Technical information, news, research, and opinion on avalanches, snow safety, and winter backcountry travel.

Wednesday, November 24, 2010

Living In The Moment

It was a long and dark December, from the rooftops I remember there was snowColdplay

NOTE: This is the second post that will address the general question of Why Is It So Complicated? This time we're going to talk about cold weather, avalanche forecasting, and persistent weak layers. Instead of engaging in endless speculation over the state of the winter snowpack, I'd like to take this opportunity to discuss the basic elements of forecasting, and what you can and cannot accomplish with forecasting.

Seeing The Future
Recently there have been a few online discussions about snow and weather conditions in the Pacific Northwest. With the current cold weather, and a generally colder winter forecast, a lot of people are wondering if persistent weaknesses will plague the snowpack this winter.

The straight answer to the above question is as follows: no one really knows what will happen in the Cascades this winter. What we know right now is that the current cold temperatures are almost certainly producing instability wherever the snowpack is shallow, and there is a high likelihood that the cold temperatures are producing surface instability in areas where the snowpack is relatively deep. It is very likely that faceting is widespread during clear, cold nights when the snow loses heat through long wave radiation loss. On solar aspects at high elevations, you might find radiation recrystallisation.

Armed with the knowledge that instability is developing in the snowpack, we can start to speculate about what might happen next.

In some areas, the instability may persist through the next couple of storm cycles, whereas in other areas, the first big avalanche cycle will clean things up. There's also the spectre of a pineapple express, which would produce widespread instability in many places, but not everywhere, while simultaneously producing a fantastic layer of "glue" to heal surface instabilities. On the other hand, a strong rain event could melt the facets and turn the snowpack into a layer of bulletproof concrete.

Of course, in the event any faceted snow is buried to a depth of about 1 metre, the existence of moderate temperatures would allow rounding to prevail in the snowpack...effectively healing instability in areas with deep snow cover. It could take a week in some areas, a couple of weeks in other areas, and in some areas the problem could indeed persist for the entire winter.

When you get down to it, just about anything could happen at this point.

Why Is It So Complicated?
In simple terms, it's complicated because no one knows the future interactions between terrain, snowpack, and weather. Therefore, no one knows whether or not persistent weaknesses will develop. As I often write on this blog, the chaotic interaction between terrain, snowpack, and weather is responsible for much of the uncertainty in backcountry avalanche forecasting. Rich Marriot writes, how can you forecast avalanches if you can't forecast the weather. The short answer is that you definitely can forecast avalanches, you just can't forecast avalanches very far into the future.

So, if the chaotic interaction between terrain, snowpack, and weather is responsible for a lot of uncertainty, it might be useful to understand why this is the case. To do this, we have to examine the discipline of forecasting. In the context of avalanche forecasting, wanting to know why there is so much uncertainty leads us directly to the following principles:
  • Information Types & Relation to Perception
  • Scales in Space and Time
These are two of the Elements of Applied Avalanche Forecasting discussed in The Avalanche Handbook. We'll discuss them next. ( Please see The Avalanche Handbook for a complete discussion of these elements. It is dangerous to issue avalanche forecasts using these elements by themselves. )

Why Forecasting Is Difficult
Forecasting is concerned with producing an accurate picture of future events. For avalanches, we can consider information types such as Class I data, Class II data, and Class III data. We can also consider the relationship between perception and data from a specific class. To define the scope of the problem, we can consider scales of space and time, or in very simple terms, we want to know where ( space ) and when ( time ).

To issue any type of forecast, you start by gathering data, and then you subject this information to some form of analysis. If you happen to be out in the backcountry on a particular afternoon, you have the tremendous luxury of hindsight made available by knowing the previous weather, and by knowing something about the current mechanical structure of the snowpack. You can also make very specific observations of the environment, including observations of the terrain and current snow deposition patterns.

On our theoretical afternoon, you also know the current weather. So while it is generally more difficult to issue a precise forecast, you also just happen to have access to an incredible amount of information on which to base any such forecast. To make things even easier, you're only concerned with issuing a forecast for a very short time, and for a very limited number of places. On the other hand, long-range forecasts are based on theoretical weather data ( Class III ), and there is always high uncertainty associated with such data. High uncertainty means that you're much more likely to make errors and blow the forecast as result.

