Mapping and modelling the habitat of giant pandas in Foping Nature Reserve, China

The fact that only about 1000 giant pandas and 29500 km2 of panda habitat are left in the west part of China makes it an urgent issue to save this endangered animal species and protect its habitat. For effective conservation of the giant panda and its habitat, a thorough evaluation of panda habitat and panda-habitat relationship based on each individual panda nature reserve is necessary and important. Mapping has been an effective approach for wildlife habitat evaluation and monitoring. Therefore, mapping is also an important step in evaluating panda habitat and further being used to analyse panda-habitat relationship. Only Foping Nature Reserve is focused in this study. The objectives of this research are: (1) to develop a highly accurate mapping method which can map panda habitat using multi-type data (remote sensing data, digital terrain data, radio tracking data, and plot data from field survey) in GIS; (2) to study panda movement patterns; and (3) to analyse panda habitat use and selection.A general introduction to the thesis is given in Chapter 1. It describes the research background and problems, and formulates the objectives and outlines of the research.In order to find a potentially better mapping algorithm, three algorithms (i.e., parallelepiped algorithm, maximum likelihood algorithm, and backpropagation neural network algorithm) were evaluated using simulated data sets as well as the remotely sensed imagery in Chapter 2. The discrimination capability of the backpropagation neural network algorithm was also explored in this chapter. The results show that the backpropagation neural network classifier has completely discriminated two spectrally discrete classes, and obtained a significantly higher mapping accuracy than the other two algorithms using both simulated data sets and remotely sensed imagery.Since different mapping techniques have complementary capabilities, two integrated mapping approaches were developed in Chapter 3 so as to combine the advantages from different mapping algorithms.' The "expert "system algorithm based on Bayesian probability theory was firstly discussed in this chapter. One integrated mapping approach is the consensus builder, which is used to adjust classification outputs in the case of a discrepancy in classification between maximum likelihood, expert system and neural network classifiers. The second approach is termed the integrated expert system and neural network classifier (ESNNC), which integrates the output of the rule-based expert system classifier with the backpropagation neural network classifier (BPNNC) before and after running the neural network system. The ESNNC produced maps with the highest accuracy compared to not only the individual backpropagation neural network classifier, expert system classifier and maximum likelihood classifier, but also the combined classifier - consensus builder.The giant panda habitat in Foping Nature Reserve was mapped using the ESNNC in Chapter 4. Two categories of panda habitat types were defined and mapped: ground- cover-based potential panda habitat types and suitability-based panda habitat types. Mapping the ground-cover-based potential panda habitat types used only field survey plot data with records of ground cover types, while mapping the suitability-based panda habitat types used not only the field survey plot data but also radio tracking data - meaning actual panda occurrence. Results show that both the ground cover based and the suitability-based panda habitat types were mapped with significantly higher accuracy compared with non-integrated classifiers: expert system, neural network and maximum likelihood classifiers. The classified maps show us that 97% of the nature reserve is covered by forest and about 68% of the nature reserve is a suitable habitat for pandas.With radio tracking data, panda movement patterns were studied in Chapter 5. The use of GIS combined with statistical tools to thoroughly analyse radio-tracking data to reveal panda movement patterns is a new aspect in panda ecological research. Results show that pandas in Foping NR occupied two distinct seasonal activity ranges (i.e., winter and summer activity ranges) and had a regular seasonal movement between the winter range below 1950 m, and the summer range above 2160 m. Pandas spent about 8 days (from June 7 to 15) to climb up to the summer habitats, while they took about 36 days (from September 1 to October 6) to descend to the winter habitats. Consequently, they spent about 243 days in their winter activity range and about 78 days in the summer activity range. Research also shows that pandas travelled shorter distances with small variation in October, December, January, February, July and August, and longer distances with larger variation in March, April, May, June and September.Analysis of wildlife habitat use and selection has been a common and important aspect of wildlife science. Little is known about panda habitat use and selection, especially about the relationship between panda presence and structures of the bamboo layer as well as the tree layer. In Chapter 6, tracking data were used to analyse panda habitat use and selection, and 110 field survey plots with measured information were analysed to identify differences of characteristics between panda-presence and panda-absence habitats. In the winter range, pandas spend more time in deciduous broadleaf forest with an elevation range of 1600 to 1800 m, a slope range of 10 to 20 degrees, and south- facing slopes. In the summer range, they use more conifer forest with an elevation range of 2400 to 2600 m, a slope range of 20 to 30 degrees. In Bashania fargesii bamboo areas with panda presence, bamboo groves have shorter and denser bamboo culms from different ages. In Fargesia Spathacea bamboo areas with panda presence, bamboo groves have higher coverage, taller and thicker bamboo culms, which are mainly one to two years old.Conclusions from the whole study are summarised in Chapter 7. It is recommended that the whole approach used in this study mayor should be applied to the neighbouring panda nature reserves in the Qinling Mountains. The uncompleted research tasks are discussed in this chapter. Therefore, this chapter has shown some possible research topics for future panda conservation studies.In summary, the following are the main findings of this research:<UL>Backpropagation neural network classifier can discriminate two classes with no overlap in their feature space.The integrated expert system and neural network classifier was developed and applied in mapping panda habitats, and obtained significantly higher overall mapping accuracy than non-integrated classifiers: expert system classifier, backpropagation neural network classifier, and maximum likelihood classifier.The integrated expert system and neural network classifier can identify a class that has only few samples, while the traditional maximum likelihood classifier fails because insufficient samples cannot form the statistical parameters to run the classification.The integrated expert system and neural network classifier successfully classified panda habitat types using multi-type input data: remote sensing data (TM1-S and 7), terrain data (elevation, slope gradient and slope direction), social data (settlement distance), radio-tracking data, as well as field survey plot data.Radio-tracking data were involved in mapping panda habitat for the first time. They can be a good indicator of suitable habitats for pandas.The movement pattern of pandas in Foping Nature Reserve was thoroughly studied and revealed using GIS combined with statistical tools. Pandas spent a very short period of 8 days in June to move from winter to summer habitats, while they used more than one month in September to descend from summer to winter habitats.The finding that pandas in Foping Nature Reserve have a shorter movement distance and a small activity range in January and February indicates these two months may be a good time for conducting a panda population survey.Panda habitat maps produced by the integrated expert system and neural network classifier with higher accuracy have been used for analysing panda habitat use and selection. Pandas in Foping Nature Reserve mainly select deciduous broadleafforest in the winter activity range, and select conifer forest and Fargesia bamboogroves in the summer activity range.The structure parameters of the bamboo layer in panda-presence habitats are significantly different from those in panda-absence habitats.