Multi-layer root zone soil moisture estimation for drought management | |
| Author | Kodikara, Jayanga Nishadi Samararathna |
| Call Number | AIT Thesis no.WM-24-22 |
| Subject(s) | Soil moisture--Data processing Soil moisture--Measurement Droughts--Management |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Soil moisture is a key controlling factor in eco-hydrological processes, governing interactions between soil-plant-atmosphere systems exchanging carbon, energy, and water fluxes. Quantification of root zone soil moisture (RZSM) layer-wise is a challenge in sustainable irrigation management because of the complexity of its variation in space, depth, and time. This study develops a model using the 1D Convolutional Neural Network (CNN) algorithm. The correlation between the soil characteristics (static variables) and meteorological variables (dynamic variables) with soil moisture (SM) at various depths (10 cm, 25 cm, 50 cm, and 100 cm) was examined to identify the input variables for the model. The developed model is applied to fill the gaps in observed SM time-series across different depths and locations in Nebraska. Observed precedence soil moisture at the topmost layer (0-10 cm) along with meteorological and soil properties of Nebraska are used as input variables. In-situ SM observations for each depth (0-10 cm, 10-25 cm, 25-50 cm, and 50-100 cm) are used as predictands. Nebraska is clustered into three regions considering the association of correlated static and dynamic variables and RZSM using silhouette scores and the k means algorithm. Then the 1D-CNN models are trained cluster-wise and layer-wise to estimate RZSM. The trained models were then used to predict the SM values to fill the gaps in the observed SM time series. Correlation analysis results show a promising correlation between SM and predictor variables (available water capacity (AWC), organic matter (OM), soil texture, elevation, moist bulk density (MBD), saturated hydraulic conductivity (SHC), rainfall, maximum and minimum relative humidity, land surface temperature, solar radiation, vapor pressure deficit, and maximum and minimum temperature. However, the strength of the correlation declines with increasing depth. Statistically significant correlations were confirmed by the calculated p-values. The inclusion of soil moisture data from the previous three days at a depth of 10 cm significantly improved the model prediction accuracy when predicting soil moisture at 10 cm. However, the model prediction accuracy for test data declined with increasing depths. Validation at nine selected soil moisture stations indicated that the multi-layer soil moisture estimates are predicted with acceptable accuracy, although the prediction accuracy declined with respect to the increasing depth. The RZSM gridded maps of daily 1 km resolution were generated using bias-corrected SMAP surface soil moisture as the three days previous soil moisture input to the developed model. Due to the lesser representativeness of SMAP surface soil moisture (SSM) to the observed soil moisture at 10 cm, the gridded maps performance was reduced in comparison to the predicted point data performance at the same soil moisture station using the same model. The calculated soil water deficit (SWD) across Nebraska showed that Nebraska has undergone mild agricultural drought conditions from 2015 to 2018. |
| Year | 2024 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Shanmugam, Mohana Sundaram |
| Examination Committee(s) | Natthachet Tangdamrongsub;Sarawut Ninsawat;Shrestha, Sangam |
| Scholarship Donor(s) | Pan Merit Belt and Road Scholarships |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2024 |