Leveraging real-time hydrologic data for the control of ...glaser.berkeley.edu/glaserdrupal/pdf/SPIE 11 branko.pdf · flood-control, irrigation, recreation, hydropower generation,

*[email protected]

Leveraging real-time hydrologic data for the control of large-scale water distribution systems in the Sierra Nevada

Branko Kerkez*a, Steven Glaser a

aUC Berkeley, Department of Civil and Environmental Engineering, Berkeley, CA 94704.

ABSTRACT

Recent water shortages, particularly evident in the state of California, are calling for better predictive capabilities, and improved management techniques for existing water distribution infrastructure. One particular example involves large-scale water distribution systems (on the scale of reservoirs and dams) in the Sierra Nevada, where the majority of the state’s water is obtained from melting snow. Current control strategies at this scale rely on sparse data sets, and are often based on statistical predictions of snowmelt. Sudden, or unexpected, snowmelt can thus often lead to dam-overtopping, or downstream flooding. This paper assesses the feasibility of employing real-time hydrologic data, acquired by large-scale wireless sensor networks (WSNs), to improve current water management strategies. A sixty node WSN, spanning a square kilometer, was deployed in the Kings River Experimental Watershed, a research site in the Southern Sierra Nevada, at an elevation of 1,600-2,000 m. The network provides real time information on a number of hydrologic variables, with a particular emphasis on parameters pertaining to snowmelt processes. We lay out a system architecture that describes how this real-time data could be coupled with hydrologic models, estimation-, optimization-, and control-techniques needed to develop an automated water management infrastructure. We also investigate how data obtained by such networks could be used to improve predictions of water quantities at nearby reservoirs. Keywords: wireless sensor networks, hydrologic monitoring, water resources

1. INTRODUCTION By most metrics, the current water infrastructure in the state of California is struggling to meet the requirements imposed by its growing population and a thriving agricultural sector[27][28][29]. Recent droughts are a motivating example that informed decision-making will require new management techniques to ensure equitable and reliable distribution of water supplies. We present a brief review of current water management practices, and propose a system-level solution to improve monitoring and forecasting of water resources in California. As with any complex system, detailed spatial and temporal data are required to meet the need of modern hydrologic models. We will show that the current state of the monitoring infrastructure is not sufficient to meet the requirements imposed by such models, and that use of modern wireless sensor networks provides a workable and affordable real-time monitoring solution. 1.1 Water in California

The current population of California is estimated at 36 million, and is expected to grow to 60 million by 2050 [6]. The state boasts one of the most productive agricultural sectors in the world, and contains a plethora of forest terrain and environmental habitats. Aside from weathering historic drought seasons, flood risks are growing due to aging levees and extreme weather scenarios.

California’s water infrastructure composes a broad, multi-use system that balances the water supply to meet goals of flood-control, irrigation, recreation, hydropower generation, and other uses. Water demands in the state are met by a combination of federal, state, and local management projects. The majority of the supply is dedicated to meeting agricultural irrigation needs, with a smaller fraction going to urban use [6]. Historically, 9-30% of the state’s energy is also generated via hydropower. The major hydrologic parameter associated with hydropower production is snowfall [1].

1.2 Current management practices

A closer look at the problem traces the origins of the water supply to the Sierra Nevada, a mountain range spanning a 600 km north-south transect of the state. Precipitation is deposited in the wintertime as snow, which acts as storage, and melts throughout the year to provide a steady supply of water. It is estimated that seasonal snow cover is the primary source of water for over 60 million people in the western United States, and that melting snow is responsible for 80% or more of soil moisture and stream flow in semi-arid mountain basins [3, 17]. As such, major state resources are devoted to the monitoring of snow related phenomena in the Sierra Nevada.