For this reason, forecasts for the immediate time frame and for a small geographic area, often called nowcasts, are the most accurate. A forecast for a few days ahead is less accurate, and a long-range forecast might not be accurate in any sense. Nowcasts are more accurate because uncertainty is lower when we actually know something about the variables affecting the current situation. Unfortunately, we usually can't know the variables that will create or affect situations in the future.

This means that for any date far enough in the future, for a large area, uncertainty is essentially unlimited. If you want to see how this works, issue a forecast for your life over the next minute. What about a forecast for the next hour. What about the next 24 hours. What about next week? What about next month? Three months from now? Three years from now?

Bayesian Logic
The dynamic, integative process humans that use to conduct avalanche forecasting can be referred to as a Bayesian activity. We say that this process is dynamic, because it is active, and because it changes as information is collected. The process itself is integrative because you must consider the evidence as a whole, rather than as separate pieces. This is reflected by the following formula:
  • Prior × Likelihood = Posterior ( or, what follows )
If we convert that formula to human-compatible terms, we get the following:
  • The combination of Past Conditions and Current Conditions = Forecast
If you want to issue a forecast for February 14th, 2011, you'll notice a rather glaring lack of information about conditions leading up to that date. The reason why is obvious: you have no information about terrain, weather, and snowpack for that date because it hasn't happened yet.

The most important characteristics of Bayesian revision is the ability for a single piece of data to change the forecast. That means, you throw out the old forecast as you acquire additional data. If you observe unstable snow, your forecast must change.

Try It Out
Here are some exercises:
  • Forecast snowpack instability for the Cascade Mountains during winter 2010-2011. ( You can also choose your home mountain range if you don't live in the Cascades. )
  • Forecast snowpack instability for Phantom Trees backcountry ski run on November 30th, 2010 at 2:30pm. ( You can also choose your own favourite backcountry ski run. )
  • When you're finished with your forecasts, write a short snippet about which forecast is more precise and why it is more precise.
  • Remember, it's easy to confuse accuracy with precision, but they are not the same thing at all.
Conclusion
With respect to the Cascades, it seems pretty likely that another big dump will produce significant instability. But such general forecasts are easy to issue because it's well known that large dumps of snow produce significant snowpack instability. When we consider the uncertainty of long range forecasting, it's very easy to see why it's all but impossible to say anything about an entire winter.

As usual, we'll just have to wait and see.

But remember, theoretical avalanches aren't dangerous. It's the real avalanches that you have to watch out for, and real avalanches always happen at a very specific place and time.

Addendum
Regardless of the specific place and time in which you find yourself, Happy Holidays to all my readers.

Wednesday, November 17, 2010

There Are No Magic Bullets

Chances are, when said and done, Who'll be the lucky ones, Who make it all the wayFive For Fighting

A complete backcountry safety system uses a mix of elements, including thorough planning, safe travel habits, rescue gear, and good judgment. Using multiple risk management elements allows you to reduce risk in a variety of different places, which is the same as not putting all your eggs in one basket. ( Putting all your eggs in one basket is referred to as risk concentration. )

To this point, many recreational skiers go lite on the trip planning, and by doing so they miss out on important opportunities to reduce risk. Then, perhaps due to poor planning, or lack of skill, the party makes a few poor decisions, which again represent missed opportunities to reduce risk.

As opportunities to manage/reduce risk are discarded, more pressure is put onto the rescue gear component of your backcountry safety system. Unfortunately, the rescue gear component of your backcountry safety system is really only designed to give you a chance at live recovery in the event of a complete burial. Rescue gear is not a comprehensive backcountry safety system.

And it's certainly not a magic bullet.

If you travel somewhere frequently, you may be tempted to avoid planning. But remember, even though the terrain remains the same, both environmental conditions and humans are subject to frequent changes. For this reason, it's a good idea to have a set of stock trip plans that you can pull out and review from the perspective of current conditions.

