Space-time analyses for forecasting future incident occurrence: a case study from Yosemite National Park using the presence and background learning algorithm

This follow up paper to the Yosemite Search and Rescue Incident Georeferencing Study has been published in the International Journal of Geographical Information Science. Many thanks to my colleagues and all of the volunteers who have helped support this project! 

Abstract

To address a spatiotemporal challenge such as incident prevention, we need information about the time and place where incidents have occurred in the past. Using geographic coordinates of previous incidents in coincidence with spatial layers corresponding to environmental variables, we can produce probability maps in geographic and temporal space. Here, we evaluate spatial statistic and machine learning approaches to answer an important space-time question: where and when are wildland search and rescue (WiSAR) incidents most likely to occur within Yosemite National Park (YNP)? We produced a monthly probability map for the year 2011 based on the presence and background learning (PBL) algorithm that successfully forecasts the most likely areas of WiSAR incident occurrence based on environmental variables (distance to anthropogenic and natural features, vegetation, elevation, and slope) and the overlap with historic incidents from 2001 to 2010. This will allow decision-makers to spatially allocate resources where and when incidents are most likely to occur. In the process, we not only answered questions related to a real-world problem but also used novel space-time analyses that give us insight into machine learning principles. The GIScience findings from this applied research have major implications for best practices in future space-time research in the fields of epidemiology and ecological niche modeling.

Download the Paper

The IJGIS will provide free access for the first 50 downloads. Since the GIScience community already subscribes to this publication, I thought I would open this up to the Search and Rescue GIS Community: Download Here

Conclusion for Search and Rescue GIS

• Both where and when an incident occurs is important.
• SAR incidents occur where visitation is likely highest (obvious) - but visitor use activity is also not well mapped in recreational areas like Yosemite. Therefore it is difficult to map risk factors independently. 
• If you don't map where an incident has occurred how will anyone else ever learn from the experience? 
• Maps are an extremely compelling tool for telling a story about a place and capturing institutional knowledge.
• GIS is an under utilized tool in Search and Rescue and this research is just beginning to scratch the surface

Acknowledgments

This research initiative is supported by the National Science Foundation (grant nos. BDI-0742986 and SBE-1031914). I would like to thank Yosemite Search and Rescue, Yosemite Volunteers-In-Parks, and the Yosemite National Park Division of Resource Management and Science for research permissions (OMB#1024-0236) and constructive suggestions. Special thanks to my Dissertation Commitee: Dr. Samuel Traina, Dr. Ruth Mostern, Dr. Yihsu Chen, labmates Wenkai Li and Otto Alvarez, co-authors Yu Liu and John Wieczorek, and especially my PhD advisor Dr. Quinghua Guo. Thank you to volunteers Diane and Greg Ambrose and Sarah Nurit for all of the Georeferencing and clerical work!

If we want to collaborate in follow up research, contact the Spatial Analysis & Research Center at University of California Merced (SpARC). 

This map below is just a point layer of cumulative incidents. Stay tuned for time-enabled maps and maps that filter by incident type.

Critical Planning and Analysis using GIS for WiSAR

By Don Ferguson 
dferguson@mix.wvu.edu

Wilderness search and rescue is understood to be an inherently spatial problem which is relative to both the subject and searcher.  From the standpoint of the subject, many decisions that are made before and after becoming lost or injured are influenced by the terrain and the environment.  These decisions may be either conscious (active) or sub-conscious (passive), and in many ways are driven by both time and space.  In WiSAR related to lost persons, it is the decisions that were made by the subject that resulted in them becoming lost.  Studies on lost person behavior have allowed searchers to categorize individuals that exhibit similar behaviors when they become lost.  These categorical behaviors are observed by plotting Initial Planning Points (Point Last Seen or Last Known Point) and Find locations then extracting information about the terrain and observing trends in the data.  When combined with a critical analysis on the influence of terrain and environment (T & E) on a specific individual, geospatial trends in lost person behavior provide valuable information that could reduce the time taken to locate a lost subject.