Currently, the main source of snow data is provided by the Snow Telemetry network (SNOTEL), which is operated by the National Resource Conservation Service (NRCS), and features 1700 snowcourse and real-time measurement locations [5, 22, 23]. A snowcourse is a pre-selected location, where manual observations of snow-depth and snow water equivalent (SWE) are made on a monthly basis. Snowcourses were selected to be representative of particular water-producing regions, and were chosen begging in in the 1930s through empirical means, with the main criteria of general accessibility and limited public disturbance [22]. Combined with few automated measurement stations throughout the state, snowcourse measurements are currently a primary means by which to derive water supply forecasts. The automated SNOTEL sites also collect real-time meteorological data that can be employed for prediction purposes.

A water supply forecasts projects the volume of water that will be expected over a specified period of time. These forecasts are currently conducted in a statistical fashion, leveraging historical data, and correlations between previously observed SNOTEL readings and water availability. Such predictive methods rely heavily on principal component analysis, and regression techniques to predict future streamflow and SWE behavior [9]. These streamflow predictions are then used by decision makers to optimize water distribution.

1.3 Challenges

A number of challenges arise with current water management practices, particularly relating to data collection and the general uncertainty associated with forecasting methods. While the study of hydrologic phenomena is a relatively mature field, the hydrologic processes pertaining to water and energy fluxes in mountainous regions, where basins are largely dominated by snowmelt, are yet not well captured by current hydrologic models [3]. In particular, it is not known to what extent, if any, recent trends in climate may affect the snowpack. Snow distribution is known to vary significantly across space, even along short transects [13,18], but current monitoring strategies provide very sparse data. Automated SNOTEL sites give convenient daily measurements but only for specific points, which are difficult to integrate into representative basin-wide estimates. SNOTEL locations were selected in part due to ease of accessibility (flat terrain, for example) and thus suffer further from a lack of being able to capture the range of possible terrain seen in most basins. In some cases, SNOTEL site readings were found to be 200% greater than the their respective basin mean [20]. Alternatively, manual snowcourses can provide more representative estimates of basin-wide snow cover, but such snow surveys are costly and can only be conducted on a monthly basis [7].

SNOTEL point measurements have been integrated into basin volumes using hypsometry and variogram analysis [7], and, in some cases, through further help of satellite observations [2]. Such methods are however still impeded by sparsely available point measurements and the inherent non-linearity associated with snow-cover distribution. Furthermore, it has been shown that spatial patterns of snow accumulation are not governed by spatial proximity, but rather by complex interactions between topographic terrain parameters [4,8,20]. To capture true basin-wide snow cover, it thus simply not enough to increase the number of point measurements, but to also take measurements in a strategic fashion.

The lack of knowledge pertaining to the spatial distribution of snow is currently preventing the implementation of deterministic, or physical, models for purposes of water forecasting. At present, water resource officials are limited to a standard set of statistical procedures to determine streamflow based on historical data. Pogano et al conducted a comprehensive analysis of such statistical methods, and showed that predictions inaccuracies were not only basin-specific, but also dependent on specific times of of the snowmelt season [21]. Statistical methods tend to perform well when predictions are made within an acceptable window around the historical mean. A continuation of the current warming trend could lead from significant deviation from this historical mean, and has the potential induce a significant amount of error into statistical forecasting methods. Aside from impacting model predictions, further warming will shift the rain-snow transition zone to higher elevations, thus accelerating the snow melting process. Rapid melting of the snowpack could reduce the amount of water available during spring and summer months, and has the potential to reduce hydropower production by 8-50% [1].

2. PROPOSED IMPROVEMENTS Broadly, the uncertainties associated with California’s water management techniques will be significantly reduced by:

1. expanding current monitoring strategies to include more densely sampled real-time data, and 2. replacing statistically based prediction tools with physically-based forecasting models.

Realizing the need for a multidisciplinary approach, we propose a system-level solution to address these needs. Figure 1 shows that these goals can be met by improving sensing and monitoring strategies, developing hydrologic modeling techniques, deriving basin-wide estimation techniques, improving forecast modeling, and addressing the control of water infrastructure.

Figure 1: The layers of the proposed systemic solution for improved water forecasting and management.

