Wildlife habitat mapping using Sentinel-2 imagery of Mehao Wildlife Sanctuary, Arunachal Pradesh, India.

Mehao Wildlife Sanctuary, situated in the state of Arunachal Pradesh, is part of an important biodiversity hotspot in the north-eastern part of India in the Himalayas. The current study deals with the identification of important wildlife habitats in the sanctuary. We used a supervised classification technique to delineate these habitats in the sanctuary, which are used by several mammals and bird species encountered during camera trap and sign surveys conducted between November 2017 and May 2020. Satellite images from Sentinel - 2A were used to classify the land use land cover (LULC) of the sanctuary. The LULC information was generated by using a maximum likelihood classifier. We classified a total of thirteen LULC classes, i.e., water, built-up, agriculture, orchard, grassland, bamboo forest, bamboo-mixed forest, riverbed, barren land, snow, wild banana, riverine forest and mixed forest. LULC classification reveals a high percentage of mixed forest, about 69.9%, followed by wild bananas at 7.2%. The commission and omission error rates, however, are high for riverbed and agriculture (0.5) and bamboo forest (0.5), respectively. The accuracy assessment showed an overall classification accuracy of 88.5% with a Kappa coefficient of 0.87. The abundance of mammals was high in the mixed forest, but Ivlev's electivity index shows that species generally avoided this habitat and preferred specialized forest habitats, such as bamboo forest, bamboo-mixed forest, grassland, riverbed and riverine forest. Our LULC map will provide a baseline for potential planning and monitoring changes of wildlife habitats in Mehao WLS.

Resource selection probability functions for gopher tortoise: providing a management tool applicable across the species' range.

The gopher tortoise (Gopherus polyphemus) is protected by conservation policy throughout its range. Efforts to protect the species from further decline demand detailed understanding of its habitat requirements, which have not yet been rigorously defined. Current methods of identifying gopher tortoise habitat typically rely on coarse soil and vegetation classifications, and are prone to over-prediction of suitable habitat. We used a logistic resource selection probability function in an information-theoretic framework to understand the relative importance of various environmental factors to gopher tortoise habitat selection, drawing on nationwide environmental datasets, and an existing tortoise survey of the Ft. Benning military base. We applied the normalized difference vegetation index (NDVI) as an index of vegetation density, and found that NDVI was strongly negatively associated with active burrow locations. Our results showed that the most parsimonious model included variables from all candidate model types (landscape features, topography, soil, vegetation), and the model groups describing soil or vegetation alone performed poorly. These results demonstrate with a rigorous quantitative approach that although soil and vegetation are important to the gopher tortoise, they are not sufficient to describe suitable habitat. More widely, our results highlight the feasibility of constructing highly accurate habitat suitability models from data that are widely available throughout the species' range. Our study shows that the widespread availability of national environmental datasets describing important components of gopher tortoise habitat, combined with existing tortoise surveys on public lands, can be leveraged to inform knowledge of habitat suitability and target recovery efforts range-wide.

Environmental drivers and spatial dependency in wildfire ignition patterns of northwestern Patagonia.

Fire management requires an understanding of the spatial characteristics of fire ignition patterns and how anthropogenic and natural factors influence ignition patterns across space. In this study we take advantage of a recent fire ignition database (855 points) to conduct a comprehensive analysis of the spatial pattern of fire ignitions in the western area of Neuquén province (57,649 km(2)), Argentina, for the 1992-2008 period. The objectives of our study were to better understand the spatial pattern and the environmental drivers of the fire ignitions, with the ultimate aim of supporting fire management. We conducted our analyses on three different levels: statistical "habitat" modelling of fire ignition (natural, anthropogenic, and all causes) based on an information theoretic approach to test several competing hypotheses on environmental drivers (i.e. topographic, climatic, anthropogenic, land cover, and their combinations); spatial point pattern analysis to quantify additional spatial autocorrelation in the ignition patterns; and quantification of potential spatial associations between fires of different causes relative to towns using a novel implementation of the independence null model. Anthropogenic fire ignitions were best predicted by the most complex habitat model including all groups of variables, whereas natural ignitions were best predicted by topographic, climatic and land-cover variables. The spatial pattern of all ignitions showed considerable clustering at intermediate distances (<40 km) not captured by the probability of fire ignitions predicted by the habitat model. There was a strong (linear) and highly significant increase in the density of fire ignitions with decreasing distance to towns (<5 km), but fire ignitions of natural and anthropogenic causes were statistically independent. A two-dimensional habitat model that quantifies differences between ignition probabilities of natural and anthropogenic causes allows fire managers to delineate target areas for consideration of major preventive treatments, strategic placement of fuel treatments, and forecasting of fire ignition. The techniques presented here can be widely applied to situations where a spatial point pattern is jointly influenced by extrinsic environmental factors and intrinsic point interactions.