Cut On the Bias

When it comes to decisions, avalanche education often centres itself around biases. ( Referred to as human factors. ) Here is some very high-end thinking on biases from Oxford University.

Feeling Rational: http://lesswrong.com/lw/hp/feeling_rational/
The Role of Bias in Bias: http://lesswrong.com/lw/he/knowing_about_biases_can_hurt_people/
The Affect Heuristic: http://lesswrong.com/lw/lg/the_affect_heuristic/

Might make your brain hurt a bit, but that's a good thing, right?

Q & A

Answers to questions sent to me via email.

Why is buried surface hoar so dangerous?
Buried surface hoar is dangerous because surface hoar crystals form a thin layer that contains a lot of air. This means that the layer on top of the surface hoar is mostly supported by air. When crushed, surface hoar crystals have the capacity to rearrange themselves into a much smaller space, which causes the layer above to fall. As the layer falls, it provides energy that causes the crushing to spread. After the crushing process is complete, you're left with two layers of snow that have no attachment to each other-delamination has occurred. Finally, gravity, which is always in effect, pulls the detached layer downhill.

All the persistent forms ( facets, depth hoar, surface hoar, and combinations of these with crusts ) are dangerous for this reason. Crusts, when found alone, are dangerous for slightly different reasons, mostly related to poor bonding at the interface between the crust and the layer above-poor bonding increases the risk of catastrophic delamination, and catastrophic delamination is required for avalanche formation. The answer provided here intentionally does not discuss weak layer parameters such as anisotropy or variations in grain morphology, which are more related to why the layers are persistent than to why they are dangerous.

What is the significance of snow crystal size?
There are two key factors related to crystal size: the first factor is that large crystals have a lower number of bonds per unit volume. This is part of the reason why the persistent forms, such as facets and surface hoar, are so weak. Networks of large crystals can be quite strong, but they are almost always relatively weaker than networks of smaller crystals.

The second factor is grain size mismatch between layers. Grain size mismatches result in weaker bonding between layers, because the smaller grains overlay the pore space between the larger grains. This means a lot of the smaller grains are simply "touching the air". Furthermore, such configurations tend to concentrate stress at the interface between the layers.

What wind speed produces wind slab?
This depends almost entirely on the condition of surface snow. It takes little more than a stiff breeze to move dry, loosely packed snow. On the other hand, hurricane force winds will have little effect on dense, frozen corn snow.

What is the source of uncertainty in persistent weak layers?
With respect to persistent weak layers, uncertainty arises from a few key factors:

• Where is the weak layer?
• What is its depth below the surface?
• How weak is the layer and surrounding interfaces?

Picture a mountain valley. You know that there is buried surface hoar in some places. You also know that its depth below the surface varies. You also know that its degree of weakness varies by location.

Now that we've discussed what you know, think about what you don't know.

Snowpack Synopsis for British Columbia Interior Ranges

A great read from James Floyer.

http://www.avalanche.ca/cac/library/researchandarticles/2011-01-06_Snowpack_Synopsis

Enjoy!

Forecasting 101

if I knew then, what I know now—from the vernacular

There are lots of options for improving your avalanche forecasting skills, but what happens after you've taken an avalanche safety training course? The idea is to get out and gain experience, but the relationship between experience and skill is tenuous. Yes, you need to get out and ski tour, but you also need to focus on increasing your skill, so that all your hard-earned experience doesn't go to waste.

I get a few emails each month in which people ask me how to break past a learning plateau, and my answer is, increasingly, to engage in some self-directed study. Of course, the question then becomes what should I study?

When I think about all things I've learned about avalanches, several things really stick out as useful. Snow metamorphism is one of the most useful topical areas, and of course, learning the best practises of risk management is also really useful. You should also study avalanche formation because the information you learn is incredibly useful when interpreting snowpack test results. Of course, all this leads in the general direction of improving your backcountry avalanche forecasting skills, and there is one resource that is utterly peerless in this regard.

Chapter six of The Avalanche Handbook lays out a theoretical framework for avalanche forecasting that I've found very useful. The elements of applied avalanche forecasting are as follows:

• Definition. Forecast snowpack instability across space and time.
• Goal. Align your perception of instability with reality.
• Information Types & Relation To Perception. The relationship between instability and Class III, II, and I factors.
• Scales in Space and Time. Considering avalanche forecasting in terms of spatial and temporal parameters.
• Human Factors and Perception. Managing your state of mind and remaining objective.
• Reasoning Processes. Learn how to apply inductive and deductive reasoning to the problem.
• Decision-Making. How to manage risk.