2.1 Sensing

Low resolution remote sensing strategies, specifically satellite data, provide information only regarding total snow-area cover. To integrate basin-wide snow volumes, depth measurements are needed across much wider areas than is currently provided by sparse point-wise SNOTEL measurements. Furthermore, this data needs to be sampled and available in real-time to maximize its potential to water resources officials. The monitoring equipment must also be designed to withstand the harsh wintertime conditions experienced in the Sierra Nevada.

Aside from snow-depth, a number of other hydrologic parameters must also me monitored to investigate the dynamics of the snowpack. This includes, but is not limited to, solar radiation and other energy fluxes that drive melt, as well as hydrologic parameters such as soil moisture that govern how melt-water is ultimately distributed throughout basins.

Ultimately, increasing the number of sensors will only provide limited information unless the sensors are placed in a fashion that maximizes the captured variability of the snowpack. Methods need to be developed to determine how many sensors should be placed, and where they should be placed, to provide measurements that are representative of wider areas.

2.2 Hydrologic modeling and analysis

Hydrologic models must be updated to incorporate observations made by densely instrumented watersheds. Current snowmelt models need to be parameterized, and new models need to be developed to determine, in real-time, the quantities of water that are expected to drain from the snowpack. Furthermore, groundwater, and soil moisture models, must be developed to determine how the basin will respond to the melting snow.

2.3 Basin-wide estimation techniques

Estimation techniques must be developed to effectively distribute snow-related point measurements to the basin scale. Snow depth and snow water equivalent values must be integrated over entire basis to determine the actual volume of

• Wireless  sensor  networks  • Sampling  design  and  optimal  sensor  placement  Sensing  

• Snowmelt  modeling  • Basin  response  analysis  

Hydrologic  Modeling  and  Analysis  

• Basin-­‐wide  snow  volume  integration  • SWE  state  estimation  Estimation  

• Basin-­‐wide  stream>low  predicitons  • Deterministic  reservoir  in>low  predictions  Forecasting  

• Optimization  for  hydropower  and  >lood-­‐control  • Veri>ication  and    model-­‐checking  Control  

water that is present in the system. Additionally, state estimation techniques should be used to infer values that are costly to measure (for example, estimating snow density without the need to measure it).

2.4 Forecasting

The above basin-wide SWE estimates can be used as inputs to snowmelt-, and other hydrologic-models to accurately predict streamflow conditions. Such forecasts should be based on deterministic mass-, and energy-balance, models rather than statistical methods, while measures of uncertainty will still need to be provided. With basin-specific stream-flow predictions it will then be possible to combine the contributions of each basin to the overall inflow into reservoirs and dams.

2.5 Control and optimization of water infrastructure

The implementation of the above methods will play a large role in reducing overall uncertainty of current reservoir operations relating to water distribution, flood-control, and hydropower. The overall forecasting windows will shrink to time steps that will make it possible to conduct real-time control of water distribution infrastructure. It will then be possible to employ a number of optimal-control, and optimization strategies, such as the methods described in [16].

3. CURRENT PROGRESS A number of the above steps have already been addressed as part of the initiative described in [17,18,19,20,21,22,25]. In particular, we will focus here on describing the technology that was employed to conduct real-time monitoring of a large water basin in the Southern Sierra Nevada. 3.1 Wireless Sensor Networks

It was determined that the only feasible solution which would permit for the monitoring of large-scale hydrologic systems in real time, would require the use of Wireless Sensor Network (WSN) technology. The wireless network, described in [11,13,15], is built upon technologies developed by Dust Networks, a company focusing on the control and monitoring of industrial processes [24]. The core component of the network is known as a mote, a tiny ultra-low-power embedded system (microcontroller and memory) with an attached low-power radio for wireless capabilities. In a mesh configuration motes relay information to their neighbors, who then relay this information to their neighbors, eventually reaching a network manager, and thus forming a truly multi-hop, redundant, and reliable network. This redundancy reduces the impact of node-loss, which allows the network to transmit data through other paths in the mesh, even if stable links collapse.

