The study area is located in Serra do Socorro (Torres Vedras, central Portugal) and covers ~1449 hectares (Fig. 1). Serra do Socorro is the steepest landscape in the municipality of Torres Vedras (Gabinete Técnico Florestal, Torres Vedras, 2015-2019). Topography is dominated by uplands and lowlands ranging from 51 meters to a maximum of 389 meters elevation, with an average value of 168 meters.

The area is characterised by a strong Atlantic wind influence and consequently, throughout the whole year, atmospheric humidity remains high which is especially noticeable in summer when compared with other regions in Portugal (more than 70% relative humidity). In terms of temperature, the maximum (26 °C) is reached during August. Precipitation values, in the fire season (especifically during July and August), do not reach 7 mm as average value per month (values are given for the historical period 1961-1990) (Gabinete Técnico Florestal, Torres Vedras, 2015-2019).

The main forest land cover type is eucalypt plantations (175 ha) surrounded by small agriculture, shrublands and uncultivated land. These plantations are managed by The Navigator Company, the leading paper company in Europe. The site presents the following land use classes: non burnable areas (5.9%), blue gum plantations (12%), extensive agricultural lands and pasture lands (56.7 %), shrublands (24%), pine stands (0.24%).

The municipality of Torres Vedras is considered under a low intensity fire regime with a long fire season. Fire history in Torres Vedras is defined by a significant number of events, mainly in summer season (maximun in October) and small size (less than 1 hectare) (Gabinete Técnico Florestal, Torres Vedras, 2015-2019). Those autumn and winter fires mainly affect shrublands and farming areas (Marques et al., 2014). Social and demographical factors also play a determining role in Serra do Socorro fire regime, given the strong links between population and agricultural activities and the common use of fire irrespective of season (Raínha & Fernandes, 2002).

Changes in fuels, carbon, timber volume and fire behaviour, were explored in response to fuel treatments over time for the present (2015, t=0) and three future periods (2018-2021-2024, t=1, t=2, t=3, respectively). A three-tiered strategy was applied as follows.

We focused on changes in two fire metrics, flame length (FL) and rate of spread (ROS) patterns over time in response to treatments. The maximum FL was observed at the earlier time periods and decreased thereafter. The maximum FL was observed in year 2018 (t2) presumably due to eucalypt stand structure in young plantations where most trees are between three and five years old. The highest ROS was observed for young eucalypt plantations and in the adjoining shrublands. We observed decreasing ROS with increasing stand age.

LTD creates both aggregated (Fig. 2) and dispersed treatment schedules (Fig. 3), each representing a different spatial treatment scenario. It might be assumed that large land owners such as The Navigator Company would focus on aggregate fuel treatments while forest owners might prefer non-aggregated treatments.

The higher fire risk levels of the periods t₀ and t₁ (Fig. 2a, 2b; 3a, 3b) resulted in the allocation of most of the fuel treatments within plantations in a juvenile stage. This is consistent with the previous fire behaviour outputs by FlamMap in terms of FL and ROS, which were largest in the young eucalypt stands. In t₂ and t₃ (Fig. 2c, 2d; 3c, 3d) fuel treatments were allocated by LTD to both, within eucalypt plantations and outside plantations, in shrublands areas. Young eucalypt plantations often have high density, crowns are low and closer to the ground and have shrubs understory and thus generate a continuous fuel layer and lead to higher fire hazard (Botequim et al., 2013).

The decreasing trend in the number of treatment projects over time was consequence of the forest fire hazard status through time (e.g. the strategic placement of project areas over time in Fig 3a,3b presented projects 1, 2, 3 and 4 while in Fig. 3c, 3d presented only projects 1 and 2). The number of resulting projects decrease as fire metric values decrease over time ( t2, t3) and do not surpass the defined thresholds thus do not trigger fuel treatments operations.

The output from fire simulation and strategic allocation of fuel treatment projects from LTD were consistent with experiences from local managment and current silviculture practices performed by the The Navigator Company to manage wildfire risk.

