Park at the Boquillas Crossing parking lot near Boquillas Canyon. After passing through the port of entry visitors are ferried across the Rio Grande on a small rowboat for a modest fee ($5 round-trip). Walking across the river is permitted only at the Boquillas Crossing, but is not recommended if the river level is high.Once across the river visitors have the option of walking to the village (1/2 mile) or paying an additional fee to ride on a burro, horse, or in a vehicle. Local guides are available. Visitors are required to check in with Mexican immigration officials upon arrival in Boquillas, and pay a small entrance fee ($3) to the Mexican Protected Area that Boquillas is situated in. A small wrist bracelet is your receipt for the entrance fee. Visitors planning to stay overnight in Mexico will need to apply for a temporary visa. Secure overnight parking at the Boquillas Crossing Port of Entry may be available.
At 8:25 p.m. Friday, the first raft of the night slid across the green river near Roma, a quick trip on still waters the length of a football field. Smugglers stopped about 10 feet offshore and began pushing a dozen migrants, some holding infants and babies, into the shallows.
Across The Rio Grande
Others wondered aloud about sending their children across the border alone. If they crossed alone, the U.S. would have to allow them to stay, the migrants knew. They had seen Border Patrol separate unaccompanied youths at the river.
Marjorie and her 6-year-old son, also victims of the kidnapping, arrived at the river around 6 p.m. A Venezuelan man helped the little boy across a shallow area of the river near the Puente Negro railroad bridge. Marjorie followed behind, clutching a bag of belongings.
Earlier in the day, a few hundred migrants had formed a line along the El Paso side of the river. Marjorie and hundreds of others from the caravan of buses who streamed across the river joined the line, where some people had started to build fires to stay warm. Others crossed back to Juárez to buy water and food for those in line.
To quantify spatial and temporal variations and long-term trends in seasonal snowpack properties across the Rio Grande headwaters (Fig. 1), we ran SnowModel simulations over a 34-yr period from 1984 to 2017. SnowModel is a spatially distributed physically based snow evolution modeling system designed for application in a wide range of environments where snow occurs (Liston and Elder 2006b). SnowModel includes the following submodels: MicroMet (Liston and Elder 2006a), a high-resolution meteorological distribution model; EnBal (Liston 1995), which computes surface energy exchanges between the snow and atmosphere; SnowPack (Liston and Hall 1995), which simulates the seasonal evolution of snow depth and SWE; SnowTran-3D (Liston and Sturm 1998; Liston et al. 2007), a 3D model that simulates snow redistribution by wind over topographically variable terrain; and SnowAssim (Liston and Hiemstra 2008), a data assimilation system that can be used to assimilate field and remote sensing observations into SnowModel simulations. Required inputs to run SnowModel include temporally varying fields of air temperature T, relative humidity, wind speed and direction, and precipitation P as well as spatially varying fields of topography and land cover. The model uses known relations between meteorological variables and the surrounding topography and land cover to distribute those variables across the domain. Surface pressure and incoming solar and longwave radiation are additional atmospheric property distributions required to run SnowModel and are computed as part of the MicroMet submodel as described in detail in Liston and Elder (2006a). SnowModel has been rigorously evaluated and shown to perform well in seasonally snow-covered environments similar to the study area (e.g., Greene et al. 1999; Hiemstra et al. 2006; Liston and Elder 2006a,b; Liston et al. 2008; Prasad et al. 2001; Sexstone et al. 2018; Sproles et al. 2013).
SnowModel simulations were run at a 3-hourly time step with a 100-m spatial grid resolution (1.12 million grid cells across the study domain). Elevation, land cover, and canopy cover fraction were provided by the U.S. Geological Survey (USGS) National Elevation Dataset ( ) and National Land Cover Database spatial datasets ( ). The effective leaf area index (LAI*) values across the domain were created by scaling the maximum LAI* for each forest class vegetation type (Table S2 in the online supplemental material) by the canopy cover fraction (e.g., Broxton et al. 2015; Sexstone et al. 2018). This study also implemented dynamic land-cover changes across the simulation time period that were updated each simulation year to represent the temporal effects of BB and wildfire disturbance on snowpack evolution. The cumulative distribution of BB-induced tree mortality from 2000 through 2015 was divided into quantiles representing no mortality, light mortality, moderate mortality, and severe mortality (Bright et al. 2013). Aerial surveys of cumulative BB mortality (USFS 2016) from each year of the simulation time period were used to reduce the spatially variable LAI* across the study domain by 0% (no mortality), 5% (light mortality), 25% (moderate mortality), and 40% (severe mortality) (Bright et al. 2013; Pugh and Gordon 2013; Sexstone et al. 2018). Furthermore, areas subject to wildfire disturbance following the 2013 West Fork Complex Fire (Fig. 1) were prescribed a category of very severe mortality with a LAI* reduction of 82%, which was the difference in LAI* between burned and unburned forest measurements presented by Gleason and Nolin (2016). Deforestation associated with forest management and salvage logging was not considered in this study.
