O histórico de incêndios florestais pode ajudar o gestor na localização de áreas de maior risco e, consequentemente, alocar de maneira mais eficaz os recursos de produção. Este trabalho mostra como o histórico de incêndios em florestas de eucalipto pode ser usado para gerar dois modelos preditivos, um de probabilidade de incêndio em nível de talhão e outro de cicatrizes que identificam áreas com cobertura florestal danificada por incêndio. O ajuste do modelo preditivo de probabilidade anual de incêndio teve como variáveis preditivas, uma combinação de variáveis biométricas (volume comercial com casca), climáticas (face de exposição do terreno, precipitação anual, precipitação total anualizada, temperatura média anual e média de umidade relativa do ar), sociais (distância para área urbana, para estradas e para assentamento rural, população municipal, densidade demográfica, e população da zona rural) e de dados processados de levantamentos a laser aerotransportados (ALS): volume estimado por métricas ALS; índice de área foliar para altura total das árvores (LAI), para frações de altura de 1 a 5 metros (LAI_1_5m) e 1 a 10 metros (LAI_1_10m), e estimativa de sub-bosque (proporção entre LAI_1_5m e LAI). Foram utilizadas como técnicas de ajuste, a regressão logística (LOGIT) e o algoritmo Random Forest (RF), que se mostrou superior após o processo de validação-cruzada (tipo \"k-fold\", com k=10). Dados ALS não se mostraram significativos, e o método RF com as variáveis volume comercial com casca, precipitação total anualizada, distância para áreas urbanas e para assentamentos e população da zona rural foi o de melhor eficácia. Esse resultado se expressou nas medidas de especificidade (classificação correta de áreas com registro de incêndio) e performance (classificação correta de áreas preditas como incendiadas). O melhor resultado revela especificidade e performance de 77%. Dentre as variáveis preditoras, a de maior importância foi a precipitação total anualizada. O modelo preditivo de cicatrizes de áreas com cobertura danificada teve sua classificação baseada em três classes: Incêndio, Colheita/Terra Nua e Plantação. O ajuste utilizou como variáveis preditivas 16 métricas multiespectrais, derivadas do sensor RapidEye, e 29 métricas ALS. A resolução espacial das predições é de 5m. Os algoritmos Support Vector Machine (SVM) e Random Forest foram usados como técnicas de classificação, que após a validação-cruzada (\"k-fold\" com k=10), identificou o RF como superior. Neste caso, a inclusão das métricas ALS ao cenário em que se usam apenas dados multiespectrais, aumentaram a sensibilidade para aspectos estruturais da vegetação, verificado para as classes \"Incêndio\" e \"Plantação\" e melhorou a acurácia das predições de 94%, para 97%, e o índice kappa de 90% para 95%. Por importância de capacidade preditiva de cicatrizes de dano, destacam-se as variáveis banda vermelho e NDVI para o RapidEye e, as variáveis relacionadas à cobertura e densidade do dossel, para os dados ALS. Os modelos gerados são úteis para gestores florestais, pois permitem melhor planejamento das operações de combate a incêndio, podendo, inclusive, reduzir custos na operação devido a melhor eficiência logística. / Historical forest fire data can help managers to locate risk areas and, consequently, allocate more efficiently production resources. This work shows how historical fire data from eucalyptus plantations can be used to generate two predictive models, one for fire probability at stand level and another of scars generated from areas with forest cover damaged by fire. The adjustment of the predictive model for fire probability used, as predictive variables, a combination of biometric (volume), climatic (aspect, annual precipitation, annualized total precipitation, annual mean temperature and mean relative air humidity), social (distance to urban area, to roads and to rural settlement, municipal population, demographic density, and rural population), and LiDAR variables: predicted volume by ALS metrics, leaf area index for tree\'s total height (LAI), and for fractions of 1 to 5 meters heigth (LAI_1_5m) and 1 to 10 meters (LAI_1_10m), and a shrub estimation (fraction between LAI_1_5_m and LAI). Logistic regression (LOGIT) and Random Forest (RF) algorithms were compared and RF achieved better accuracy after the 10-fold cross-validation. Adding LiDAR data resulted non significance, and the best adjustment for RF method used wood volume, annualized total precipitation, distance to urban areas, distance to settlements and rural population. The model predictive performance was evaluated by computing the specificity (correct classification of areas with fire registry) and performance (correct classification of areas predicted as burned). The best model yelds specificity and performance of 77%. Among the predictive variables, the one that presented the greatest importance was the annualized total precipitation. The predictive fire scars model had its classification based on three classes: Fire, Harvest / BareLand and Plantation. The adjustment used as predictive variables, 16 multispectral metrics, derived from the RapidEye sensor, and 29 ALS metrics. The spatial resolution of the predictions is 5m. The algorithms Support Vector Machine (SVM) and Random Forest were used as classification techniques, and, after the 10-fold cross-validation RF reached the best tune. In this case, combining ALS metrics to the scenario that used only multispectral data, the sensitivity increased for vegetation structure, verified for the \"Fire\" and \"Plantation\" classes, and improved the prediction accuracy from 94% to 97%, and the kappa index from 90% to 95%. Red band and NDVI were the dominant factors from RapidEye to predict fire scars pixels, and variables related to canopy cover and canopy density were the most important variables from the ALS data. The generated models are useful for forest managers, as they allow better planning of fire-fighting operations, and may even reduce operating costs due to better logistics efficiency.

