A hybrid integrated deep learning model for the prediction of citywide spatio-temporal flow volumes

ABSTRACT

The spatio-temporal residual network (ST-ResNet) leverages the power of deep learning (DL) for predicting the volume of citywide spatio-temporal flows. However, this model, neglects the dynamic dependency of the input flows in the temporal dimension, which affects what spatio-temporal features may be captured in the result. This study introduces a long short-term memory (LSTM) neural network into the ST-ResNet to form a hybrid integrated-DL model to predict the volumes of citywide spatio-temporal flows (called HIDLST). The new model can dynamically learn the temporal dependency among flows via the feedback connection in the LSTM to improve accurate captures of spatio-temporal features in the flows. We test the HIDLST model by predicting the volumes of citywide taxi flows in Beijing, China. We tune the hyperparameters of the HIDLST model to optimize the prediction accuracy. A comparative study shows that the proposed model consistently outperforms ST-ResNet and several other typical DL-based models on prediction accuracy. Furthermore, we discuss the distribution of prediction errors and the contributions of the different spatio-temporal patterns.     

The recognition of former melt flux through high‐strain zones

High‐strain zones are potential pathways of melt migration through the crust. However, the identification of melt‐present high‐strain deformation is commonly limited to cases where the interpreted volume of melt “frozen” within the high‐strain zone is high (>10%). In this contribution, we examine high‐strain zones in the Pembroke Granulite, an otherwise low‐strain outcrop of volcanic arc lower crust exposed in Fiordland, New Zealand. These high‐strain zones display compositional layering, flaser‐shaped mineral grains, and closely spaced foliation planes indicative of high‐strain deformation. Asymmetric leucosome surrounding peritectic garnet grains suggest deformation was synchronous with minor amounts of in situ partial melting. High‐strain zones lack typical mylonite microstructures and instead display typical equilibrium microstructures, such as straight grain boundaries, 120° triple junctions, and subhedral grain shapes. We identify five key microstructures indicative of the former presence of melt within the high‐strain zones: (a) small dihedral angles of interstitial phases; (b) elongate interstitial grains; (c) small aggregates of quartz grains with xenomorphic plagioclase grains connected in three dimensions; (d) fine‐grained, K‐feldspar bearing, multiphase aggregates with or without augite rims; and (e) mm‐ to cm‐scale felsic dykelets. Preservation of key microstructures indicates that deformation ceased as conditions crossed the solidus, breaking the positive feedback loop between deformation and the presence of melt. We propose that microstructures indicative of the former presence of melt, such as the five identified above, may be used as a tool for recognising rocks formed during melt‐present high‐strain deformation where low (<5%) volumes of leucosome are “frozen” within the high‐strain zone.     

Rill flow resistance law under equilibrium bed‐load transport conditions

In this paper, a recently deduced flow resistance equation for open channel flow was tested under equilibrium bed‐load transport conditions in a rill. First, the flow resistance equation was deduced applying dimensional analysis and the incomplete self‐similarity condition for the flow velocity distribution. Then, the following steps were carried out for developing the analysis: (a) a relationship (Equation  13 ) between the Γ function of the velocity profile, the rill slope, and the Froude number was calibrated by the available measurements by Jiang et al.; (b) a relationship (Equation  17 ) between the Γ function, the rill slope, the Shields number, and the Froude number was calibrated by the same measurements; and (c) the Darcy–Weisbach friction factor values measured by Jiang et al. were compared with those calculated by the rill flow resistance equation with Γ estimated by Equations  13 and 17 . This last comparison demonstrated that the rill flow resistance equation, in which slope and Shields number, representative of sediment transport effects, are introduced, is characterized by the lowest values of the estimate errors.     

Effects of antecedent moisture and macroporosity on infiltration and water flow in frozen soil

Infiltration into frozen soil plays an important role in soil freeze–thaw and snowmelt-driven hydrological processes. To better understand the complex thermal energy and water transport mechanisms involved, the influence of antecedent moisture content and macroporosity on infiltration into frozen soil was investigated. Ponded infiltration experiments on frozen macroporous and non-macroporous soil columns revealed that dry macroporous soil produced infiltration rates reaching 10³ to 10⁴ mm day⁻¹, two to three orders of magnitude larger than dry non-macroporous soil. Results suggest that rapid infiltration and drainage were a result of preferential flow through initially air-filled macropores. Using recorded flow rates and measured macropore characteristics, calculations indicated that a combination of both saturated flow and unsaturated film flow likely occurred within macropores. Under wet conditions, regardless of the presence of macropores, infiltration was restricted by the slow thawing rate of pore ice, producing infiltration rates of 2.8 to 5.0 mm day⁻¹. Reduced preferential flow under wet conditions was attributed to a combination of soil swelling, due to smectite-rich clay (that reduced macropore volume), and pore ice blockage within macropores. In comparison, dry soil column experiments demonstrated that macropores provided conduits for water and thermal energy to bypass the frozen matrix during infiltration, reducing thaw rates compared with non-macroporous soils. Overall, results showed the dominant control of antecedent moisture content on the initiation, timing, and magnitude of infiltration and flow in frozen macroporous soils, as well as the important role of macropore connectivity. The study provides an important data set that can aid the development of hydrological models that consider the interacting effects of soil freeze–thaw and preferential flow on snowmelt partitioning in cold regions.     

