UC Davis

This dissertation is a three-part interconnected geographic study of fine scale soil moisture using monitoring, modeling, and remote sensing in coastal northern California. Although the focus is on hydrology and soil moisture, soil moisture naturally lends itself to be a unifying theme in physical geography connecting hydrology, agriculture, ecology, climate, wildfire, remote sensing, and humans through management actions. The objective of the three chapters is to advance the characterization and process understanding of soil-plant-water dynamics spatially and temporally and to provide high resolution soil moisture data that can be used by decision makers to improve situational awareness and planning through streamflow forecasting, landslide detection, drought warning and fire danger.In Chapter 1, monitoring data was collected in the context of a paired watershed study and soil and tree moisture probes were used to examine soil water-tree-ecosystem dynamics in Sonoma County, California. The results of this study showed: 1) there were noticeable effects on soil moisture and evapotranspiration from mechanical tree thinning and prescribed burns compared to an untreated watershed, 2) soil moisture decreased at the surface and increased in the root zone post-treatment, relative to an untreated watershed, 3) evapotranspiration decreased by 90 mm in the post treatment year for the treated watershed, 4) soil and tree moisture in situ sensors can be used as a surrogate for discrete and time consuming Live Fuel Moisture (LFM) measurements, and 5) soil moisture thresholds can be used to help identify wildfire danger and drought conditions. Improving real-time wildfire danger and drought conditions using existing or low-cost in situ sensors can improve situational awareness and planning. In Chapter 2, a novel L-band Unoccupied Aerial System (UAS) was used to map surface (0-10 cm depth) soil moisture at a high resolution in a complex, natural landscape. L-band UAS was used to map soil moisture at 5-50 meters resolution in a 3,200-acre mixed grassland-forested landscape in Sonoma County, California. The L-band dielectric constant model produced soil moisture maps with an unbiased root mean squared error (ubRMSE) of 0.07 m3/m3. A random forest machine learning model was also applied and produced surface soil moisture maps at 10-m and 30-m using optical UAS and thermal infrared (TIR) satellite data with accuracy levels of 0.020 m3/m3 and 0.017 m3/m3 ubRMSE, respectively. Improved fine-scale soil moisture maps developed using UAS-based systems may be used to help reduce fire risk and improve hydrologic models, streamflow forecasting, and early detection of landslides. In Chapter 3, high resolution root zone soil moisture, live fuel moisture, and fire danger maps were developed using machine learning, water balance, and regression models that connected the results from Chapter 1 and Chapter 2. Root zone soil moisture maps were developed using a random forest model that resulted in R2 values from 0.71 to 0.80. LFM maps were made using relationships developed in Chapter 1 and modeled soil moisture anomalies with an R2 of 0.82 compared against LFM data. The soil moisture anomaly and LFM results were applied to produce retrospective maps of fire danger that indicated predominantly medium to high fire danger during two historical fire ignition dates. An accumulated soil moisture anomaly (aSMA) showed interannual and long-term patterns and characterized an above-normal aSMA that switched to a below-normal aSMA right before the large wildfires that occurred between 2015-2020. Chapter 1 showed that root zone soil moisture was a better predictor of LFM than surface soil moisture and in Chapter 3, root zone soil moisture anomalies were shown to be an accurate predictor for LFM maps that can be useful to assess wildfire danger conditions at a high temporal and spatial scale. High resolution fire danger and LFM maps can be used to improve awareness of landscape conditions and early drought warning for decision makers.