Economic analysis of geospatial technologies for wildfire suppression
Hayley Hesseln A , Gregory S. Amacher B D and Aaron Deskins CA Centre for Studies in Agriculture, Law and the Environment, University of Saskatchewan, 51 Campus Drive, Saskatoon, SK, S7N 5A8, Canada.
B Department of Forest Resources and Environmental Conservation, College of Natural Resources and Environment, Virginia Polytechnic Institute and State University, Blacksburg, VA 24060, USA.
C National Center for Landscape Fire Analysis, University of Montana, Missoula, MT 59812, USA.
D Corresponding author. Email: gamacher@vt.edu
International Journal of Wildland Fire 19(4) 468-477 https://doi.org/10.1071/WF08155
Submitted: 11 September 2008 Accepted: 11 September 2009 Published: 24 June 2010
Abstract
Geospatial technologies used to fight large fires are becoming increasingly available, yet no rigorous study exists of their effects on suppression costs or fire losses, nor do we know whether these technologies allow more efficient combination of firefighting assets used to suppress fires. The high cost of these technologies merits an assessment of these values. Using data from all large-scale fires originating on US Forest Service land greater than 1620 ha in the Northern Rocky Mountains for the 2000–03 fire seasons, we estimate random parameter models of total fire expenditures, agency fire suppression costs, fire duration, and area burned. Site factors, geospatial technology use, and firefighting assets are used as explanatory variables in these regressions. In addition, stochastic cost frontier models are estimated for suppression costs to judge the efficiency of input use for fires with and without geospatial technology use. We find that although geospatial technology use does not appear to significantly increase suppression costs when other factors are controlled, it does seem to allow more efficient allocation of resources such as labour and capital by fire managers seeking to minimise the costs of controlling large fires. Both of these results suggest that the high cost of using these technologies may be offset by improvements in the use of costly firefighting assets by fire managers.
Acknowledgements
We thank the National Center for Landscape Fire Analysis (NCLFA) at the University of Montana, Missoula, MT, for funding this work, and Lloyd Queen and the staff at NCLFA for advice that helped improve our study, and help with obtaining data. All remaining errors are our own.
Adams R, Crocker T , Katz R (1984) Assessing the adequacy of natural science information: a Bayesian approach. The Review of Economics and Statistics 66, 568–575.
| Crossref | GoogleScholarGoogle Scholar |
Amacher G, Malik A , Haight R (2005) Not getting burned: the importance of fire prevention in forest management. Land Economics 81, 284–302.
Bauer P (1990) Recent developments in the econometrics estimation of frontiers. Journal of Economics 46, 39–56.
Carter D , Cubbage F (1994) Technical efficiency and industry evolution in southern US pulpwood harvesting. Canadian Journal of Forest Research 24, 217–224.
| Crossref | GoogleScholarGoogle Scholar |
Gillmeister W, Moffitt J , Bhowmilk P (1990) Information value in weed management. Northeastern Journal of Agricultural Economics 19, 24–28.
Harris R (1993) Measuring efficiency in New Zealand manufacturing in 1986/7 using a frontier production function approach. New Zealand Economic Papers 27, 57–79.
| Crossref | GoogleScholarGoogle Scholar |
Murnane RJ (2006) Catastrophe risk models for wildfires in the wildland–urban interface: what insurers need. Natural Hazards Review 7(4), 150–156.
| Crossref | GoogleScholarGoogle Scholar |
Peck S , Richels R (1987) The value of information to the acidic deposition debates. Journal of Business & Economic Statistics 5, 205–217.
| Crossref | GoogleScholarGoogle Scholar |
A Fires of this size were sampled because they are those where geospatial technologies are considered a viable firefighting tool. Not all fires made use of these technologies, though, as we will discuss later.
B Obviously, the importance of the cost constraint to minimisation of losses would depend in part on how much the government weighs the cost of resources against actual and expected fire losses. For fires that threaten communities, in theory no cost is spared to minimise losses. For other fires, the fire manager may be more cost-conscious in the choice of resources and technologies. The frontier approach applies more in the latter than the former case. The role of geospatial technology is implicit here, as some fires do not have enough risk associated with the fire, measured in terms of expected and actual losses, to encourage use of the most expensive geospatial resources by the fire managers.
C In stochastic terms, Eqns 2b and 3 would contain an error term that represents our econometric estimation of the stochastic cost frontier function (see Grebner and Amacher 2000). This error term follows a half normal distribution.
D Obviously, the caveat applies that our results are meant to describe suppression only for fires originating from US Forest Service land.
E Random sampling was not an issue because all fires were sampled.
F This is another reason why a random parameters model, discussed below, is appropriate for testing the cost and loss effects of geospatial technologies in our data.
G Furthermore, if prices are fixed, then this is equivalent to estimating costs as a function of expenditures on fire assets, which is consistent with many other cost functions that have been estimated in the forest economics literature.
H The results of these residual regressions led to estimated t statistics for geospatial use of 0.648 using the fire expenditure regression to construct residuals, and –0.281 using the suppression cost regression to construct residuals. The F statistics for the residual regressions for suppression cost and expenditures were 0.2 and 1.1 respectively. By these findings, geospatial technology use is not endogenous with regard to the error in the regressions reported later, and in general the errors are not correlated with explanatory variables. Area burned is also a potentially endogenous variable, but its t statistic in the suppression cost residual regression was –0.245, again insignificant at the 10% and better levels.
I For technical details concerning the use of these models for cost functions, see Bauer (1990) and Greene (1993).
J A plus or minus sign indicates that the estimated regression coefficient is positive or negative respectively in the table.
K Another problem with weather is that detailed information is not available for critical moments of fire suppression when assets were employed. Such data might be collected in the future if agency managers were interviewed during fire suppression activities.
L Total precipitation data ranged from only 0 to 0.61 cm.