A sixty-node Wireless Sensor Network, spanning a square kilometer, was deployed in the Kings River Experimental Watershed, a research site in the Southern Sierra Nevada Mountains, at an elevation of 1,600-2,000 m (figures 2,3,5). This prototype network is part of the NSF sponsored Southern Sierra Critical Zone Observatory (SSCZO) initiative, which aims to set up independent observatories to study the dynamic interactions between the solid Earth and its outer fluid envelope. The network provides real time information on a number of hydrologic variables, with a particular emphasis on those pertaining to snowmelt processes. The network is located in the northern part of the basin and is shown on Figure 2. Twenty-five non-wireless sensors are also co-located in the central portion of the watershed (see figure 2). A number of streams, each instrumented to measure flow, come together in the western portion of the basin and drain in a larger creek. This mid-elevation site contains dense forest cover (pine and cedar), and a number of open meadows. The harsh wintertime conditions, along with fluctuating humidity and temperature make the site a particularly challenging environment for wireless sensing. The network is built upon a time synchronized mesh protocol (TSMP) by DUST Networks [24][25], which enables ultra-low powered, reliable performance through 1) tight time synchronization, 2) slotted communications, 3) frequency channel diversity, and 4) self-meshing automated network formation. Network wide time synchronization in the proprietary TSMP operates on the order of milliseconds, and is loosely based on the IEEE15.4e workgroup standard [26]. Motes are equipped with accurate on-board crystals to keep track of relative time, and are all synced to the network manager upon joining the network. Motes communicate within time widows called slots, and remain in a low-power state for the duration of network operations. A series of slots form a frame, which repeats indefinitely. When more motes join the network, they are assigned communication on unallocated slots. The procedure is shown in figure 4, while more

information on slotted communication is given in [24][25][26]. Multiple paths between each network node (see figure 3) create link diversity, and allow for efficient communications even when stable nodes fade from the network. To mitigate outside radio interference, DUST Networks motes are equipped with the ability to frequency channel hop. In the case of the CCSSZO network, during their allotted slot, motes randomly transmit data on any number of the 15 available channels along the 2.4GHz band. Utilizing this channel diversity has the overall effect on reducing interference due to multipath fading, and reducing interference from radio technologies that share the same frequency band (Bluetooth, WIFI, etc.). In most cases, radio transmissions account for more than 95% of all power used by WSNs [25]. All of the above features, which are not readily available on most other platforms, significantly reduce the amount of time that a mote spends transmitting data and listening to incoming communications. This permits the DUST Networks system to have a near 99% duty cycle, allowing each node to last two years on standard household batteries. For reliability and efficiency purposes, our deployment makes a distinction between two types of network nodes: a sensor-node is a node equipped with a custom data logger, sensors, and a mote; while a repeater node is a mote. Repeater nodes act as extender for sensor-nodes that are out of range, while serving to create further redundancy in the mesh topology.

Figure 2: The Kings River Experimental Watershed. The wireless sensor network is located in the northern portion of the delineated basin.

In 60-node SSCZO network, each of the 24 total sensor-nodes is equipped with the sensors listed in table 1, which provides over 300 total sensors for the entire SSCZO site. The remaining repeater nodes are battery powered, while each sensor node contains 7Ah lead-acid battery and 10 Watt solar panel. The network manager, which schedules mote communications and controls the mesh topology, is located at the base of a 40m eddy-flux tower. It is interfaced to an embedded Linux computer, which is connected to a mobile modem, providing real-time data that is transmitted to offsite servers for aggregation and analysis.

SSCZO

WSN

Parameter Sensor Manufacturer Quantity

Snow depth Ultrasonic Judd Communications Inc 1 Soil moisture, soil temperature EC-TM Decagon Inc 4

Matric potential MPS-1 Decagon Inc 4 Solar radiation LI-200 LI-COR Inc 1

Humidity and Temperature SHT15 Sensirion Inc 1

Table 1: The sensors, and quantity per node, deployed in the SSCZO study.

