Background/Question/Methods: Tropical montane forests in the Western Ghats in southern India exist as a mosaic of forests and grasslands separated by a sharp, natural edge. Carbon-isotope studies have revealed that the forests within the ecosystem have experienced expansions and contractions coinciding with glacial advances and retreats respectively. While the ecosystem has been extensively researched, a landscape level analysis of the processes driving the pattern has been lacking. This information is critical to inform ongoing efforts to restore the historical extent of the ecosystem. This study was conducted on a tropical montane forest ecosystem in the Eravikulam National Park located in peninsular India. Remotely sensed Landsat imagery was used to identify patches of shola forests and a digital elevation model (DEM) was used to determine possible causative agents driving the pattern. Slope, Aspect (decomposed into North-South and East-West aspects), Wetness index, Terrain shape index and Elevation were derived from the DEM and used as covariates. A Logistic regression model was fitted to the data to identify the effect of covariates on the occurrence of forests.Results/Conclusions: The logistic regression model correctly classified ~74% of random points as sholas. Area under the receiver operating characteristics plot measured 0.847 (SE 0.011, p<0.0001). In the presence of other covariates, elevation was identified as the strongest determinant of the presence of forests (p<0.0001). Northern (p=0.002) aspects were predicted as more likely to have forests than grasslands. Significant interactions were observed between wetness index and eastern aspects, wetness index and slope and slope and terrain shape index. While, monsoonal rainfall patterns might be expressed in the influence of East/West aspects, field observations confirm the occurrence of shola forests on concave valleys with steep slopes. These observations are indicative of the importance of the interaction between slope, wetness index and terrain shape. Overall, the model tended to over-predict the distribution of forests as compared to their present distribution. We also have reason to suspect that the occurrence of sholas might be correlated with phreatic aquifer zones and soil depth patterns. We believe that further refinement of the model may be required to enhance the grain of analysis and by incorporating data on the underlying geology and soil distribution.
