loading page

Correction of SMAP (Soil Moisture Active Passive) Satellite Retrieved Soil Moisture Data using Machine Learning Techniques over North West Region of Bangladesh
  • Hamidul Haque,
  • Nasreen Jahan
Hamidul Haque
Bangladesh Univsersity of Engineering and Technology

Corresponding Author:[email protected]

Author Profile
Nasreen Jahan
Bangladesh University of Engineering and Technology
Author Profile


Bangladesh, part of Indo-Gangetic-Brahmaputra Plain, is frequently affected by floods and droughts. As the economy of Bangladesh is still agriculture based, effective measurement of soil moisture will not only strengthen the irrigation management but also improve the hydrological modelling and drought prediction. But only nine agro-meteorological stations of Bangladesh measure the soil moisture four times a month which creates a vacuum to scientifically manage her water resources. SMAP (Soil Moisture Active Passive) satellite of NASA provides an unprecedented opportunity for full scale measurement of soil moisture over this region. Field measurements of soil moisture from April 2015 were used to assess the effectiveness of the SMAP’s measurement over the North West Region of Bangladesh which suffers from frequent dry spells. Initially the Root mean squared error (RMSE) between the SMAP and observed soil moisture were found to vary between 12.28 to 16.72% for the available stations. The results showed a bias in SMAP data and it was significantly reduced using bias correction. Later multiple linear regression, based on supplementary climate data in addition to SMAP observations, was applied to obtain an improved estimate of soil moisture and the RMSE were reduced to 1.19 to 3.18%. Lastly, different machine learning techniques (i.e. ANN, SVR, XGBoost etc.) were used to reduce the bias further. This study demonstrates a promising potential of using the SMAP data in soil moisture estimation over Bangladesh for its effective water resources management.