Integrated Geospatial Tools for Search and Rescue (IGT4SAR) is a dynamic tool developed to take advantage of using Geographic Information Systems to model lost person behavior and provide a critical analysis on the influence of T & E.  The primary advantage of a GIS is that it allows a user to interact with spatial data and even create new data from existing information, for example estimating cellular coverage across the search area using a digital elevation model and cell tower attributes.  No more is a search analyst limited to extracting information from a printed, topographical map that is most likely outdated.

WiSAR operations that involve a lost subject are plagued with uncertainty.  Where did the subject leave the trail, or is the object found by a search team an actual clue related to the lost subject?  In order to deal with the cognitive complexity of all this uncertainty search analysts often resort to developing scenarios, or hypotheses, to describe what is believed to have happened to the lost subject.  These scenarios provide justification for applying resources to specific geographical locations.  This is critical as lost person searches often cover large geographical areas and have few resources with which to search.  Thus a method is required to assist in prioritizing the search area otherwise the search effort is limited to merely purposeful wandering which is typically less effective than targeted searching.  Among other functions, GIS allows analysts a way to “play-out” various scenarios to determine what is possible and likely. 

Several Lost Person Behavioral models are built directly into IGT4SAR, for example using data provided in Robert Koester’s text on Lost Person Behavior, concentric rings are automatically drawn around the IPP based on subject category that represent the recorded distances to find locations of similar individuals.  Similarly, a Track Offset Model provides a visual representation of how far from a linear feature such as a trail or road in which the subject was found.  The Find Locations tool provides a means of re-classifying spatial data such as roads, trails and hydrology vector data along with a land cover surface raster to display the typical types of features where similar subject were found.

Expanding on the concept of reclassifying raster data, a similar approach is used to create a Least Cost Path Surface that represent the potential subject mobility, or distance travelled, over a period of time.  This model takes terrain features such as slope, access to travel aides (roads, trails, etc) and barriers (bodies of water) along with land cover to estimate how far a subject could have traveled over a given period of time.  Analysis of historical data from Yosemite National Park suggested subjects (predominately hiker category) did not travel more than 1.5 hours walking distance from the IPP.  Know this information could drastically reduce your search area. See Jared Doke's MS Thesis for more on this study. 

Combining these estimates together with specific information you know about the subject and the local T & E, an analyst can “play-out” various scenarios to see what makes sense and assist is assigning a probability of most likely occurrence to the various regions.   This ultimately leads to a Probability Density plot of the search area.   While assigning numerical values in the form of Probability to various regions of the search area may be slightly misleading as it gives the impression that a rigorous quanitfied analysis has been done as opposed to the qualified (Bayesian) analysis, the numerical values allow for easier tracking of progress within the search area in the form of Probability of Detection and Probability of Success.  While to some this may sound complex, GIS is well equipped for handling these types of analysis and these concepts are built into the functionality of IGT4SAR.

To learn more about Integrated Geospatial Tools for Search and Rescue (IGT4SAR) and for using GIS for critical analysis and planning of lost person incidents subscribe to the YouTube videos at:

http://www.youtube.com/channel/UCrWNjhnpNOiEAATDzNw3lFg

If you have development skills or are a GIS Specialist and would like to test these tools please see the GitHub Repo: https://github.com/dferguso/MapSAR_Ex

Expert versus Machine: A Comparison of Two Suitability Models for Emergency Helicopter Landing Areas in Yosemite National Park

Ever wondered if you could use GIS to have a better understanding of where you can land a helicopter for your area? Well, I hope this sheds some light on two possible options. This was the first WiSAR project from my Phd Research. 

View Larger Map

Here is an interactive web map of Yosemite National Park with the output from the maximum entropy algorithm. Green areas are more likely to be suitable landing areas. Blue dots are landing zones used in the past.

Basically in this study we compare two methods for creating a helicopter landing suitability map (see above). We tested the landing suitability map using actual landing areas used by Yosemite helitack during operations. The idea was to test two GIScience methods, but also to give dispatchers and operations an extra location-based tool they can use to decide if it is worth scouting for a landing zone or gearing up for alternatives (litter carry out, short-haul, heli-rappel, etc.) 