When I first read this chapter, my mind certainly did a few backflips and double-takes. How do you take something this complex and apply it in the real-world? After a few more reads, and after some directed study of the material, including extensive Q & A, I finally was able to wrap my brain around the concepts.

I'm prepared to make a fairly steep claim in this post, and it is as follows: there is no better way to learn how to forecast avalanches than to study the material in this chapter until you know it forwards and backwards. Naturally, I've prepared an exam that you can use to gauge your progression.

Elements of Avalanche Forecasting Exam

124 questions from chapter 6 of The Avalanche Handbook. As with my other exams: Yes, I am fully aware that this exam is ridiculous, but it was a lot of work, so have fun.

Capsule

This chapter discusses avalanche forecasting. The seven elements of avalanche forecasting are discussed, as well as the basis for each. The elements of forecasting are linked through risk concepts ( especially perception / perceptual errors related to data sampling ) to form a framework suitable for learning basic forecasting. Specific forecasting techniques and procedures are not included because, given the multitude of forecasting contexts and the dynamic, evolutionary nature of the forecasting process, there is not really a specific process used to issue a forecast. At the end of this chapter you should understand the basis for avalanche forecasting, data types, human perception, and the links between these forecast components and risk concepts such as error, probability, and decision-making.

List Of Sections

1. Forecasting And Avalanche Forecasting.
   This section contains a detailed overview of forecasting, avalanche forecasting,
   and including relevant details on methods and pitfalls.
2. The Seven Elements Of Avalanche Forecasting.
   This section contains in-depth analysis of each element of avalanche forecasting.
3. Common Biases And Decision Traps In Avalanche Forecasting.
   This section contains an in-depth explanation of biases and decision traps, along
   with techniques to neutralize these biases and avoid decision traps.

Forecasting And Avalanche Forecasting

1. Around what element is modern avalanche forecasting framed? From what perspective are forecasts issued?
2. List all seven elements of applied avalanche forecasting.
3. Avalanche forecasting is a ________ problem.
4. All seven elements are ________.
5. Most avalanche accidents occur as a result of human errors. True or False.
6. Provide the definition of forecasting.
7. Define the root cause of most avalanche accidents.
8. How is avalanche forecasting linked to risk analysis?
9. Is avalanche forecasting limited only to estimates of instability? Explain why or why not.

I. Definition Of Avalanche Forecasting

1. Define avalanche forecasting.
2. Define the major physical uncertainty with respect to avalanche forecasting.
3. Avalanche forecasting is defined in terms of ________.
4. Whereas traditionally, avalanche forecasting was defined in terms of ________.
5. In avalanche forecasting, what type of information is most highly prized?
6. To what does triggering level refer?
7. Provide three examples of types of forecasting relative to triggers.
8. How do most slab avalanches release?
9. Some avalanches release without the need for an external trigger. True or False.
10. If true, explain. If false, explain.
11. Upon what does the energy required to release a slab avalanche depend?
12. What is the primary reason avalanche forecasting is probabilistic, with a risk-based character?

II. Goal Of Avalanche Forecasting

1. Define the goal of avalanche forecasting. Discuss the primary sources of uncertainty.
2. State the goal of avalanche forecasting from the human perspective.
3. How is this goal accomplished?
4. Define relevant information in the context of avalanche forecasting.
5. There is a strong link between quantity of information and accuracy of decisions. True or False.
6. There is a strong link between confidence in a decision and the resulting accuracy. True or False.
7. Briefly discuss the role of redundant information in statistical predictions.
8. List and describe each data classification.
9. Discuss ensemble forecasts.