Characterization of sudden and sustained base flow jump hydrologic behaviour in the humid seasonal tropics of the Panama Canal Watershed

Base flows are important for tropical regions with pronounced dry seasons, which are facing increasing water demands. Base flow generation, however, is one of the most challenging hydrological processes to characterize in the tropics. In many years during the May–December wet season in the Panama Canal Watershed (PCW), base flows in rivers abruptly increase. This increase persists until the start of the December–April dry season. Understanding this unusual base flow jump (BFJ) behaviour is critical to improve water provisioning in the seasonal tropics, especially during droughts and extended dry seasons. This study developed an integrated approach combining piecewise regression on cumulative average base flow and sensitivity analysis to calculate the timing and magnitude of BFJ. Rainfall, forest cover, mean land surface slope, catchment area, and estimated subsurface storage were tested as predictors for the occurrence and magnitude of the BFJs in seven subcatchments of the PCW. Sensitivity analysis on correlated predictors allowed ranking of predictor contributions due to isolated and cross-correlation effects. Correlations between observed BFJs and BFJs predicted by watershed and rainfall-related predictors were 0.92 and 0.65 for BFJ timing and magnitude, respectively. Forest cover was the second most significant predictor after cumulative rainfall for jump magnitude, owing to larger subsurface storage and groundwater recharge in forests than pastures. Catchments in the mountainous eastern PCW always generated larger jumps due to their higher rainfall and greater forest cover than the western PCW catchments. The cross-correlations between predictors contributed to more than 50% of the jump variances. The results demonstrate the importance of rainfall gradient and catchment characteristics in affecting the sudden and sustained BFJs, which can help inform land management decisions intended to enhance water supplies in the tropics. This study underscores the need for more research to further understand the hydrological processes involved in the BFJ phenomenon, including better BFJ models and field characterizations, to help improve tropical ecosystem services under a changing environment.     

Characterization of complex groundwater flows in the environment of singular buildings by combining hydrogeological and non-destructive geophysical (ground-penetrating radar) techniques: Punta Begoña Galleries (Getxo,Spain)

Locating and quantifying groundwater flow in many built-up areas are a priority with regard to its complete restoration. In this work, a hydrogeological survey of the surroundings of the Punta Begoña Galleries (Getxo, Bizkaia), built on a coastal cliff, was completed by using ground penetrating radar (GPR) testing. Thus, the preliminary characterization of soils and rocks in accessible areas of the cliff was first improved by hydrogeological information gathered from a single survey borehole, including permeability measurements by low pressure injection tests (LPTs) and continuous water level monitoring. As a complementary method, the non-destructive GPR technique was performed during both dry and wet hydrological periods and in tandem with the injection tests, providing more complete spatial and temporal images of water flows. Specifically, GPR allows mapping of flow paths in soils and assessing the continuity of fractures in rock masses. Altogether, this complementary approach provides greater knowledge of complex underground flow dynamics in built environments, thus making it easier to make decisions for their management.     

Dynamic,discontinuous stream networks: hydrologically driven variations in active drainage density,flowing channels and stream order

Despite decades of research on the ecological consequences of stream network expansion, contraction and fragmentation, surprisingly little is known about the hydrological mechanisms that shape these processes. Here, we present field surveys of the active drainage networks of four California headwater streams (4–27 km²) spanning diverse topographic, geologic and climatic settings. We show that these stream networks dynamically expand, contract, disconnect and reconnect across all the sites we studied. Stream networks at all four sites contract and disconnect during seasonal flow recessions, with their total active network length, and thus their active drainage densities, decreasing by factors of two to three across the range of flows captured in our field surveys. The total flowing lengths of the active stream networks are approximate power‐law functions of unit discharge, with scaling exponents averaging 0.27 ± 0.04 (range: 0.18–0.40). The number of points where surface flow originates obey similar power‐law relationships, as do the lengths and origination points of flowing networks that are continuously connected to the outlet, with scaling exponents averaging 0.36–0.48. Even stream order shifts seasonally by up to two Strahler orders in our study catchments. Broadly, similar stream length scaling has been observed in catchments spanning widely varying geologic, topographic and climatic settings and spanning more than two orders of magnitude in size, suggesting that network extension/contraction is a general phenomenon that may have a general explanation. Points of emergence or disappearance of surface flow represent the balance between subsurface transmissivity in the hyporheic zone and the delivery of water from upstream. Thus the dynamics of stream network expansion and contraction, and connection and disconnection, may offer important clues to the spatial structure of the hyporheic zone, and to patterns and processes of runoff generation. Copyright © 2014 John Wiley & Sons, Ltd.     