Human conditions: Are you tired? Maybe a bit hungover? How's your hydration and calorie intake? Are you really jonesing for a fix? What about your friends? Environmental conditions: Is avalanche danger high? What do you think the snowpack is doing on your intended route? What does the public avalanche bulletin have to say?

Conclusion
Three major elements of a complete backcountry safety system:
  • Planning: Thorough pre-trip analysis of terrain, snowpack, weather, and people involved.
  • Traveling: Safe travel habits, avalanche forecasting, managing yourself, and good judgment.
  • Rescue: Beacon, shovel, probe, spotting, searching, extraction.

Wednesday, November 10, 2010

Dust In The Wind

All we are is dust in the windKansas

NOTE: This is the first in a series of posts that will address the general question of Why Is It So Complicated? While teaching often involves simplification, it's important to remember that you can also use complexity to teach. Despite conventional wisdom, complexity is not always the enemy of simple, and simplicity does not always improve understanding.

Introduction
Today's post is going to discuss wind loading, more specifically the ins-and-outs of using wind speed and direction to forecast wind loading. The main difficulty in using a simple model, is that the simple model doesn't account for turbulence, and turbulence has an incredible influence on the "patterns" of wind deposited snow.

Video 1. You can't see clear air turbulence, but watch this time-lapse video. The presence of clouds makes the turbulence very easy to see. At about 1 minute, you can watch some backwards loading, where turbulent vortices load a slope that is facing against the wind. This is referred to by the very technical term wind slab where you least expect it.


The public avalanche bulletin often contains a forecast of aspects on which you might expect to find wind loading, but it's important to remember that a.) the public avalanche bulletin covers a very large area, b.) avalanche forecasters have a lot of knowledge about the interaction between terrain and weather, as well as a deep body of experience about a variety of such situations in the past, and c.) they also have access to very sophisticated computer models. Forecasters with years of experience at a specific location will also develop a very good sense of how a specific storm will influence loading in certain areas.

Since most of us normal folks don't have that experience, we need to stick with the tried and true.

Character Of The Data
Wind speed and direction is Class III information, which means there is high uncertainty about its relationship to avalanche formation. High uncertainty exists because there are many ways to interpret the data, and as a result, this data may or may not reveal useful information about instability. This is because the physics behind wind flow are intensely complicated and the specific outcomes are very possibly unknowable.

Prevailing Wind
When we talk about wind, there are two important elements: the prevailing wind and local winds. The prevailing wind is located high above rough mountain terrain, where airflow is unimpeded by obstacles. We can express the prevailing wind with a direction such as north or northwest, and with a speed such as 20 knots. Broadly, the prevailing wind is influenced by horizontal pressure differences and the rotation of air around pressure centres.

An additional complication of using prevailing wind is that the varying shape and behaviour of cyclonic weather systems means that the prevailing wind speed and direction can change during the passage of a storm, and it can be difficult to predict these changes because of the dynamic nature of weather systems. The reality is that each storm is a completely unique event that is driven by a set of completely unique parameters that will never happen again. Once you start to understand the complexity, it becomes easier to understand how to use this complexity to your advantage during the process of backcountry avalanche forecasting.

As usual, this complexity provides very useful clues about what not to do.

Local Wind
Anyway, things become awfully complicated when a large storm moves over the mountains and the air starts interacting with terrain. This is referred to as local wind. Local wind is formed from a complex stew of frontal lifting, orographic lifting, convective lifting, and convergence-based lifting, in addition to frictional forces.

Picture a mountain valley, along with all its nooks, crannies, and crevices. As air moves through complex mountain terrain, it hits obstacles that cause it to become turbulent. Large features block and channel the large scale flow, while airflow over, around, and through smaller barriers generates localised turbulence. On this blog I frequently refer to the chaotic interaction between terrain and weather, and these crazy wind patterns are a very large part of the chaos. Naturally, the mere presence of chaos raises uncertainty because chaos suggests that uncertainty is the only state of affairs.