III. Human Factors And Perception

1. Discuss the scale of failure in human perception with respect to avalanches.
2. Define perception.
3. Discuss the two general components of human influences.
4. Connect risk-taking propensity to perception.
5. What relationships does the Risk-Decision Matrix display?
6. Define Operational Risk Band [ ORB ].
7. What is the upper boundary of the ORB?
8. What is the lower boundary of the ORB?
9. Provide a list of Type I errors.
10. Provide a list of Type II errors.
11. Define target risk.
12. What are the consequences of Type I errors.
13. What are the consequences of Type II errors.
14. What is the relationship between uncertainty and perception?
15. Who coined the term risk homeostasis?
16. Explain risk homeostasis and provide an example.
17. List two or three items that improve perception.
18. List two or three items that degrade perception.
19. When might biases have a small effect on perception of instability?
20. When might biases have a large effect on perception of instability?
21. Discuss absolute instability relative to perception.
22. Discuss conditional instability relative to perception.
23. Write a brief explanation of the implications of perception of instability and the public danger scale.
24. Draw the continuum of instability and describe perception at three points.
25. Why is the link between data sampling and perception so important?
26. What does White ( 1974 ) argue about perception of hazard?
27. What is shown by statistics that compare fatalities to the public danger scale?
28. Why is perception better for instability in new snow?
29. Is randomness desired in the sampling process for avalanche forecasting?
30. Why are slopeside stability tests sometimes compared to playing the lottery?

IV. Reasoning Processes

1. Define the two main types of reasoning used in avalanche forecasting.
2. Provide an example of each type of reasoning.
3. Snow stability is dynamic and evolutionary, therefore avalanche forecasting is ________ and evolutionary.
4. What underpins the dynamic nature of avalanche forecasting?
5. The dynamic process of ________ ________ about instability using ________ ________ is somewhat analogous to ________ ________ using ________ ________ as ________ ________.
6. For a given avalanche path, when does the ideal forecast period begin?
7. Explain the answer to the previous question, in the context of forecast revision.
8. Define Bayes Theorem.
9. In avalanche forecasting, relative to Bayes Theorem, what constitutes the prior?
10. In avalanche forecasting, relative to Bayes Theorem, what constitutes the likelihood?
11. In avalanche forecasting, relative to Bayes Theorem, what constitutes the posterior?
12. Discuss the relationship between inductive reasoning, datums, and forecast revisions relative to instability.
13. How does an avalanche atlas fit into the context of avalanche forecasting as a Bayesian process?
14. In the context of avalanche forecasting, instability is not highly time-dependent. True or False.
15. If this is false, explain why.
16. List three examples of the deductive elements of avalanche forecasting.
17. Compare the evolutionary character of reasoning for helicopter backcountry skiing and ordinary backcountry skiing.

V. Information Types And Relation To Perception

1. Information for avalanche forecasting consists of two types. Explain each and include examples.
2. Should one always have an opinion about instability before attempting risky activities in avalanche terrain?
3. Describe the correct course of action if information relevant to the case at hand is missing at the beginning of a forecast period.
4. Describe a method to implement the correct course of action for the previous question.
5. If avalanche forecasting is Bayesian, what information type constitutes the likelihood?
6. If avalanche forecasting is Bayesian, what information type constitutes the prior?
7. Can data derived from computer models constitute the prior?
8. Define informational entropy.
9. Why is wind speed and direct data harder to interpret than cracking of the snow cover?
10. Define highly correlated.
11. What is necessary for dealing with highly correlated data?
12. Does the class of information ( I, II, III ) always give the most priority to Class I information?
13. Describe the theory of weighting data.

VI. Scales In Space And Time

1. Define spatial scale relative to avalanche forecasting.
2. Define temporal scale relative to avalanche forecasting.
3. Define scale-matching.
4. Provide an example of what might happen if scale-matching is not performed.
5. Explain the three primary spatial scales.
6. Difficulty of forecasting is inversely proportional to scale. True or False.
7. If true, explain. If false, explain.
8. Does failure to perform scale-matching result in many needless accidents?
9. Define now-cast.
10. Which is more difficult: forecasting stability for next Wednesday or next Thursday?
11. Why does chaos influence avalanche forecasting?

VI. Decision-Making

1. Provide a basic list of steps used to issue an avalanche forecast.
2. Why formalize the decision-making process?
3. Relative to avalanche forecasting, what underpins the fundamental residual risk with all decisions?

Conclusions

1. Why is a chain of events difficult to construct for avalanche forecasting?
2. Where do terrain and snow climate fit into the avalanche forecasting?

