Location: Hydrology and Remote Sensing LaboratoryTitle: Spatial–temporal modeling of root zone soil moisture dynamics using random forest and remote sensing
|KISEKKA, I. - University Of California|
|RAC PEDDINTI, S. - University Of California|
|Kustas, William - Bill|
|BAMBACH, N. - University Of California|
|BASTIAANSSEN, W.G.M - Delft University|
Submitted to: Irrigation Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/18/2022
Publication Date: 2/6/2022
Citation: Kisekka, I., Rac Peddinti, S., Kustas, W.P., McElrone, A.J., Bambach, N., McKee, L.G., Bastiaanssen, W. 2022. Spatial–temporal modeling of root zone soil moisture dynamics using random forest and remote sensing . Irrigation Science. https://doi.org/10.1007/s00271-022-00775-1.
Interpretive Summary: The increased frequency of extreme drought and increased regulation of groundwater use are making irrigated agriculture more challenging in California. Information on soil moisture is important for the development of appropriate irrigation systems to achieve the benefits of precision agriculture and agricultural sustainability. This study compared two remote sensing-based methods and a machine learning data driven method for predicting spatial-temporal root zone soil moisture distribution in a vineyard. When sufficient observed data for training are available, data-driven models based on machine learning provided much more accurate estimates of root zone soil moisture than the remote sensing-based methods. However, with refinements to the remote sensing algorithms the remote sensing-based methods may potentially be enhanced to produce accurate root zone soil moisture at multiple spatial resolutions from plot and field to landscape and watershed scales. Therefore, combining remote sensing with machine-learning techniques has the potential to provide precision agricultural water management applications for achieving agricultural sustainability in water-limited regions.
Technical Abstract: Spatial–temporal root zone soil moisture (RZSM) information collected at different scales is useful for a variety of agricultural, hydrologic, and climate applications. Spatial RZSM data can be retrieved from satellites or inferred from empirical equations and process-based model simulations. Machine learning applications for evaluating RZSM across numerous spatial-temporal scales are less generalizable than process-based models. However, data-driven machine learning approaches offer a unique opportunity to develop complex models of soil moisture without making assumptions about the processes involved in the studied region. In this study, comparisons were made between two empirical models, pySEBAL and EFSOIL, which were based on evaporation fraction (EF) and soil properties, and a data-driven model based on the Random Forest (RF) ensemble algorithm. These approaches were evaluated to demonstrate their capabilities for RZSM estimation. The EF obtained from Landsat was used as the major input for all three models, along with other meteorological and soil physical properties. The RF model was trained using in-situ soil moisture data from TDR sensors installed in a vineyard from 2018 to 2020. The predictor variables comprised of meteorological, soil, and evaporation fractions. The results reveal that there is a strong correlation between the in-situ measured soil moisture and the RF predicted soil moisture at all sensor locations. The empirical models pySEBAL and EFSOIL failed to accurately predict RZSM values from all monitored locations. However, the high RZSM generated by pySEBAL demonstrated the presence of field variability within the vineyard. We also demonstrated that the RF model may be used to forecast RZSM with fewer monitoring locations when ground measurements are combined with machine learning to produce high spatially detailed soil moisture maps using Landsat satellite data.