Integrating SMAP and CYGNSS data for daily soil moisture and agricultural drought monitoring in Nghe An province, Vietnam

Abstract

In the context of climate change, droughts are increasingly frequent and severe, affecting broader regions. Consequently, effective drought monitoring is crucial for risk management and understanding climate change impacts. Soil moisture estimation using satellite data is a pivotal metric for developing time-series agricultural drought monitoring maps. This study proposes a methodology for constructing soil moisture and agricultural drought maps for Nghe An Province, Vietnam, using the SMAP dataset along with soil moisture estimations from CYGNSS data and additional ancillary data. The Self-Attention-based Imputation for Time Series (SAITS) model, employing self-attention mechanisms to impute missing values in multivariate time series, is used to construct the soil moisture dataset from SMAP, resulting in complete datasets with a training loss RMSE_SAITS $=$ 0.073 ${\mathrm{cm}}^3/{\mathrm{cm}}^3$. Additionally, leveraging a Random Forest Regression model, CYGNSS data combined with meteorological, topographic, and soil texture information enable the estimation of daily soil moisture values, exhibiting a strong correlation with R $=$ 0.889. Subsequently, integration of the two soil moisture products from SMAP and CYGNSS yields a dataset with a spatial resolution of 1km and a temporal resolution of 1 day. The soil moisture results were compared with moisture data from ERA5 (R $=$ 0.75, ubRMSE $=$ 0.055 ${\mathrm{cm}}^3/{\mathrm{cm}}^3$) and in-situ data in Nghe An province (R $=$ 0.709, ubRMSE $=$ 0.017 ${\mathrm{cm}}^3/{\mathrm{cm}}^3$). Finally, the Standardized Soil Moisture Index is calculated to transform the time-series soil moisture data into a standardized normal distribution, generating agricultural drought maps with 9 different levels. This study represents a significant advancement in agricultural drought monitoring, highlighting the immense potential of machine learning techniques when combined with satellite-based soil moisture data. Our approach effectively monitors drought in Nghe An Province, Vietnam, with broader applicability to other regions worldwide.