Biases

1. Discuss "search for supportive evidence" and how to neutralize this bias.
2. Discuss "inconsistency" and how to neutralize this bias.
3. Discuss "conservatism" and how to neutralize this bias.
4. Discuss "recency" and how to neutralize this bias.
5. Discuss "frequency" and how to neutralize this bias.
6. Discuss "availability" and how to neutralize this bias.
7. Discuss "illusory correlations" and how to neutralize this bias.
8. Discuss "selective perception" and how to neutralize this bias.
9. Discuss "expert halo" and how to neutralize this bias.
10. Discuss "underestimating uncertainty" and how to neutralize this bias.
11. Discuss "excessive optimism" and how to neutralize this bias.
12. Discuss "anchoring" and how to neutralize this bias.
13. Discuss "rules of thumb" and how to neutralize this bias.
14. Discuss "guide-client relationship" and how to neutralize this bias.
15. Discuss "social proof" and how to neutralize this bias.

Chapter 6 Exam Answers

245 answers to questions from chapter 6 of The Avalanche Handbook.

Forecasting And Avalanche Forecasting

1. Modern avalanche forecasting framed around instability and conducted from the perspective of the trigger.
2. The seven elements of applied avalanche forecasting are as follows:
   
   • Definition
   • Goal
   • Human Factors And Perception
   • Reasoning Processes
   • Information Types & Relation To Perception
   • Scales Of Space And Time
   • Decision-Making
3. Avalanche forecasting is a dynamic problem. Forecasting avalanches
   means dealing with uncertainty in the form of spatial and temporal variability
   in the seasonal snowpack, including incremental in snow and weather conditions.
4. All seven elements are interconnected.
5. True. Most victims trigger the avalanche themselves. This suggests that the
   perception ( i.e. "the snow is stable" ) did not match reality at the time of the
   accident.
6. Forecasting is the prediction of current and future events.
7. Errors in human perception are at the root of most avalanche accidents.
8. The link between avalanche forecasting and risk analysis is formed when
   decision-making follows the prediction issued by an avalanche forecast. Since
   these decisions involve a chance of losses, the process of avalanche forecasting
   and the resulting decision-making is the equivalent of a risk analysis.
9. Avalanche forecasting is not limited to estimates of instability. There is
   a connection between avalanche forecasting, decision-making, and the inherent risk
   of those decisions.

I. Definition Of Avalanche Forecasting

1. Avalanche forecasting is the prediction across space and time of
   current and future snowpack instability relative to a specific triggering level.
2. The spatial and temporal variability of the seasonal snow cover is the
   principle physical uncertainty in avalanche forecasting.
3. Avalanche forecasting is defined in terms of instability.
4. Whereas traditionally, avalanche forecasting was defined in terms of stability.
5. Information that reveals instability.
6. Triggering level refers to the amount of energy required to release an avalanche.
7. Forecasting for natural releases, forecasting for skier triggering, forecasting for explosive triggering.
8. Most slab avalanches release from overloading by precipitation or wind.
9. True.
10. Sometimes slabs release due to temperature change. Temperature change ( usually an increase )
    results in slab motion which leads to deformation. If there are pre-existing weaknesses such
    as slip surfaces or shear bands, the additional deformation produces the propagating shear
    fractures required to release an avalanche.
11. The energy required to release a slab depends largely on the size of imperfections and
    the parameters of the load applied at any given time. In this case, parameters of load applied means
    intensity, which is expressed roughly by the amount of force and the rate at
    which the force is applied ( the balance between shear stress intensity
    and shear fracture toughness in the weak layer ).
12. At all times, but especially during times of conditional instability ( the
    prevailing state ), the size, state, quantity, and distribution of weaknesses and
    imperfections ( such as weak zones and weak interfaces ), and the energy required to
    trigger a slab release on any such weakness, are unknown. Therefore avalanche forecasting
    can be reduced to encounter probability and trigger probability, i.e.
    "what is the chance of encountering a critical imperfection and how much energy
    will it take to trigger an avalanche". These probabilities give avalanche forecasting
    its risk-based character.

