Outils pour utilisateurs

Outils du site

Panneau latéral

Liste de liens :


Task 3.1 – Risk maps based on fire sizes distribution

Goal: The objective is to propose a new method to compute every day a risk map for the next day. The risk will be computed from the distribution of potential fire sizes and the associated economic costs. It should eventually replace today's maps, that are mainly based on ignition probabilities and loosely based on the fire potential size, in order to better preventively deploy firefighters during summer days.

Inria (Vivien all) SPE (Filippi), LISA (Belgodere, Giannoni, Jouve) T6 to 18

An accurate risk map should depend on three factors: (1) the probability of ignition, which depends on the state of the vegetation and the atmospheric conditions, (2) the probability density function of the fire size (or burned area) in case it ignites, (3) a quantifiable economic and environmental cost for the potentially burned areas. There are already models for the ignition probability. Our forecasting system should include several of them to obtain an ensemble of ignition probabilities, to be calibrated using past observations. Also, we will make use of the evolution in the SURFEX model planned in Task 1. Once ignited, the potential fire size (after one hour, a few hours or one day) depends on the vegetation state, the meteorological conditions and the fire dynamics. These are all highly uncertain, hence in the context of risk assessment, we need to compute the distribution of the fire size, not just a deterministic size. Our strategy is (a) to make use of multimodel ensembles of propagation models (all implemented in the ForeFire software), (b) to carry out model reduction (statistical emulation) on these models to launch large Monte Carlo simulations. Using scores to evaluate ensemble simulations with past observations (fire sizes), like the variance of the rank histogram or the Brier score, the variance on the inputs will be calibrated. While scientific development, parameterisation and validation must be performed. Technical demonstration for these codes has been performed in 2015 (see figure).

Mass simulation demonstrator output. Left simulated map; center FFMC indice (current operational indice); right emulated map. Data from 24 July 2009, grey shapes correspond to the 3 large fires that occurred on this day.

Finally, the risk can be computed as the expected economic and environmental cost (see task 3.2) under the calibrated fire size distribution. This computation should be carried out all over the target territory Balagna for tests (corresponding to the “Homme Milieu” Observatory study area) before scaling at the national level. Salis (2013) launched 100,000 simulations randomly distributed according to the historical ignition frequency in Sardinia. Finney (2011) derived economic values of the burned areas to evaluate the risks. Our contribution will include ensemble simulations at every target location and the use of ensemble-oriented scores for calibration. This task will continue in Task 2.1 where link to on-demand simulation will be made.


             3.1.1 (T18) Test data set (Balagna region) with model inputs and observations.
             3.1.2 (T24) Report (paper) on risk assessment.
             3.1.3 (T24) Verified software for risk forecast.

Resources: PhD (shared with Tasks 2.1 and 3.2 - ensembles) Risk: One difficulty originally lies in the model reduction which requires that the number of significant inputs should be kept low (below 100) but this has been relieved recently at INRIA (results to be published). The largest concern is now the availability of good quality data to perform the simulations.

Task 3.2 – Evaluate economical and environmental impact of fire simulation

Goal: Develop method and code to compute, for each possible fire scenario and/or simulation, an environmental and financial cost of the fire.

LISA (Antoine, DeTotto all) SPE (Filippi), Inria (All) T6 to 30

The problem of costing a fire is complex, and cannot be simplified as a sum of all assets that has been impacted. In this task, an analysis will be carried out from both economic and environmental points of view. Assessing the cost of a given scenario requires an estimate of the social value of the damaged places for each simulation. The social value includes both strictly economic costs and external effects, such as environmental amenities. This will be achieved using the hedonic prices method, providing the original (undamaged) value. Assessing this cost also requires knowing the regeneration process of those places, in order to convert a periodic cost into an intertemporal cost. A bibliographic search will be carried out for each type of damaged places (natural, agricultural, houses…). Possible human costs are also of a major importance, but will not be considered. Their estimation is not simple and would require a large research effort beyond the scope of this project. For each simulation of a given scenario from the model described in Task 3.1, a cost of the fire will be computed, by aggregation of the value loss of each damaged place. The area of destruction will be provided by spatial analyses of the burnt area (intersection between the fire perimeter and the properties map). A theoretical analysis will indicate whether this aggregation should be performed through a simple sum, or through a more complex function that would account for the increasing marginal cost of the fire size. For each scenario, the costs computed in the different simulations will be aggregated in a scenario cost, through a configurable function that exhibits a risk-aversion property. The main goal is not to provide an ambitious economical costing model, but implementing a robust and coherent cost function that in relevance with available data. Overall, this may help designing an optimal (cost-minimizing) tool that can support decision in both forest fighting and planning. Data from the “Observatoire Homme Milieu” will be used to test the functions based on former fires (Balagna region), with 10-year reanalysis planned for validation.


             3.2.1 (T30) Prototype method implemented.
             3.2.2 (T36) Publication/Report on reanalysis perform in Balagna.

Resources: PhD (shared with Tasks 2.1 and 3.1 - ensembles) Risk: Data at a national level may be hard to find, to scale up the system. In this case, only Balagna region will be activated in the prototype platform.

risk.txt · Dernière modification: 2018/05/22 15:20 par