Use of a watershed hydrologic model to estimate interbasin groundwater flow in a Costa Rican rainforest

The watershed hydrologic model TOPMODEL was used to estimate interbasin groundwater flow (IGF) into a small lowland rainforest watershed in Costa Rica. IGF is a common hydrological process but often difficult to quantify. Four‐year simulations (2006–2009) using three different model approaches gave estimates of IGF that were very similar to each other (10.1, 10.2, and 9.8 m/year) and to an earlier estimate (10.0 m/year) based on 1998–2002 data from a budget study that did not use a hydrologic simulation model, providing confidence in the new estimates and suggesting each of the three model approaches is viable. Results show no significant temporal variation in IGF during 2006–2009 (or between this period and the earlier study from 1998–2002). Simulations of the 16 consecutive 3‐month periods in 2006–2009 gave 16 values of IGF rate with a mean (10.1 m/year, standard deviation = 0.6 m/year) very similar to the estimates above from the 4‐year simulations. This suggests the modified version of TOPMODEL can be used to model stream discharge and estimate IGF for sub‐annual time periods during which change in water storage is not necessarily equal to zero. Thus, simple watershed models may be used to estimate IGF based on even relatively short calibration periods, making such models useful tools in the study of this widespread hydrological process that affects water and chemical fluxes and budgets but is often difficult and costly to quantify. Copyright © 2013 John Wiley & Sons, Ltd.     

The effect of storm movement on infiltration,runoff and soil erosion in a semi-arid catchment

Rainfall characteristics are key factors influencing infiltration and runoff generation in catchment hydrology, particularly for arid and semiarid catchments. Although the effect of storm movement on rainfall-runoff processes has been evaluated and emphasized since the 1960s, the effect on the infiltration process has barely been considered. In this study, a physically based distributed hydrological model (InHM) was applied to a typical semi-arid catchment (Shejiagou, 4.26 km²) located in the Loess Plateau, China, to investigate the effect of storm movement on infiltration, runoff and soil erosion at the catchment scale. Simulations of 84 scenarios of storm movement were conducted, including storms moving across the catchment in both the upstream and downstream directions along the main channel, while in each direction considering four storm moving speeds, three rainfall depths and two storm ranges. The simulation results showed that, on both the hillslopes facing downstream (facing south) and in the main channel, the duration of the overland flow process under the upstream-moving storms was longer than that under the downstream-moving storms. Thus, the duration and volume of infiltration under upstream-moving storms were larger in these areas. For the Shejiagou catchment, as there are more hillslopes facing downstream, more infiltration occurred under the upstream-moving storms than the downstream-moving storms. Therefore, downstream-moving storms generated up to 69% larger total runoff and up to 351% more soil loss in the catchment than upstream-moving storms. The difference in infiltration between the storms moving upstream and downstream decreased as the storm moving speed increased. The relative difference in total runoff and sediment yield between the storms moving upstream and downstream decreased with increasing rainfall depth and storm speed. The results of this study revealed that the infiltration differences under moving storms largely influenced the total runoff and sediment yield at the catchment scale, which is of importance in runoff prediction and flood management. The infiltration differences may be a potential factor leading to different groundwater, vegetation cover and ecology conditions for the different sides of the hillslopes.     

Conditional statistics of Reynolds shear stress in flow over 2D dunes in the presence of surface waves

This study presents the analysis of the velocity fluctuations to describe the conditional statistics of Reynolds shear stress in flow over two‐dimensional dunes in the presence of surface waves of varying frequency. The flow velocity measurements over the dunes are made using a 16‐MHz 3D acoustic Doppler velocimeter. The joint probability distributions of the normalized stream‐wise and vertical velocity fluctuations at different vertical locations are calculated in the trough region of a selected dune in quasi‐steady region of the flow. Third‐order moments of the stream‐wise and vertical velocity components over one dune length are also calculated throughout the flow depth for understanding the effect of surface waves on relative contributions to the Reynolds shear stress due to the four quadrant events. The structure of instantaneous Reynolds stresses is analysed using quadrant analysis technique. It has been shown that the contributions of second and fourth quadrant events to the Reynolds shear stress increase with increase in the frequency of surface waves. In fact, the largest contribution to turbulent stresses comes from the second quadrant. The cumulant discard method is applied to describe the statistical properties of the covariance term u′w′. Conditional statistics and conditional sampling are used to compare the experimental and theoretical relative contributions to the Reynolds shear stress from the four quadrant events. Copyright © 2013 John Wiley & Sons, Ltd.