UC Berkeley

The use of wireless sensor networks(WSN) to study the mountain hydrologic cycle has been proposed as a supplement to existing systems used for ground-based hydrologic and ecological monitoring. Historically, the study of mountain hydrology and the water cycle has been largely observational, with meteorological forcing and hydrological variables extrapolated from a few infrequent manual measurements. Recent developments in Internet of Things (IoT) technology are revolutionizing the field of mountain hydrology. Low-power WSNs can now generate denser data in real-time and for a fraction of the cost of labor-intensive manual measurement campaigns. This research details the requirements and different technical options, describes the technology deployed in the American River basin, and discusses the methods associated with modeling large-scale environmental monitoring in extreme conditions. The American

River Hydrologic Observatory (ARHO) project has deployed fourteen low-power wireless IoT networks throughout the American River basin to monitor California’s snowpack. A network of sensors for spatially representative water-balance measurements was developed and deployed across the 2154 km2 snow-dominated portion of the upper American River basin, primarily to measure changes in snowdepth and soil-water storage, air temperature, and humidity. The WSNs, each has 10 measurement nodes that were strategically placed within a 1-km2 area, across different elevations, aspects, slopes and canopy covers.

The research evaluates the accuracy of a machine-learning based path loss model to estimate the expected radio transmission distances. The model is trained on 42,157,324 RSSI samples collected over seven months from the ARHO WSN. The 2218 links in the network span across the upper portion of the American River basin and are deployed in a complex environment, with large variations of terrain attributes and vegetation coverage. The model is based on an ensemble regression tree machine learning algorithm (Random Forest). The independent variables used in the model include path distance, canopy coverage, terrain variability, and path angle. The accuracy of this model is compared to several well-known canonical and empirical propagation models. This model showed a 37% reduction in the average prediction error compared to the canonical/empirical model with the best performance. The research presents an in-depth discussion on the strengths and limitations of the proposed approach as well as opportunities for further research.

Compared to existing operational sensor installations, the ARHO WSN reduces hydrologic uncertainty in at least three ways. First, redundant measurements improved estimation of

lapse rates for air and dew-point temperature. Second, distributed measurements captured local variability and constrained uncertainty in the air and dew-point temperature, snow accumulation and derived hydrologic attributes important for modeling and prediction. Third, the distributed relative-humidity measurements offer a unique capability to monitor upper- basin patterns in dew-point temperature and better characterize elevation gradient of water vapor-pressure deficit. Network statistics during the first year of operation demonstrated that the ARHO WSN was robust for cold, wet and windy conditions in the basin.

Using daily dew-point temperature and the amount of snow accumulation at each node to estimate the fraction of rain versus snow resulted in an underestimate of total precipitation be- low the 0 o C dew-point elevation, which averaged 1730 m across 10 precipitation events. Blending lower-elevation rain-gauge data with higher-elevation sensor-node data for each event pro- vided precipitation estimates that were on average 15-30% higher than using either set of measurements alone. Using data from the current operational snow-pillow sites give even lower estimates of basin-wide precipitation. Given the increasing importance of liquid precipitation in a

warming climate, a strategy that blends distributed measurements of both liquid and solid precipitation will provide more-accurate basin-wide precipitation estimates. The distributed, representative sensor-network measurements also improve upon operational estimates of snow- pack water storage, snowmelt amount and snowmelt timing across the basin.