We evaluate the accuracy of a machine-learning-based path loss model trained on 42,157,324 RSSI samples collected over one year from an environmental wireless sensor network using 2.4 GHz radios. The 2218 links in the network span a 2000 km² basin and are deployed in a complex environment, with large variations of terrain attributes and vegetation coverage. Four candidate machine-learning algorithms were evaluated in order to find the one with lowest error: Random Forest, Adaboost, Neural Networks, and K-Neareast-Neighbors. Of the candidate models, Random Forest showed the lowest error. The independent variables used in the model include path distance, canopy coverage, terrain variability, and path angle. We compare the accuracy of this model to several well-known canonical (Free Space, plane earth) and empirical propagation models (Weissberger, ITU-R, COST235). Unlike canonical models, machine-learning algorithms are not problem-specific: they rely on an extensive dataset and a flexible model architecture to make predictions. We show how this model achieves a 37% reduction in the average prediction error compared to the canonical/empirical model with the best performance. The article presents a in-depth discussion on the strengths and limitations of the proposed approach as well as opportunities for further research.

Monitoring the snow pack is crucial for many stakeholders, whether for hydro-power optimization, water management or flood control. Traditional forecasting relies on regression methods, which often results in snow melt runoff predictions of low accuracy in non-average years. Existing ground-based real-time measurement systems do not cover enough physiographic variability and are mostly installed at low elevations. We present the hardware and software design of a state-of-the-art distributed Wireless Sensor Network (WSN)-based autonomous measurement system with real-time remote data transmission that gathers data of snow depth, air temperature, air relative humidity, soil moisture, soil temperature, and solar radiation in physiographically representative locations. Elevation, aspect, slope and vegetation are used to select network locations, and distribute sensors throughout a given network location, since they govern snow pack variability at various scales. Three WSNs were installed in the Sierra Nevada of Northern California throughout the North Fork of the Feather River, upstream of the Oroville dam and multiple powerhouses along the river. The WSNs gathered hydrologic variables and network health statistics throughout the 2017 water year, one of northern Sierra’s wettest years on record. These networks leverage an ultra-low-power wireless technology to interconnect their components and offer recovery features, resilience to data loss due to weather and wildlife disturbances and real-time topological visualizations of the network health. Data show considerable spatial variability of snow depth, even within a 1 km ${}^2$ network location. Combined with existing systems, these WSNs can better detect precipitation timing and phase in, monitor sub-daily dynamics of infiltration and surface runoff during precipitation or snow melt, and inform hydro power managers about actual ablation and end-of-season date across the landscape.

Core Ideas: An algorithm for predicting soil‐water storage was evaluated for 6 yr. Algorithm accuracy was high during wet‐up, snow cover, and early recession. Accuracy declined during late recession and during dry periods. Accuracy declined and relative independent‐variable rank changed in dry years. Transient snow cover in dry periods altered soil‐water storage patterns. Using 6 yr (Water Year [WY] 2009–WY 2014) of hourly in situ measurements from a spatially distributed water‐balance cluster, we quantified the long‐term accuracy of an algorithm used to predict spatial patterns of depth‐integrated soil‐water storage within the rain–snow transition zone of the southern Sierra Nevada. The algorithm—the multivariate, non‐parametric regression‐tree estimator Random Forest—was used to predict soil‐water storage based on a combination of attributes at each instrument cluster (soil texture, topographic wetness index, elevation, northness, and canopy cover). Out‐of‐bag R ² (similar to cross‐validation for Random Forest) was used to quantify the accuracy of the estimator for unobserved data. Accuracy was consistently high during the wet‐up, snow‐cover, and early recession periods of average and wet years. The accuracy declined at the end of a 3‐yr dry period, and the relative rank of the independent variables in the model shifted. Soil texture was the highest‐ranked independent variable across all years, followed by elevation and northness. Topographic wetness increased in importance during dry periods. Northness exhibited high importance during the wet‐up and early recession periods of most water years. During dry years, the importance of elevation declined. In dry years, notable differences in soil‐water storage at each depth include lower‐than‐average storage in the deeper regolith at the beginning of the water year and lower storage in near‐surface layers during the winter resulting from transient snow cover.

Existing methods for wireless-sensor network (WSN) topology optimization employ simplifying assumptions of a fixed communication radius between network nodes, which is ill-suited for IoT networks deployed in complex terrain. This article proposes a data-driven approach to WSN topology optimization, employing a Bayesian link classifier trained on LIDAR-derived terrain characteristics and an in-situ survey of link quality. The classifier is trained to predict where good network links (packet-delivery ratio, PDR>0.5) are likely to form in a region given complex terrain attributes. Then, given numerous candidate wireless node placements throughout the domain, the classifier is used to construct an undirected weighted graph of the potential connectivity across the domain. Edge weights in the connectivity graph are proportional to the probability of forming a good link between the nodes. A novel modified cycle-union (MCyU) algorithm for generating a 2-vertex-connected, Steiner minimal network is then applied to the undirected weighted graph of potential network element placements. This ensures a survivable network design, while maximizing the probability of good links within the final network. The total number and spatial distribution of network elements produced by the algorithm is compared to an existing WSN, deployed for environmental monitoring in remote regions. In addition, the MCyU algorithm has been evaluated in three graph test cases to compare with state-of-the-art solutions, where MCyU outperforms in terms of weight minimization and computation time.