Meteorological forcing data were provided by the 1/8 grid spacing North American Land Data Assimilation System (NLDAS-2) reanalysis forcing dataset (Fig. 1; ; Mitchell et al. 2004; Xia et al. 2012). The NLDAS-2 reanalysis forcing dataset extends back to January 1979; however, given the anomalously wet period in the early 1980s across the region, we began our analysis in 1984 to avoid potential spurious trends (Harpold et al. 2012). Hourly NLDAS-2 T, P, relative humidity, and wind speed and direction data were aggregated to 3-hourly values to correspond with the model simulation time step. The 3-hourly NLDAS-2 forcing data along with mean elevation of each NLDAS-2 grid cell was used by MicroMet to downscale and create the 100-m spatial resolution meteorological forcing data required by SnowModel (e.g., Liston and Elder 2006a; Liston and Hiemstra 2011; Liston et al. 2008; Sexstone et al. 2018) based on Northern Hemisphere monthly lapse rates (refer to Liston and Elder 2006a) and the 100-m digital elevation model. A 2C rain/snow threshold air temperature was used in SnowModel for partitioning between rain and snow precipitation (Auer 1974). The SnowAssim data assimilation system (Liston and Hiemstra 2008) was used in this application of SnowModel (e.g., Fletcher et al. 2012; Liston et al. 2008, 2018) to adjust biases in NLDAS-2 winter P forcing data (e.g., Henn et al. 2018; Sexstone et al. 2018). Differences between observed and simulated SWE before peak snow accumulation each year at six long-term SNOTEL stations across the study domain were used to spatially interpolate winter season P adjustment factors for each year of the simulation period (Liston and Hiemstra 2008; Table S3).
Field snow surveys were conducted in 2016 and 2017 at 73 locations within the study area (Fig. 1) to measure snow properties in both forested and nonforested locations. At each field measurement location (n = 73), nine snow depth measurements were collected using a snow depth probe across a 20 m 20 m grid and a mean determined for each location. At 46 of the measurement locations, snow density was measured using either a Federal snow sampling tube (Kinar and Pomeroy 2015) or snowpit observations using a 250-cm3 snow density wedge cutter (Kinar and Pomeroy 2015), and SWE was computed based on the mean snow depth snow density. Field measurements were directly compared to simulated snow depth and SWE at grid cells that overlapped the field measurement locations on each date. Field snow survey data are available in Sexstone (2020).
Simulated SCA between 2000 and 2017 was evaluated using remotely sensed MODIS SCA data from the MOD10A2 product ( ). The 8-day MODIS SCA product was used in favor of the daily MODIS SCA product to minimize the influence of cloud coverage on the model evaluation of snow-cover duration. Model grid cells were classified as snow covered if the simulated SWE was greater than 10 mm on the same day (Gascoin et al. 2013). The MOD10A2 maximum SCA across the model domain (observed by MODIS for every 8-day period) was compared to the maximum SCA simulated by SnowModel during the same period.
Simulated SWE across the Rio Grande headwaters was also evaluated using the National Aeronautics and Space Administration (NASA) Jet Propulsion Laboratory (JPL) Airborne Snow Observatory (ASO) dataset that was collected near peak snow accumulation in the basin on 3 April 2016 (Painter 2018; _50 M_SWE/versions/1). ASO performs airborne surveys to estimate snow depths at a 3-m resolution by differencing snow-off from snow-on elevations obtained from lidar (Painter et al. 2016). The ASO SWE product is then derived at a 50-m spatial resolution by combining the ASO snow depths with simulated snow density fields (Hedrick et al. 2018; Painter et al. 2016). We resampled the 50-m ASO SWE dataset to the 100-m SnowModel grid and directly compared it to the SnowModel SWE simulated on 3 April 2016.
Simulated snow-cover duration across the model domain was highly related to MODIS SCA observations (R2 = 0.90; p value 0.05; Fig. 2). 2ff7e9595c
コメント