II. Goal Of Avalanche Forecasting

1. The goal of avalanche forecasting is the reduction of uncertainty introduced
   by three key sources:
   
   • Temporal and spatial variability of the snow cover, including terrain influences.
   • Incremental changes to the snowpack from snow and weather conditions.
   • Human factors, especially variations in perception.
2. From the human perspective, avalanche forecasting seeks to align perception
   and reality, i.e. human perception of instability across the spatial and temporal
   scales should match reality as closely as possible.
3. Aligning human perception with reality is accomplished by performing objective
   analysis on data relevant to the case at hand ( using the scientific method ).
4. In the context of avalanche forecasting, relevant information is any information
   that reveals instability.
5. False.
6. False.
7. Redundant information degrades the accuracy of predictions.
8. Data classifications are as follows:
   
   • Class I. Instability Factors. Non-numerical, mostly observable phenomena.
     Includes avalanche occurrences, shear quality, instability tests, cracking of the
     snow cover, and whumpfing.
   • Class II. Snowpack Factors. Rule-based. Snow stratigraphy, snow temperature,
     grain size and type, snow density.
   • Class III. Snow and Weather ( meteorological ) Factors. Precipitation, wind,
     temperature, radiation, weather forecast.
9. An ensemble forecast is the average of several predictions. It is thought that
   ensemble averages are more accurate. These forecasts are often used as a hedge against
   chaos. Ensemble forecasts have improved weather forecasting.

III. Human Factors And Perception

1. Failures in human perception with respect to avalanches run from the
   level of individual to the level of government and
   society. It is possible for a single individual to experience a
   serious perception failure and trigger an avalanche while skiing. On the
   other end of the scale, it is possible for an entire society to experience
   a serious perception failure and fail to allocate sufficient resources to
   avalanche forecasting, hazard mapping, and zoning.
2. Perception is a view of reality based on information processing by
   the senses.
3. Human influences on perception are roughly divided into the following categories:
   
   • Basic personality traits and behaviour ( risk propensity ).
   • Individual perception and its effect on decision-making.
4. The relationship between risk propensity and decision-making is highly
   complex, but in general risk-taking propensity is governed by the total
   sum of one's life experiences.
5. The Risk-Decision matrix displays the relationship between
   risk propensity, perception, and decision-making.
6. The operational risk band is a framework defined
   by the upper and lower limits of risk. To avoid errors that
   result in either accidents or excessive conservatism, the results
   of all decisions should fall inside the operational risk band. This
   is an important component of formalized decision-making.
7. The upper limit of the ORB is a Type I error, usually resulting in an accident.
8. The lower limit of the ORB is a Type II error, usually resulting in lost opportunity.
9. Reluctance to claim the snowpack is unstable unless hard proof is at hand.
10. Failure to open an important transportation corridor.
11. Target risk is the maximum risk an individual is willing to accept for
    a given reward. Target risk optimizes the difference between potential gains
    and potential losses. Behaviour modification is the typical method
    by which people seek to achieve target risk. For example, people might
    be willing to take a serious risk for a serious reward but usually are
    unwilling to take a serious risk for a small reward. Achieving target risk
    means that options are weighed based on the difference between risk and the reward
    across the series of options, with the option having the largest difference chosen
    most frequently. ( Relative to the individual and their risk propensity,
    which is of course, a complex subject by itself ). In the bigger picture,
    it is extremely important to understand how one's perception of risk and reward,
    influence decision-making. The ORB is a framework used
    to formalize decision-making with customizable upper and lower limits on risk.
    The upper and lower limits are set by an organization. ( Or an individual although
    most individuals probably do not consciously consider the ORB in their decision making process. )
12. Death, accidents, injuries.
13. Serious financial losses, lost opportunities, bruised egos.
14. Variations in perception increase with uncertainty.
15. Gerald Wilde
16. When safety devices are used, people modify their behaviour to
    maintain the same level of risk as before. When avalanche beacons
    are used, people choose to ski riskier terrain than they would ski
    without an avalanche beacon. Therefore the overall level of risk
    remains the same. The long and the short of this effect is that
    using a safety device will affect your decision-making and this awareness
    is a critical element of objective decision-making.
17. Targeted education and experience improve perception.
18. Biases degrade perception.
19. Biases have a small effect on perception of instability
    when instability is widespread and the triggering energy is low.
20. Biases have a large effect on perception of instability when
    instability is not quite isolated and the triggering level is a
    bit higher than usual.
21. During times of absolute instability, most people, especially
    experienced people, agree that the snowpack is unstable. Therefore
    variations in perception are small.
22. During times of conditional instability ( the prevailing state ),
    people may or may not agree about the quantity or location of instability, nor
    about the required triggering energy. Therefore variations in perception
    are large.
23. Perception of instability relative to the public danger scale clearly
    shows that many fatalities are linked to the Considerable danger level, which
    proves that perception during conditional instability is poorest ( or
    has the largest variations, depending on your perspective ).
24. Diagram omitted.
25. Data sampling is one of the most crucial inputs into any forecast. In fact,
    it is fair to say that data sampling forms the basis of forecasting,
    especially for backcountry travel. Therefore, if the data sampling is subject
    to bias, the forecast is not objective. For example, if a slopeside test
    reveals nothing about instability, it can be easy to conclude that instability
    is not present. However the choice of test location plays a critical role in
    the test results. This is a perfect example of how biased data sampling
    could lead to a disaster.
26. White argues that perception of hazard does not improve with the level of general education, i.e., high school graduates vs. college graduates.
27. Most accidents occur during Moderate or Considerable danger.
28. Storm snow instabilities are found near the surface; this type of
    instability is much easier to find or detect through skiing. Storm
    snow instabilities are also subject to far less perceptual error than
    deep instabilities because biases strongly affect deep instabilities,
    especially when instability persists for a long time. ( i.e. Recency or Frequency. )
29. No
30. The temporal and spatial variability of the snowpack, in addition
    to the danger of accessing real avalanche starting zones, often mean
    that the results of slopeside tests are, for all intents and purposes,
    random or chaotic ( like the lottery ). In addition, data sampling is
    subject to bias ( leading to serious perceptual errors ) that can add an element of
    Russian Roulette. In this case, not only might you "not win" any money,
    you also might suffer serious injury or loss of life.