The wireless portion of the SSCZO network has been operational since May 2009, and is continuing to provide a dense set of data upon which future studies of the basin’s water balance will be based. The efficacy of the network to serve as a real-time hydrologic-, and water-monitoring platform has successfully been validated. A 20-node network, built upon the same technology, is already operational in northern California [12], and more network deployments are planned for the Summer of 2011. This will create a multi-network, or network-of-networks, infrastructure that will serve as a footprint for a grander vision to built a real-time water information system for the state of California.

Figure 3: A snapshot of the state of a subset of the SSCZO WSN. Lines indicate paths between nodes. The reliable mesh ensures multiple connections between nodes.

Figure 4: Time synchronization communications using the TSMP protocol. Source: Dust Networks [25].

3.2 Modeling, estimation, and sensor placement

Aside from developing techniques to effectively deploy WSNs, this initiative has also taken steps to address a number of other layers of the proposed goals in figure 1. A snowmelt model, which employs hybrid systems theory, has been developed and validated on real-world data sets [10]. A sensor placement schemes has been investigated to determine the optimal location of snow-depth sensors, which will permit future monitoring sites to be instrumented with the minimum amount of sensors, while maximizing the amount and value of acquired information. These methods have

been extended to evaluate multi-dimensional Gaussian Process regression techniques that will be used to distribute point-wise snowdepth readings across an entire basin [14]. Coupled with the snowmelt model, these methods will provide volumetric estimates of snow throughout a basin, which will in the future be used to deterministically forecast streamflow conditions.

4. LEVERAGING REAL-TIME DATA To motivate the use of real-time hydrologic data for purposes of forecasting and control, this section will briefly investigate the effects of streamflow at the SSCZO WSN site on water inflow at a nearby dam and reservoir. The location of the SSCZO and the Pine Flat reservoir are shown on the map figure 4 (provided by the USGS). Figure 2 shows that the majority of tributaries on the SSCZO site flow into four main streams, which then aggregate into Dinkey Creek in the south-western portion of the basin. This creek then flows, along with a number of other western tributaries, into the more substantial larger Kings River, (figure 4). The river is received at the Pine Flat reservoir as its primary inflow. The reservoir is operated by the Army Corps of Engineers and is currently being used for flood control, and water storage purposes.

Figure 5: Location of the SSCZO site and nearby Pine Flat reservoir. The sites are separated by about 30km.

Figure 5 plots the stream discharge (in m3/day) originating at the site of the SSCZO WSN. The figure also shows the simultaneous water inflow into the Pine Flat reservoir. The time period corresponds with April-September of 2009, which corresponds with the peak of the snowmelt season. The majority of snowmelt originated during the month of May, as is evident by the relatively large discharge characteristics during this period. There is a clear correlation between peaks of discharge in the stream, and eventual water inflow into the reservoirs. A simple cross correlation reveals an average lag of about 10-14 days between the two systems, which suggest that the SSCZO stream flows could serve as legitimate proxies for predicting water levels at the reservoir. Given that a number of other tributaries flow into the Kings River, the correlation between the two flows in figure 5 should not be interpreted as an exact relationship. Rather, similar WSN deployments at the origins of the other tributaries would shed further light on the contribution of multiple basins to the reservoir inflow. More importantly, when snowmelt, and hydrologic models are coupled with the WSN readings, the predictive window on reservoir inflow will widen, and will provide accurate, physically based forecasts on a monthly basis, or beyond. The real-time nature of the monitoring system will also permit for real-time corrections to forecasts, something that is difficult with the sparse datasets that are currently available to management officials. To avoid flooding or dam-overtopping scenarios, which results due to unexpected weather, and lack of coordination between dam operators, control of reservoirs could be automated by leveraging this real-time data. The safety of water-distribution systems could also be investigated through formal verification techniques to ensure reliable error-prone operations.

