectively, against SMAP core validation site in situ measurements and 0.075 m3/m3, 0.58, and 0.48, respectively, against International Soil Moisture Network (ISMN) measurements. At the core sites, the NN retrievals have a significantly higher skill than the GEOS-5 model estimates and a slightly lower correlation skill than the SMAP Level-2 Passive (L2P) product. Against ISMN measurements, the model estimates and the retrieval products had a more comparable skill. A triple collocation analysis against AMSR2 and ASCAT soil moisture retrievals showed that the NN and L2P retrieval errors have a similar spatial distribution, which differs slightly from that of the model.