IV. Reasoning Process

1. The two main types of reasoning used in avalanche forecasting are as follows:
   
   • Inductive. Inductive reasoning is intuitive and integrative; much more
     difficult to characterize than deductive reasoning. ( The inductive
     reasoning process differs from person-to-person. ) Inductive reasoning
     relies on a conclusion to establish a truth.
   • Deductive. Deductive reasoning relies on models, procedures, and
     data to arrive at a result. Deductive reasoning relies on a truth to
     reach a conclusion.
2. An example of each type of reasoning is as follows:
   
   • Inductive. Looking a steep slope that has shed its snow
     after a storm and understanding why the slope is safe to ski.
   • Deductive. Examining weather station data.
3. Snow stability is dynamic and evolutionary, therefore avalanche forecasting is dynamic and evolutionary.
4. Rapid changes to the snowpack, across both space and time, underpin the dynamic nature of avalanche forecasting.
5. The dynamic process of integrating information about instability using inductive reasoning is somewhat analogous to Bayesian revision using updated information as time proceeds.
6. With the first snowfall of the season. However, an avalanche atlas that contains
   historical information about the path is also very useful in assembling probabilities
   used in forecasting.
7. New information can make previous information worthless; since avalanche
   forecasting is dynamic and evolutionary, the process of forecast revision is
   on-going. However all new information, including information that reveals
   instability, must be integrated into the complete seasonal forecast, including
   any historic data available.
8. The definition of Bayes Theorem is as follows:
   
   • Posterior a Likelihood × Prior and proportional to
9. The previous forecast constitutes the prior if avalanche forecasting is viewed
   as a Bayesian process.
10. Singular information relevant to the current situation constitutes the likelihood
    if avalanche forecasting is viewed as a Bayesian process.
11. The new forecast constitutes the posterior if avalanche forecasting is viewed as
    a Bayesian process.
12. One datum can completely change the opinion of the outcome if the datum
    reveals information about instability. This is particularly true if a low-entropy
    ( low uncertainty ) datum reveals instability. For example, forecasting is conducted
    before a ski trip and moment-by-moment during the trip. Even if the current forecast
    is "isolated instability", the appearance of cracks beneath the skis changes the
    entire forecast immediately. In this case the current forecast is revised from "isolated
    instability" to "high instability" regardless of the prior forecast and the posterior
    ( the prediction ) is revised to "avalanche" from "no avalanche".
13. An avalanche atlas constitutes distributional information ( the prior ) in the
    context of avalanche forecasting.
14. False.
15. Instability is highly time-dependent; for example, solar warming
    can increase instability for a few hours during the afternoon. Instability may
    fall almost immediately when the slope falls into shade.
16. Information from models, rules-based systems, and telemetry data.
17. The information database for helicopter skiing is very deep and detailed.
    Information on prior seasons is available as well. For ordinary backcountry skiing,
    the information database is much smaller and relatively little historical information
    is available.

