Wrist-worn accelerometers are gaining popularity for measurement of physical activity. However, few methods for predicting physical activity intensity from wrist-worn accelerometer data have been tested on data not used to create the methods (out-of-sample data). This study utilized two previously collected data sets [Ball State University (BSU) and Michigan State University (MSU)] in which participants wore a GENEActiv accelerometer on the left wrist while performing sedentary, lifestyle, ambulatory, and exercise activities in simulated free-living settings. Activity intensity was determined via direct observation. Four machine learning models (plus 2 combination methods) and six feature sets were used to predict activity intensity (30-s intervals) with the accelerometer data. Leave-one-out cross-validation and out-of-sample testing were performed to evaluate accuracy in activity intensity prediction, and classification accuracies were used to determine differences among feature sets and machine learning models. In out-of-sample testing, the random forest model (77.3–78.5%) had higher accuracy than other machine learning models (70.9–76.4%) and accuracy similar to combination methods (77.0–77.9%). Feature sets utilizing frequency-domain features had improved accuracy over other feature sets in leave-one-out cross-validation (92.6–92.8% vs. 87.8–91.9% in MSU data set; 79.3–80.2% vs. 76.7–78.4% in BSU data set) but similar or worse accuracy in out-of-sample testing (74.0–77.4% vs. 74.1–79.1% in MSU data set; 76.1–77.0% vs. 75.5–77.3% in BSU data set). All machine learning models outperformed the euclidean norm minus one/GGIR method in out-of-sample testing (69.5–78.5% vs. 53.6–70.6%). From these results, we recommend out-of-sample testing to confirm generalizability of machine learning models. Additionally, random forest models and feature sets with only time-domain features provided the best accuracy for activity intensity prediction from a wrist-worn accelerometer. NEW & NOTEWORTHY This study includes in-sample and out-of-sample cross-validation of an alternate method for deriving meaningful physical activity outcomes from accelerometer data collected with a wrist-worn accelerometer. This method uses machine learning to directly predict activity intensity. By so doing, this study provides a classification model that may avoid high errors present with energy expenditure prediction while still allowing researchers to assess adherence to physical activity guidelines.
We developed a new ordinary differential equation (ODE) to represent subsurface flows from a hillslope with artificial subsurface drainage (tiling) into an adjacent channel. Our ODE is based on a derived storage‐discharge relationship from high‐resolutions HYDRUS3D simulations that solve 3D partial differential Richards' equations for a hillslope (plot scale) with internal drainage surfaces, variable soil depth, hydraulic conductivity, and varying slope in the direction of flow. The ODE does not capture the hysteresis in the storage‐drainage relation that can only be captured using partial differential equations (PDEs), however the simplification does not have major impacts on the modeled flows. The nondimensional representation of the equation is consistent with, and provides an explanation for, the empirical equation used in classical hydrological models such as the ARNO (Arno River) and Variable Infiltration Capacity (VIC) model to simulate subsurface flows, which removes the need for parameter calibration. Our equation captures the features not captured by other ODE simplifications such as DRAINMOD. The ODE can be coupled to the Hillslope Link Model (HLM) flood forecasting system allowing real‐time regional simulation of streamflow fluctuations and floods over large domains. We show that (i) the new equation improves the capability to model watershed scale hydrograph recessions and to better capture the timing above flood thresholds for multiple spatial scales from 0.1 to 18,000 km2; (ii)using realistic values of slope combined with typical values of tile separation and soil depth provides good predictive power at gauged basin outlets; and (iii) parameters can be derived for particular situations using the 3D PDE framework.
This study evaluates the potential for a newly proposed non-linear subsurface flux equation to improve the performance of the hydrological Hillslope Link Model (HLM). The equation contains parameters that are functionally related to the hillslope steepness and the presence of tile drainage. As a result, the equation provides better representation of hydrograph recession curves, hydrograph timing, and total runoff volume. The authors explore the new parameterization’s potential by comparing a set of diagnostic and prognostic setups in HLM. In the diagnostic approach, they configure 12 different scenarios with spatially uniform parameters over the state of Iowa. In the prognostic case, they use information from topographical maps and known locations of tile drainage to distribute parameter values. To assess performance improvements, they compare simulation results to streamflow observations during a 17-year period (2002–2018) at 140 U.S. Geological Survey (USGS) gauging stations. The operational setup of the HLM model used at the Iowa Flood Center (IFC) serves as a benchmark to quantify the overall improvement of the model. In particular, the new equation provides better representation of recession curves and the total streamflow volumes. However, when comparing the diagnostic and prognostic setups, the authors found discrepancies in the spatial distribution of hillslope scale parameters. The results suggest that more work is required when using maps of physical attributes to parameterize hydrological models. The findings also demonstrate that the diagnostic approach is a useful strategy to evaluate models and assess changes in their formulations.