Clouds, along with other sophisticated meteorological data, can provide a fairly good visual model of this turbulence at a large scale, but these data are almost no use at the small scales required for effective backcountry avalanche forecasting. At small scales, we can use weather stations to measure local wind, but because of resource limitations, our picture of local wind speed and direction is actually extremely limited.

Thankfully, despite these problems, there are a few really good techniques that we can use to determine if wind loading has occurred. In simple terms, even though the wind is invisible, and even though the chaos of turbulence makes it very hard to determine local wind flow, we can observe the physical environment for the signals of wind loading.

Examples
What do we know about the complexity of wind speed and direction at a single location? What about the chaotic complexity of turbulence?

Figure 1.1. Wind speed and direction at 1 hour increments for Crystal Mountain "Green Valley" weather station over the course of a single day. Since the weather station measures wind on an hourly basis, there is a lot of missing data here.


Figure 1.2. Wind speed and direction at 1 hour increments for Crystal Mountain "Green Valley" weather station over the course of a single day. Here, the pattern matching software in our brains sees two broad patterns that may be associated with a major wind direction change that occurs when a cyclonic weather system passes over the weather station. Or, maybe our brains are seeing a pattern where no pattern exists.


Figure 1.3. Wind speed and direction at 1 hour increments for Crystal Mountain "Green Valley" weather station over the course of a single day. Here our brain is tempted to view "clusters" of wind directions, but even if these "clusters" do occur, our brain is simply incapable of accounting for turbulence, and these "patterns" may very well arise from turbulence and contribute to additional turbulence. It's pretty crazy, isn't it?


Figure 1.4. Complex deposition and drift patterns near the toe of the Sulphide Glacier. Can you identify the wind slab? Does the wind slab have a simple shape? Is the layering simple or complicated?
Figure 1.5. How complicated is turbulence? I'll engage in a gross oversimplification, but you can clearly see the vortex generation that is one of the hallmarks of turbulent fluids such as air. In the mountains, this effect produces cornices in some situations, and in general it produces loading patterns that are incredibly intricate. These loading patterns produce wind slabs of varying size and depth. The depth of a wind slab often varies greatly across a very small area, which further raises uncertainty about the likelihood of avalanche formation.


Figure 1.6. From MIT. The difference between laminar and turbulent flow. Imagine this in three dimensions, and then imagine trying to relate this model to snow deposition patterns. By now it's pretty obvious that you can't actually do this. It's worth mentioning that a supercomputer with 1024 cores might be able to produce a relatively accurate simulation. However, small inaccuracies in the input data, including inaccuracies that are immeasurably small, and/or inaccuracies in the mathematics of the simulation, will produce severe errors at every scale. This means that you'll end up with a simulation that "looks accurate", but actually has no relationship to the real world.

Conclusion
So what's a guy or gal to do? Well, given the complexity and uncertainty that arises, observations of local weather and snow deposition patterns remain the gold standard for discerning the parameters of wind loading. You can safely assume that any combination of wind and snow means that some wind loading has occurred.
  • Significant amounts of snow are removed from windward slopes.
  • Significant amounts of snow can be deposited on lee slopes.
  • Drifting patterns. Drifts point in the same direction as airflow.
  • Snow build up, or the lack thereof, on trees and rocks.
  • Cornices point in the direction of airflow and indicate loading below.
  • Look for ripple patterns on the snow surface.
  • Observe the snow's texture, especially its hardness.
  • Wind slab often sounds hollow when you step on it.
  • Wind slab has an intricate shape and layering.
  • Complex stratigraphy raises uncertainty.
  • Delicate and/or intricate transitions between clean and dirty snow may exist.
  • Given the foregoing, never try to outsmart the snowpack. It simply can't be done.
Obviously, there's a theme: stick with tried and true methods of using visual observations to assess wind loading. These methods are far better than trying to model wind loading by using the prevailing wind direction.

Futher reading that merely hints at the unreal complexity of the problem:

Friday, November 5, 2010

Great Learning Resources

We Don't Need No EducationPink Floyd

Here are two absolutely wonderful resources. Free registration required.