V. Information Types And Relation To Perception

1. Avalanche forecasting relies on the following types of information:
   
   • Singular Information. Information relevant to the current situation and near future.
   • Distributional Information. Information from the past or from similar situations in the past.
2. Yes
3. Go find the information.
4. Go test ski a few slopes, dig a few snow pits, read current telemetry and weather forecasts.
5. Yes
6. Informational entropy refers to the level of uncertainty associated with
   any data. Relative to avalanche forecasting, both cracking in the snow cover
   and natural avalanche releases provide extremely reliable, i.e. low uncertainty,
   information about instability. On the other hand, a report of wind speed and
   direction is indirectly linked to instability. Understanding and linking concepts
   is required to convert high entropy data into low entropy data.
7. Cracking of snow cover is an obvious sign of instability; cracks mean
   that propagating shear fractures are occurring. Wind speed and direction
   is linked indirectly to instability.
8. Highly correlated means there is an indirect relationship between
   two elements in a system. This loosely coupled relationship means that
   a change to one element in the system may or may not result in a change
   to the other element in the system and/or the change may be difficult to
   predict or ascertain.
9. Generally speaking, resolving highly correlated
   data requires a thorough conceptual understanding of the systems and data
   involved in order to create a link between the systems, or to refine the data
   into a format relevant to the case at hand ( e.g., a report of wind speed
   and direction must be linked with a visual observation of wind-loaded snow ).
   This linkage can only formed if the observer understands the concepts of,
   and relationships between, distributional data ( wind speed and direction )
   and singular data ( the case at hand, i.e. the visual observation of wind-loaded
   snow ). Even if wind-loaded snow is observed, it may be far away and still
   irrelevant to the case at hand.
10. Yes. Priority is given to Class I observations because this type of
    data reveals positive information ( highly prized ) about instability. Class II
    data only reveals the potential for instability and Class III data only
    reveals elements which might ( or might not ) contribute to instability.
11. In general, any datum which reveals instability is considered more important
    than any datum that contains little or no information about instability, regardless
    of its class.

VI. Scales In Space And Time

1. Avalanche forecasting operates at three primary spatial scales that
   refer to the geographic area of the forecast: synoptic scale, meso scale,
   and micro scale.
2. Avalanche forecasting operates along the temporal scale, including
   the distant past, recent past, the present, and the near future. Avalanche
   forecasting typically does not operate past the near future because of the
   chaotic nature of the data require ( i.e. the accuracy of long range
   weather forecasts is far from assured ).
3. Scale matching involves resolving the scale of information to the
   scale of the forecast. For example, one should not rely solely on a synoptic
   scale forecast for decision-making at the micro scale. The synoptic forecast
   is important but cannot take precedence over information relevant to the
   current situation. If one observes natural avalanches or cracking in the
   snow cover, one can assume high instability regardless of the information
   contained in the synoptic scale forecast. Fundamentally, scale matching is necessary
   because information found at one scale cannot be simply applied to
   another scale. Seeing cracks in the snow cover at one location in the mountains does not
   mean the snow is unstable for 100 miles in every direction.
   Quantity, rate, and duration of snowfall is another good example. Most
   quantitative precipitation forecasts are issued at the synoptic or meso
   scale. At the micro scale ( very local ) one may find far less or far
   more snow than indicated by a synoptic scale forecast.
4. Despite local signs of instability, a backcountry traveler might rely on
   a synoptic scale forecast and decide to ski an unstable slope. This could
   result in an avalanche. This is a good example of the link between the
   dynamic, evolutionary nature of avalanche forecasting and the use of
   singular and distributional information. The synoptic scale forecast
   constitutes distributional information; local signs of instability constitute
   singular information relevant to the case at hand.
5. The three primary spatial scales are as follows:
   
   • Synoptic. This is the largest scale: 1000 square kilometers.
   • Meso. This is the middle scale: 100 square kilometers.
   • Micro. This is the smallest scale: 1 square kilometer.
6. True.
7. As the scale decreases, difficulty of forecasting increases and the need
   for accurate information relevant to the case at hand increases as well.
8. Yes.
9. A now-cast is a forecast of instability for the present moment.
10. It is more difficult to forecast instability for next Thursday than for next Wednesday.
11. Weather is strongly linked to avalanche formation. The ability to
    successfully forecast avalanches is strongly influence by the essentially
    chaotic nature of weather.