Abstract
"Landing a rescue helicopter in a wilderness environment, such as Yosemite National Park, requires suitable areas that are flat, devoid of tree canopy, and not within close proximity to other hazards. The objective of this study was to identify helicopter landing areas that are most likely to exist based on available geographic data using two GIScience methods. The first approach produced an expert model that was derived from predefined feature constraints based on existing knowledge of helicopter landing area requirements (weighted overlay algorithm). 

Yosemite Search and Rescue loading a patient into Helicopter 551 - NPS

The second model is derived using a machine learning technique (maximum entropy algorithm, Maxent) that derives feature constraints from existing presence-only points; that is, geographic one-class data. Both models yielded similar output and successfully classified test coordinates, but Maxent was more efficient and required no user-defined weighting that is typically subject to human bias or disagreement. The pros and cons of each approach are discussed and the comparison reveals important considerations for a variety of future land suitability studies, including ecological niche modeling. The conclusion is that the two approaches complement each other. Overall, we produced an effective geographic information system product to support the identification of suitable landing areas in emergent rescue situations. To our knowledge, this is the first GIScience study focused on estimating the location of landing zones for a search-and-rescue application." Doherty et al. 2013. Professional Geographer 65:3

Next steps are to:

1) Test these methods in other mountainous environments 

2) Upload the python scripts for the generating raster from vector for formatting into Maxent

3) Upload the ArcGIS ModelBuilder model as a python geoprocessing script to GitHub

4) Integrate the model as a tool in MapSAR, ArcGIS Desktop template.

I would love to collaborate on this project and help answer any questions you might have. 

Special thanks to my co-authors Dr. Guo and Otto Alvarez, Yosemite Search and Rescue, UC Merced, the National Science Foundation, and Liz Sarow from Esri for the inspiration. 

Collecting and Analyzing Missing Person Data for Alzheimer Patients in British Columbia

As part of my project for my GIS Certificate at the University of the Fraser Valley, I was required to complete a project of my choosing. As a SAR Manager for Ridge it only seemed natural to work on something that hopefully would be of some use and interest to the SAR community.

With Robert Koester’s book Lost Person Behavior and the ISRID database as a starting point, I looked at the information available from the local SAR teams in my area (southwestern British Columbia).

One team in particular, Surrey Search and Rescue, performs a large number of searches for missing Dementia subjects. Between Surrey SAR, Coquitlam SAR, and Ridge Meadows SAR I was able to gather the results of 51 searches spanning the years 2001 to 2012.

Using ESRI’s ArcMap 10.1 (see www.mapsar.net for more info), I plotted the PLS and Found locations of these subjects and from that, calculated their mean distance travelled as well as the over-all mean direction of travel. This data is based on searches that occur in the urban setting of a Westcoast Canadian city. Given more time and resources, a study of several diverse Canadian cities may yield results that differ from or are similar to the outcome of this project.

These types of searches are perhaps the most frustrating for a SAR team. With more studies like this and a well-populated database such as the ISRID database, SAR Managers may have one more tool to assist them in bringing about a successful conclusion to a difficult task.

I would like to acknowledge the guidance from my professor, Dr. Scott Shupe for making most of this course understandable and enjoyable.

- Rick Laing
rmsarmanager@gmail.com

Stay tuned for a similar study from Yosemite National Park!

Paul Doherty Dissertation Defense: Search and Rescue in Yosemite

The applications of GISystems to wilderness search and rescue, an overview within a GIScience context and examples from Yosemite National Park. Click on this link for flyer with more information

When: Wednesday, May 8th 1030h PST [new time]

Where: University of California Merced, 5200 Lake Rd Merced, California 95340, Science and Engineering Room 200 - click below for an interactive map. Please send me an email at pjdohert@gmail.com if you have not already confirmed your attendance.

Webcast: My YouTube Channel



I will also be giving a public presentation at the Yosemite Open Forum next week on Tuesday, May 14th at Noon. Click here for more details.

For more information on my research see my research page.