Multitier multiscale sensing: A new paradigm for actuated sensing

Many urgent environmental sensing applications, such as the characterization of solar radiation and microclimates within forest ecosystems or the distribution of contaminants in river and lake systems, require observation of phenomena distributed over large spatial domains and characterized by complex spatial and temporal dynamics. Sensing systems including distributed static sensor nodes, robotic actuated sensing, adaptive sensing, and a combination of static sensor nodes and actuated sensors have been proposed to enable efficient and high performance characterization of such phenomena. As we will explain, the first three methods have fundamental limitations that inhibit their ability to characterize complex spatial and temporal dynamics over large spatial domains with high fidelity. On the other hand, the last proposed method can characterize spatio-temporal dynamics with high fidelity; however, this method has unwanted trade-offs that are inherent to the methodology.;We propose a generalized hierarchical method termed Multitier Multiscale Sensing that introduces a hierarchy of sensors according to sensing fidelity, spatial and temporal coverage, and mobility characteristics. This hierarchy scales from a single to many tiers. We implemented a version of the first three algorithms mentioned and compared their performance via the application of monitoring solar radiation intensity, which is important in determining plant growth, in a forest understory. Extensive empirical evaluations were performed using a simulation system based on synthetic data and data collected in-field, as well as with in-field robotic actuated sensing systems. We explore the influence of system parameters - including sensor measurement time, robotic speed, and number of robots - and analyze the effect these parameters have on algorithm execution time and on the number of observations required to achieve a given performance metric. Results show that for monitoring dynamic spatio-temporal phenomena, Multitier Multiscale Sensing generally outperforms the other classes of sensing algorithms. For other phenomena that are characterized by spatial and temporal dynamics over large spatial domains - such as additional environmental phenomena as well as other diverse applications including public health monitoring, precision agriculture, and security that fit the phenomena criteria - we would expect similar trends.;We also present a more intelligent version of Multitier Multiscale Sensing called MUST - a MUltiscale approach with STochastic modeling. MUST was motivated in part by framing the problem of high fidelity environmental sensing as a sensor placement and path planning problem. Path planning involves adaptively selecting only a small subset of locations to make observations, while still efficiently predicting data at unobserved locations. MUST is a hierarchical approach that models the phenomena as a stochastic Gaussian Process that is exploited to select a near-optimal subset of observation locations. We discuss in detail our proposed algorithm for the application of monitoring light intensity in a forest understory. We performed extensive empirical evaluations using a simulation system based on field data and on a physical cabled robotic system to validate the effectiveness of our proposed algorithm. Results indicate that the hybrid approach of MUST considerably and consistently outperforms a provably efficient path planning algorithm for light intensity monitoring.