VI. Decision-Making

1. The following is a basic list of steps used to issue an avalanche forecast:
   
   1. Data collection and integration.
   2. Analysis.
   3. Objective decision-making.
2. Formalizing the decision-making process prevents ( or reduces ) bias.
3. The spatial and temporal variability of the snowpack, in conjunction with
   incremental changes due to snow and weather and variations in human perception,
   creates the fundamental residual risk associated with avalanche forecasting.

Conclusions

1. Complex links between the elements of applied avalanche forecast
   make it difficult to conceive avalanche forecasting as a chain of events.
2. Terrain and climate are viewed as distributional information. In general,
   terrain and climate are included, implicitly, in most aspects of avalanche
   forecasting.

Biases

1. The search for supportive evidence is expressed as a willingness
   to gather facts that support the desired conclusion while disregarding
   facts that support an alternate, or undesired, conclusion. To prevent
   this bias, always search for information that reveals instability.
2. Inconsistency is expressed by applying different sets of
   decision-making criteria to similar situations. One might use the
   presence of existing ski tracks to justify the decision to descend
   a steep slope. In this case, the decision is based solely upon the
   existence of ski tracks, when without the presence of ski tracks
   one might use an entirely different set of criteria to evaluate
   instability.
3. Conservatism is expressed by failure to change one's mind when
   new information or evidence becomes available. This can affect
   evaluation of instability in either direction, and of course,
   this is linked directly to decision-making. Keep an open mind
   and use a formalized decision-making process to neutralize this bias.
4. Recency is expressed by allowing events from the most recent
   past to dominate decision-making at the expense of events in the
   less-recent past. Consider the current situation ( singular )
   and past situations ( distributional ) when making-decisions.
   This bias is very important when instability persists for a long
   time.
5. Frequency is expressed by allowing very frequent events to
   dominate decision making at the expense of less-frequent events.
   Consider the current situation ( singular )
   and past situations ( distributional ) when making-decisions.
6. Availability is expressed when decision-making is dominated
   by specific events easily recalled from memory at the expense of
   information relevant to the case at hand.
7. Illusory correlations is expressed a link is "seen" between
   data when no such link exists. Deductive reasoning is strongly
   affected by this bias.
8. Selective perception is expressed by viewing a problem in the
   context of one's own background and experience. Allow everyone
   to have input, especially people with different backgrounds.
9. Expert halo is expressed by allowing one person's expertise
   ( real or perceived ) to dominate decision-making. Everyone in
   the group ( skiers, forecasters ) should contribute to the decision.
10. Underestimating uncertainty ( denial ) is a method of coping
    with anxiety, especially when the outcome is time-pressured or
    may have a serious outcome. Consider distributional and singular
    information inside a formalized decision-making process to neutralize
    this bias.
11. Excessive optimism is expressed by denial. Seek the opinion of
    a disinterested third party to neutralize this bias.
12. Anchoring is expressed when initial information is given
    more weight in the forecasting process than new information. During
    a ski tour, signs of instability might not be present. However if
    signs of instability appear, it is possible to try and extrapolate
    to the best case scenario, i.e., instability is isolated in this
    location only. While this might be true, it is important to remember
    that "might" is the operational word.
13. A rule of thumb is expressed by a rule that greatly oversimplifies
    the problem. Consistency ( staticity ) in situation is required for a rule of thumb
    to work properly and avalanche forecasting considers a variety of
    very specific, and dynamic, situations. Using a rule of thumb
    almost guarantees that one will overlook important information
    and this will have a negative effect on the decision-making
    process.
14. Clients sometimes pressure guides to travel over terrain that
    is too dangerous. An inexperienced client should not be allowed
    to override the instability assessment of an experienced guide.
15. Social proof is expressed by seeing other people doing something
    without consequences and believing that one can do the same thing
    without consequences. Formalize the decision-making process.