Sprungmarken

 

Diploma Thesis

Character and magnitude of correlations between vegetation patterns and grazing on the summer pastures in the Fergana Range, Tian Shan (Kyrgyzstan)

Melanie Kappes (10/2006)

Support: Carl Beierkuhnlein, Sebastian Schmidtlein

Abstract

Introduction and intention: Alpine vegetation patterns are the result of complex interactions of environmental gradients. In addition to natural factors, pasturing impacts play a crucial role. For sustainable use, rangeland management, regulating factors like stock species and numbers, time range etc., must be adapted to the vegetation response. On that account a model is needed to serve as an exploration tool of the relationships, a method for display of the current situation and as a basis for predictions of vegetation response to changing pasturing intensities. This intention led to the present study, conducted on a summer pasture in southern Kyrgyzstan (Tian Shan, 41°02’N; 73°33’E), investigating three hypotheses: (1) The impact of grazing diminishes the influence of natural gradients concerning species distribution and vegetation patterns. (2) Due to rugged terrain and grassland vegetation with only slight spectral variation, modeling by means of spectral and environmental information provides significantly better results than conventional supervised classification, applied merely to satellite imagery. (3) Modeling based on environmental information is an applicable method to predict the reaction of the vegetation to changing grazing intensity by means of implementation of scenarios.


Methods: For the modeling process Partial Least Squares Regression (PLSR) was adopted to relate predictors based on a Digital Elevation Model (DEM), the camp positions and ASTER satellite imagery to information on the vegetation (DCA-axes scores of the first three axes). Two approaches were established: the first was based on predictors of DEM and ASTER providing a profound fundament as it combines spectral and environmental information reflecting the current situation. The second approach was based on gradients derived from the DEM and Extracted Accessibility, a measure for grazing intensity deduced from weighted distances to camp positions. In addition to modeling of the current situation, it enables predictions on changing pasturing intensities by the implementation of scenarios. Validation was conducted with the Mantel test, evaluating the similarity between the original species per plot matrix and the modeled DCA-axes scores. For noise removal, improved pattern recognition and interpretation purposes segmentation of the maps of DCA-axes into vegetation classes, derived by Two-Way Indicator Species Analysis (TWINSPAN), was applied. The result was compared to the product of a supervised classification (SC), only based on ASTER bands as well as deduced ratios and indices.

Results: TWINSPAN revealed six vegetation classes: (1) Aflatunia ulmifolia & Spiraea hyperhicifolia, (2) Galium pseudorivale & Carduus nutans, (3) Origanum tytthanthum & Geranium collinum, (4) Ligularia thomsonii & Origanum tytthanthum, (5) Malva neglecta & Urtica dioica and (6) Allium atrosanguineum & Aster alpinus. During the PLSR modeling process altitude, aspect, slope and Extracted Accessibility were found to be the main influencing gradients, being implemented in both modeling approaches. The segmented maps present similar patterns: class one mainly dominates steep areas at lower altitude whereas class five covers the flat areas. Class four bridges to slightly steeper slopes and higher altitudes while classes two and three are situated at medial slopes and altitudes. Class six spans the highest elevations at about 3000m a.s.l. Supervised classification, too, exhibited similar but highly turbulent patterns. Scenarios of decreasing pasturing pressure resulted in declining areas covered by classes four and five while one, two, three and six increased. These tendencies reversed for scenarios of increased pasturing pressure.


Discussion: (1) The investigation of the influence of different gradients on vegetation patterns by means of DCA and PLSR encountered a systematic obstacle since altitude and grazing pressure are correlated due to the location of most tents at low elevations. Thus neither the full range of grazing at every height nor a reference for no grazing exists. Hence the effect of pasturing in comparison to other gradients can not be estimated exactly. (2) Both predictive modeling approaches, based on the two predictor sets, resulted in similar patterns and showed similar model quality. In contrast, SC validation indicated low quality which is displayed by scattered and noisy patterns possibly resulting from to high spectral similarity of vegetation classes, elevated amounts of bare soil disturbing the vegetation reflection, etc. (3) Since Extracted Accessibility was implemented in the best model created by PLSR, adoption of scenarios, presuming higher and lower pasturing intensity, respectively, was possible. Elevated grazing resulted in an enlargement of clusters four and five which actually proved most affected by grazing, featuring indicators of grazing, nitriphication and trampling. In contrast, classes one, two, three and six expanded for decreasing pasturing intensity.


Conclusions: A measure for pasturing intensity was established. A simple model could be created revealing the main influencing factors, displaying the recent situation and enabling predictions. Although the system is very complex and the model suffers from theoretical and practical limitations, it conduced to better understanding of the systems behavior and clarification of further challenges.

last modified 2010-10-18