Repository logo

Browsing by Author "Chavez, Jose Luis, committee member"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemEmbargo
    Enhancing rootzone soil moisture estimation using remote sensing, regional characteristics, and machine learning
    (Colorado State University. Libraries, 2023) Sahaar, Ahmad Shukran, author; Niemann, Jeffrey D., advisor; Chavez, Jose Luis, committee member; Green, Timothy R., committee member; Butters, Gregory, committee member
    Accurate estimation of root-zone soil moisture (θ ̄) is essential for various agricultural applications, including crop yield estimation, precision irrigation, and groundwater management. This dissertation encompasses three interconnected studies that collectively investigate different approaches for improving soil moisture estimation. The first study delves into the utilization of remote sensing methods, particularly optical and thermal satellite imagery, to estimate fine-resolution (30 m) root-zone soil moisture across diverse regions. Traditionally, these methods relied on empirical relationships with evaporative fraction Λ_SEB or evaporative index Λ_PET. However, it has been shown that a single relationship does not universally apply to all regions. This study evaluates the influence of regional soil, vegetation, and climatic conditions on the shape and strength of these relationships using global sensitivity analysis. The results highlight that soil characteristics, such as clay and silt content, and vegetation properties, like leaf area index and rooting depth, play pivotal roles in determining these relationships. Moreover, the impact of annual precipitation in defining climatic regions is crucial. Consequently, region-specific relationships are proposed, adapting to local conditions and potentially enhancing soil moisture estimates. The second study extends this investigation by applying the regionally adapted relationships for the Λ_SEB " vs." θ ̄ and Λ_PET " vs." θ ̄ to estimate rootzone soil moisture (θ ̄) from remote sensing data across four study regions. The results consistently demonstrate the superior performance of the regionally adapted relationships over a single empirical relationship, with a substantial reduction in root mean squared error. These adapted relationships are particularly effective in arid and semiarid regions. The third study explores the application of machine learning models, including XGBoost, CatBoost, RF, LightGBM, and artificial neural networks, to predict soil moisture levels across various climates and depths in the contiguous United States. The findings emphasize the high accuracy and effectiveness of machine learning models, especially XGBoost, in predicting soil moisture across diverse climate regions. XGBoost outperforms other models, making it a potentially valuable tool for soil moisture prediction in environmental monitoring and management. The study also highlights the influence of climate and soil depth on prediction accuracy, with deeper layers having improved forecasts. Additionally, feature importance analysis identifies key predictors for predicting soil moisture, such as elevation, aridity index, soil composition, and depth. These findings contribute to the advancement of soil moisture monitoring and management, with practical applications in agriculture and environmental sciences.