SSCZO Site

Pine Flat Reservoir

Figure 5: Stream discharge at the location of the SSCZO WSN, and reservoir inflow at the Pine Flat dam.

5. CONCLUSIONS This paper provided a brief overview of the current challenges associated with the state of water-resource monitoring and forecasting in the state of California. In particular, it was shown that the major source of water in the state is obtained from melting snow during the spring and summer time periods, and that much room for improvement exists with current monitoring and forecasting techniques. A system-level solution was proposed to address a number of these challenges, incorporating a multidisciplinary approach that develops new monitoring techniques, and aims to shift from statistical to physically based forecasting tools. Wireless sensor networks were introduced as a viable and energy efficient means by which to instrument large-scale mountain basins. The data obtained by these networks will lay ground for a suite of improved hydrologic models that can be coupled to permit for the automated, real-time optimization and control of existing water infrastructure.

!"

#!!!!!!"

$!!!!!!"

%!!!!!!"

&!!!!!!"

'!!!!!!!"

'#!!!!!!"

'$!!!!!!"

'%!!!!!!"

'&!!!!!!"

#!!!!!!!"

!"

(!!"

'!!!"

'(!!"

#!!!"

#(!!"

)!!!"

)(!!"

$*'*!+"

$*&*!+"

$*'(*!+"

$*##*!+"

$*#+*!+"

(*%*!+"

(*')*!+"

(*#!*!+"

(*#,*!+"

%*)*!+"

%*'!*!+"

%*',*!+"

%*#$*!+"

,*'*!+"

,*&*!+"

,*'(*!+"

,*##*!+"

,*#+*!+"

&*(*!+"

&*'#*!+"

&*'+*!+"

&*#%*!+"

+*#*!+"

+*+*!+"

+*'%*!+"

-./012"345671/80"

9050/:;4/"<=>;?"St

ream

Dis

char

ge (m

3 /d)

Res

ervo

ir In

flow

(m3 /d

)

REFERENCES

[1] Aspen Environmental Group and M. Cubed, Potential Changes in hydropower production from global climate change in California and the western United States (2005), prepared for the CA Energy Comission.

[2] Bales, R. C., K. A. Dressler, B. Imam, S. R. Fassnacht, and D. Lampkin (2008), Fractional snow cover in the Colorado and Rio Grande basins, 1995–2002, Water Resour. Res., 44, W01425, doi:10.1029/2006WR005377.

[3] Bales R, Molotch N., Painter T., Dettinger M., Rice R, and Dozier J. Mountain hydrology of the western united states (2006), Water Resources Research, 42(W08432), 2006.

[4] Blöschl, G. (1999), Scaling issues in snow hydrology. Hydrological Processes, 13: 2149–2175. doi: 10.1002/(SICI)1099-1085(199910)13:14/15<2149::AID-HYP847>3.0.CO;2-8

[5] California cooperative snow surveys, http://www.water.ca.gov/floodmgmt/hafoo/hb/sss/, last accessed Feb. 2 2011. [6] California water plan (2009), http://www.waterplan.water.ca.gov/cwpu2009/index.cfm, last accessed Feb. 2 2011. [7] Dressler, K. A., S. R. Fassnacht, R. C. Bales, (2006), A Comparison of Snow Telemetry and Snow Course

Measurements in the Colorado River Basin. J. Hydrometeor, 7, 705–712. [8] Erickson, T. A., M. W. Williams, and A. Winstral (2005), Persistence of topographic controls on the spatial

distribution of snow in rugged mountain terrain, Colorado, United States, Water Resour. Res., 41, W04014, doi:10.1029/2003WR002973.

[9] Garen, D. C., 1992: Improved techniques in regression-based stream-flow volume forecasting. J. Water Resour. Plan. Manage., 118, 654–670.

[10] Kerkez B., Glaser S., Dracup J., and Bales R., (2010). A hybrid system model of seasonal snowpack water balance. In Proceedings of the 13th ACM international conference on Hybrid systems: computation and control (HSCC '10).

