02910naa a2200289 a 450000100080000000500110000800800410001902400550006010000160011524501660013126000090029752019880030665000130229465300220230765300290232965300140235870000200237270000230239270000250241570000210244070000240246170000200248570000190250570000200252470000170254477300590256120868202018-05-02       2018    bl uuuu     u00u1 u    #d7 ahttps://doi.org/10.1016/j.ecoleng.2017.12.0142DOI1 aREIS, L. P.  aEstimation of mortality and survival of individual trees after harvesting wood using artificial neural networks in the amazon rain forest.h[electronic resource]  c2018  aModeling individual trees in tropical rain forests in the Amazon allows for the safe use of scarce resources in a sustainable way. Unfortunately, in the Brazilian Amazon, rain forest growth and production models are not yet used to estimate future forest stock. Thus, forest management plans do not present technical-scientific support that guarantees sustainable production of wood throughout the cutting cycle. Therefore, this work aims to estimate the survival and mortality of individual trees in a selectively harvested forest using Artificial Neural Networks (ANN) to support silvicultural decisions in forest management in the Amazon rain forest. In 1979, a selective harvest was carried out, with 72.5 m3 ha-1 in an area of 64 ha in Floresta Nacional do Tapajós, in the state of Pará, Brazil. In 1981, 36 permanent plots were installed at random and inventoried. Nine successive measurements were carried from 1982 to 2012. In the modeling, classification, survival, and mortality, training and ANN testing were performed, using input variables such as: different semi-distance-independent competition indices (DSICI), diameter measured (dbh), forest class (FC), trunk identification class (TIC), competition index (CI), growth groups (GG), liana infestation intensity (liana); and crown lighting (CL); Damage to tree (D) and tree rotting (R). The categorical output variables (Classification) were Dead or Surviving tree. Overall efficiency of the classification was above 89% in training and above 90% in the test for all ANNs. Survival classification hit rate was above 99% in the test and training for all ANNs but the mortality score was low, with hit rates below 6%. The overall Kappa coefficient was below 8% for all ANNs (ranked ?poor?) but all ANNs were above 55% in the survival classification (ranked ?good?). ANN estimates the individual survival of trees more accurately but this does not occur with mortality, which is a rarer event than survival.  aFloresta  aGestão florestal  aInteligência artificial  aModelagem1 aSOUZA, A. L. de1 aREIS, P. C. M. dos1 aFREITAS, L. J. M. de1 aSOARES, C. P. B.1 aTORRES, C. M. M. E.1 aSILVA, L. F. da1 aRUSCHEL, A. R.1 aRÊGO, L. J. S.1 aLEITE, H. G.  tEcological Engineeringgv. 112, p. 140-147, Mar. 2018.