Numerical recipes for disaster: changing hazard and the stand-origin-map

Quantifying Fire Cycle from Dendroecological Records Using Survival Analyses

“…Our simulations allowed us to single out temporal variations in fire activity as the most important factor interfering with estimation of FC from cross-sectional sampling, which confirms previous criticism made in that regard [25,28]. That is because survival analyses are based on TSF data and not on complete fire-return intervals, which can be treated interchangeably only if the failure process (fire) is assumed to be stationary [28].…”

“…In the context of fire history reconstruction obtained from cross-sectional field surveys, such as those simulated in the present study for our case study, sources of uncertainty in the estimation of FC are multiple and include: (1) the variation inherent to sampling; (2) the spatial stochasticity of the phenomenon, which involves the random burning of a landscape in which stands with various TSF values are affected; and (3) temporal variations in fire activity, which interfere with the stationarity assumption that is needed to substitute observed TSF for complete fire intervals [28]. By integrating those sources of uncertainty and by testing a variety of estimation methods in a range of conditions that are representative of the field conditions for which those reconstructions can be useful, we managed to identify the most accurate and robust method as well as the factors that limit its application and reliability.…”

“…Left-censoring is also possible, and it occurs when the lifespan of an individual began prior to the beginning of the experiment with no way of retrospectively determining the moment at which it began. In the present situation, the event of interest is a stand-replacing fire and, in an ideal dataset, two successive fire events would have to be known to observe a closed interval, i.e., uncensored information, e.g., [28,32]. Most of the time, however, it is not possible to obtain such complete information with landscape-scale dendroecologically reconstructed fire history data, especially in regions where fire-free intervals are long.…”

“…That is because survival analyses are based on TSF data and not on complete fire-return intervals, which can be treated interchangeably only if the failure process (fire) is assumed to be stationary [28]. For all methods, increases in fire activity are associated with an underestimation of the length of the FC, and vice versa, although this bias is almost negligible for the FC obtained using Cox regressions.…”

“…The general objective of this paper is to assess and compare the accuracy and precision of three different SA-based methods for estimating FC from dendroecological reconstructions of fire history by means of a modeling experiment designed to simulate a range of conditions that are representative of what is typically encountered in such studies [10,11,14,[22][23][24]. More specifically, we compare the robustness of three methods in the face of three sources of uncertainty that interfere or limit the estimation of historical FCs: (1) the length of the FC, which, when put in relationship with longevity of the main three species, directly affects the proportion of censored observations; (2) the sampling effort, which is a ubiquitous constraint in empirical field studies; and (3) past variation in fire activity, which has been recognized as a potential bias and source of assumption violation for some analytical approaches [25][26][27][28]. Most of these criticisms are related to the assumption that the disturbance process is stationary, which has been suggested to be unrealistic.…”