Hydrological droughts of 2017-2018 explained by the Bayesian reconstruction of GRACE(-FO) fields

Shaoxing Mo, Yulong Zhong, Ehsan Forootan, Xiaoqing Shi, Wei Feng, Xin Yin, Jichun Wu

Research output: Contribution to journalJournal articleResearchpeer-review

1 Citation (Scopus)
17 Downloads (Pure)


Hydrological droughts are events of prolonged water scarcity and cause many devastating impacts. It is, therefore, extremely crucial to understand their spatiotemporal evolution to guide prevention and mitigation policies. The Gravity Recovery and Climate Experiment (GRACE, April 2002–June 2017) and GRACE Follow-On (GRACE-FO, June 2018 until the present) missions have been used to study large-scale droughts of almost two decades. But characterizing droughts during the between missions gap period of 2017 and 2018 has not been well addressed and will be covered here. To bridge the gap, an innovative Bayesian convolutional neural network is developed to reconstruct the missing signals from hydroclimatic inputs. The reconstruction fields and existing signals are then used to explore regions that have experienced consecutive water storage deficits during the 2017–2018 gap. We found many regions of the northern midlatitudes exhibiting moderate to exceptional droughts in terms of water storage deficits, among which parts of Pakistan and Afghanistan, and Iberian Peninsula experienced exceptional droughts lasting for more than 1 year with the maximum deficits (−4.4 ± 0.8 and −7.2 ± 1.1 cm, respectively) being over 50% of the seasonal storage variations. Comparisons with climate indicators show that the identified droughts are predominantly caused by continuous below-normal precipitation. The recovery process correlates generally well with the accumulation rate of precipitation surpluses (the correlation coefficient (R) can be up to 0.92). Besides, the reconstructed signals, which have R > 0.7 with the testing GRACE(-FO) data in over 90% of the globe, reliably maintain the data continuity and therefore they are recommended for hydro-climatological studies.

Original languageEnglish
Article numbere2022WR031997
JournalWater Resources Research
Issue number9
Number of pages39
Publication statusPublished - Sept 2022


  • convolutional neural network
  • deep learning
  • gap filling
  • hydrological drought


Dive into the research topics of 'Hydrological droughts of 2017-2018 explained by the Bayesian reconstruction of GRACE(-FO) fields'. Together they form a unique fingerprint.

Cite this