The LTD program selected areas for treatments based on fire hazard and fire risk. In Portugal, a typical eucalypt full rotation may include up to 2 or 3 coppice cuts, each cut being followed by a stool thinning that may leave an average number of shoots per stool ranging from 1 to 2. Harvest ages typically range from 10 to 12 (Soares & Tomé, 2001). Prescriptions further include several shrub cleanings over a rotation (i.e., 1 to 3 fuel treatments per cycle - Duncker et al., 2012). The specific fuel treatments, including slash removal and prescribed burning would be carried out in these areas according to local conditions to modify fuel structure and fuel load changes. In our study area, prescribed burning as fuel management is the treatment preferred for the removal of accumulated fuel loads in blue gum plantations (i.e., after stool selection) (Pinto et al., 2013). Other punctual operations depending on the stand status might include mechanical treatments used separately or in concert with prescribed fire to reduce fuel loads and alter fuel structure and thus minimise fire impacts (Fernandes et al., 2011). Nevertheless, several studies suggest that the final size of individual fires can be impacted by landscape treatment rates as low as 5% (e.g. Cochrane et al., 2012).

The methods we used can potentially have wider application to explore fuel and fire management options in eucalypt plantations. By coupling FlamMap and LTD to examine temporal changes in plantations we were able to add a temporal analyses to the strategic implementation of fuel management. This temporal aspect was made possible by pre-existing research that led to the development of growth and yield model, canopy equations and fuel models descriptions.

Given the fire risk and cost of performing forest and shrublands fuel treatments across large areas, effective allocation of resources is critical. It is very challenging to strategically select fuel treatment locations on space and time with heterogeneous landscape i.e., topography, fuel models, and highly valuables resources susceptible to potential fire damages (Wei & Long, 2014). In this sense, we demonstrated a new approach for prioritising forest management through time in eucalypt stands in Serra do Socorro.

The framework of our method joined models of forest stand dynamics with optimisation models to provide a new decision support system for locating areas for fuel treatments. The study demonstrates new ways to develop efficient risk mitigation practices focusing on spatial-time-investment fuel treatment strategies in fire-prone landscape. The simulation methods also provide a way to communicate wildfire risk to the diverse array of landowners and other stakeholders in Serra do Socorro and adjacent areas in Central Portugal. Thus the methodology can be implemented as part of operational-planning decision-making and helps efficient forest management when answering spatial-temporal question such as where and when fuel treatments are required under the desired objectives.

The literature provides little guidance for determining the required fire behaviour knowledge (e.g. the use of crown fire models developed for conifers and the spotting potential of eucalypts) for an effective protection of eucalypt plantations (Fernandes et al., 2011). Indeed, the limitations of simulation models has been suggested by several authors (Alexander & Cruz, 2013; Cruz & Alexander, 2010). A large number of observations obtained from carrying out experimental simulations studies (e.g. FlamMap5) showed that, in many cases, they are instrumental to predict fire behaviour characteristics and can quantitatively replicate large wildfire events, in terms of predicted burned area (Finney et al., 2011), and size and shape of perimeters (Ager et al., 2014a; Salis et al., 2014). Wildfire simulations should be regarded as general indicators of wildfire exposure and fuel management planning for prioritising fire protection efforts (Salis et al., 2013). We attempted to minmise errors with the application of FlamMap by using calibrated fuel models for the study area (Fernandes et al., 2009). We acknowledge that even if the strategic location of fuel treatments projects are in line with local manager experiences and silviculture practices, uncertainties in the the fuel model and weather data input (i.e., wind conditions) contribute to variability in our modeled outputs (Oliveira et al., 2016).

In future work we will conduct sensitivity analyses to test a wide range of scenarios that vary in terms of intensity of area treated (hectares). We also suggest that future studies include social and demographic factors since these play a role in Serra do Socorro fire regime as evidenced by the strong links between population and agricultural activities and the common use of fire irrespective of season (Raínha & Fernandes, 2002). Coupling mitigation efforts for both social and biophysical risk factors in terms of human ignitions and fuel management is likely the most efficient mitigation strategy for wildfire risk management in fire prone eucalypt plantations in Portugal. Moreover, a better understanding of wildfire risk transmission between eucalypt plantations and surrounding communities (Ager et al., 2014b; Haas et al., 2015; Oliveira et al., 2016) will facilitate the development of strategic fuel treatment plans aimed at reducing future wildfire impacts.