Effects Of Forest Fragmentation On Oak (Quercus Liaotungensis) Populations Of A Small Watershed

The restoration of fragmented landscape is an urgent task and a new challenge of regional biodiversity conservation. In this research, a fragmented Quercus liaotungensis forest landscape in a small watershed was studied. On the basis of spatial structure analysis of the whole watershed and the Quercus liaotungensis patches, and through the DBH area dynamic relationship between dominance tree and companion tree modeling using BP-artificial neural networks, a spatial realistic metapopulation model was developed to model the patch dynamics. Using the metapopulation model the dynamic trend of the Quercus liaotungensis populations' spatial-temporal distribution pattern was predicted. The results showed that:1) The Quercus liaotungensis patch and the Betula platyphylla & Be tula dahurica patch are the main types in watershed; The distribution of area, perimeter and number character of patch types is unequal; The mean patch area and perimeter of each patch type obey a same fractal regularity, and all patch types have a lower folding degree; The variance of isolation among the patch types is high; The different information of landscape spatial quantitive characters can be showed by landscape diversity index which is calculating by path type area, patch type perimeter and patch type number; Using patch type area calculate the dominance index can reflect the landscape in a real manner. Compare with the Betula platyphylla & Betula dahurica patch, the fragmentation degree of Quercus liaotungensis patch is high.2) The Quercus liaotungensis patches has a relative uniform spatial distribution inwatershed. The structure of it represents a type of metapopulation spatial pattern. The variance of patch area and perimeter of the Quercus liaotungensis patches is high, has an unequal distribution; The patch perimeter folding degree of the Quercus liaotungensis patches obey the same regularity with other patch types.3) The BP-artifical neural network model with the 5 input variables, Quercus liaotungensis, Betula dahurica, Acer mono, Fraxinus rhynchophylla and Juglans mandshurica tree species' DBH area and 1 output variable, the sum of other companion tree species' DBH area, when the hiden layer neuron number is 20, learn rate 77 =0.01, momentum B=0.8 and train 30000 epochs, the test accuracy is 0.93.4) The general dynamic trend of the Quercus liaotungensis patches from the year 1987 to the year 2002 can be conclude as: the patch occupancy, patch number, patch area and path perimeter of the Quercus liaotungensis are continuous increasing. When the disturbance are ceasing, the fragmented Quercus liaotungensis population can restore by itself in nature condition.