Our objective is to report mortality rates for ponderosa pine trees in Oregon ten to eleven years after removing a fire-scarred partial cross-section from them, and five years after an initial survey of post-sampling mortality. We surveyed 138 live trees from which we removed fire-scarred partial crosssections in 1994/95 and 387 similarly sized, unsampled neighbor trees of the same species. These trees were from 78 plots distributed over about 5,000 ha at two sites in northeastern Oregon. The annual mortality rate for sectioned trees from 1994/95 to 2005 was 3.6% compared to 2.1% for the neighbor trees. However, many of the trees that died between 2000 and 2005 were likely killed by two prescribed fires at one of the sites. Excluding all trees in the plots burned by these fires (regardless of whether they died or not), the annual mortality rate for sectioned trees was 1.4% (identical to the rate from 1994/95 to 2000) compared to 1.0% for neighbor trees. During these fires, a greater proportion of sectioned trees died than did catfaced neighbor trees (80% versus 64%) but the difference was not significant.

No description available.

This research investigates soil respiration (Rs) in a boreal jack pine (Pinus banksiana Lamb.) fire scar chronosequence at Sharpsand Creek, Ontario, Canada. During two field campaigns in 2006 and 2007, Rs was measured in a chronosequence of fire scars in the range 0 to 59 years since fire. Mean Rs adjusted for soil temperature (Ts) and soil moisture (Ms) (Rs T,M) ranged from 0.56 μmol CO2/m2/s (32 years post fire) to 8.18 μmol CO2/m2/s (58 years post fire). Coefficient of variation (CV) of Rs adjusted for Ts and Ms ranged from 20% (16 years post fire) to 56% (58 years post fire). Across the field site, there was a significant exponential relationship between Rs adjusted for soil organic carbon (Cs) and Ts (P = 1.24*10-06; Q10 = 2.21) but no effect of Ms on Rs adjusted for Cs and Ts for the range 0.21 to 0.77 volumetric Ms (P = 0.702). Rs T,M significantly (P = 0.030) decreased after burning mature forest, though no significant (P > 0.1) difference could be detected between recently burned and unburned young forest. Rs was measured in recently burned boreal jack pine fire scar age categories that differed in their burn history and there was a significant difference in Rs T,M between previously 32 v 16 year old (P = 0.000) and previously 32 v 59 year old (P = 0.044) scars. There was a strong significant exponential increase in S R T,M with time since fire (r2 = 0.999; P = 0.006) for the chronosequence 0, 16 and 59 years post fire, and for all these age categories, Rs T,M was significantly different from one another (P < 0.05). The Joint UK Land Environment Simulator (JULES) was used to model vegetation re-growth over successional time at Sharpsand Creek, though it appeared to perform poorly in simulating leaf area index and canopy height. JULES probably over estimated heterotrophic Rs at Sharpsand Creek when Ts corrected simulated values were compared with measured Rs T,M. The results of this study contribute to a better quantitative understanding of Rs in boreal jack pine fire scars and will facilitate improvements in C cycle modelling. Further work is needed in quantifying autotrophic and heterotrophic contributions to soil respiration in jack pine systems, monitoring soil respiration for extended time periods after fire and improving the ability of JULES to simulate successional vegetation re-growth.