[11] Kerkez B., Glaser S., Remote Sensing and Detection Strategies of Large Scale Natural Systems Subjected to Extreme Conditions, The Fifth International Workshop on Advanced Smart Materials and Smart Structures technology. Boston, MA, July 2009.

[12] Kerkez B., Rice R., Glaser S.D., Bales R.C., Saksa P.C., Design and development of a wireless sensor network to monitor snow depth in multiple catchments in the American River basin, California: hardware selection and sensor placement techniques. (2010). Abstract IN34A-07. Fall 2010 meeting American Geophysical Union, San Francisco.

[13] Kerkez, B., Glaser, S.D., and Bales, R., (2011a). Monitoring Hydrologic Phenomena: Wireless Sensor Networks, in review.

[14] Kerkez, B., Glaser, S.D., and Bales, R., (2011b). Monitoring Hydrologic Phenomena: Sensor Placement and Sampling Design, in review.

[15] Kerkez. B., Meadows M., Glaser S. D., Bales R. C., The Science of Wireless Sensor Networks: Improving engineered systems through scientific analysis (2009). Fall 2009 meeting American Geophysical Union, San Francisco.

[16] Labadie John, Optimal Operation of Multireservoir Systems: State-of-the-Art Review, J. Water Resour. Plng. and Mgmt. 130, 93 (2004), DOI:10.1061/(ASCE)0733-9496(2004)130:2(93)

[17] Marks D., Cooley K., Robertson D., and Winstral A., Long-term snow database, reynolds creekxperimental watershed, idaho, usa. (2001), Water Resources Research, 37:2835{2838, 2001}

[18] Meadows M.W., Kerkez B., Hartsough P.C., Lucas R.G., Bales R.C., Hopmans J.W., Glaser S.D., Comparing plot-scale sensor measurements to the watershed level: a comprehensive case study of snow depth and soil moisture in the southern Sierra Nevada, California. (2010). Abstract C13C-03. Fall 2010 meeting American Geophysical Union, San Francisco.

[19] Molotch, N. P., and R. C. Bales (2005), Scaling snow observations from the point to the grid element: Implications for observation network design, Water Resour. Res., 41, W11421, doi:10.1029/2005WR004229.

[20] Molotch, N. P., Colee, M. T., Bales, R. C. and Dozier, J. (2005), Estimating the spatial distribution of snow water equivalent in an alpine basin using binary regression tree models: the impact of digital elevation data and independent variable selection. Hydrological Processes, 19: 1459–1479. doi: 10.1002/hyp.5586

[21] Pagano, Thomas, David Garen, Soroosh Sorooshian, 2004: Evaluation of Official Western U.S. Seasonal Water Supply Outlooks, 1922–2002. J. Hydrometeor, 5, 896–909.

[22] Rice, R., and R. C. Bales (2010), Embedded-sensor network design for snow cover measurements around snow pillow and snow course sites in the Sierra Nevada of California, Water Resour. Res., 46, W03537, doi:10.1029/2008WR007318.

[23] SNOTEL data and products, http://www.wcc.nrcs.usda.gov/snow/, last accessed February 2 2011. [24] DUST Networks: http://dsutnetworks.com. Last accessed February 2, 2011.

[25] DUST Networks ,Technical Overview of Time Synchronized Mesh Protocol (2006), Whitepaper. [26] IEEE Standard for Information technology -- Telecommunications and information exchange between systems --

Local and metropolitan area networks -- Specific requirements. Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs), IEEE Std., Rev. 2006, 8 September 2006.

[27] Western Education Foundation, A briefing on California Water Issues (2009), whitepaper, http://www.watereducation.org/, last accessed February 2, 2011.

[28] Sierra Club, Water issues in California, http://www.sierraclub.org/ca/water/index.asp, last accessed February 2, 2011.

[29] Association of California Water Agencies, California Water (2011), http://www.calwatercrisis.org/program.htm, last accessed February 2, 2011.