|
Spatial ecology and GIS based topics
IN340 Mapping forest change in space and time in Indonesia (Weeks 2 - 8; need to have completed IL001) The forests of Lambusango change across the landscape, with some forest on recent limestones, others on much older limestones, sandstones and cherts and others still on ultramafic rocks, one of the rarest forest types in southeast Asia. These forests are also changing with encroachment being a problem at points around the edges. One team is carrying out detailed and broad scale mapping of forest characteristics on a variety of geology and at differing stages of degradation and regeneration and use these data to classify satellite imagery from Lambusango. However, there are a variety of challenges including a continuum of forest biomass variation rather than discrete classes of forest type, and the limitations of satellite image resolution. This project will gather data on forest characteristics, such as tree size and stand density, in a selection of forest types. Then high resolution imagery will be processed, trying to delineate individual trees and investigating whether data derived from the imagery correlates with forest characteristics. This would allow broad scale forest characteristics maps to be created, giving a better understanding of how variables such as forest biomass vary across the Lambusango Forest. This topic requires that you have some prior experience of processing remotely-sensed imagery as you will have to process your data independently once you get home.
MN341 Species distribution modelling in Madagascar (Weeks 1 - 6) Distribution models allow a set of spatial records for a given species (from our databases) to be integrated with maps of environmental covariates (e.g. elevation, climate, land cover) in order to construct and validate a statistical model of the probability that a given species will be found in a particular landscape unit. These models can then be expressed as a habitat suitability map. It will be possible for students in 2012 to join one of the science teams and contribute to collecting field data for lemurs, forest birds, wetland birds, or reptiles and amphibians and then use our entire dataset to make models for a set of species using either GLM or Maxent. Outputs from these studies would be very helpful as the maps produced can feed directly into our systematic conservation planning process and inform the management of the Mahamavo region. High quality maps are also excellent communication tools for explaining the significance of the site to decision makers.
MN342 Landscape ecology in Madagascar (Weeks 1 - 6) By conducting biodiversity surveys, we build up a knowledge base concerning patterns in the environment. However, in order to make resilient conservation plans for a dynamic future characterised by land cover change, climate change, human population growth and infrastructure development, we need to be able to understand the processes which are affecting the distribution and density of species within the landscape. It would be possible to join the teams conducting field surveys of lemurs, forest birds or reptiles to contribute to data collection, then return to base camp and use our full database linked to our spatial data to infer population processes from patterns of biodiversity. In particular, it would be very useful to test to what extent various species in a particular guild are affected by patch size, edge effects, isolation and compactness and therefore predict the likely consequences for biodiversity of habitat fragmentation in future environmental scenarios.
MN343 Community ecology in Madagascar (Weeks 1 - 6) Which processes, including habitat and ecological interactions, structure communities of forest birds, reptiles and lemurs in Mahamavo? In terms of habitat, there is scope for comparison of primary and secondary dry forest and exploration of the effects of gradients in moisture between relatively moist and highly xeric forests. This might permit the identification of indicator species for particular forest types. A more sophisticated approach would be to use Mantel tests to test a suite of competing hypothesis about the environmental processes which explain pairwise dissimilarity in the community of reptiles/birds/lemurs between lots of pairs of sample units as a function of distance, difference in environmental variables such as moisture, and difference in habitat configuration. Additionally, it would be possible to test whether ecological interactions, especially competition, within a taxonomic group may be structuring the community. This could be achieved by co-occurrence tests or generalised dissimilarity models. For some groups, development of ecological dissimilarity (ED) based monitoring indicators for environmental condition which track communities through ecological space through time would be a very promising direction to investigate. Alternative directions to take might be to make distribution models and then maps of beta-diversity or to use numerical classification to make maps of community types.
MN344 Remote sensing and environmental modelling in Madagascar (Weeks 1 - 6) Freely available satellite images can enable frequent synoptic observations of whole landscapes. However, it is critical to develop and implement methods that convert this stream of raw data into useful knowledge about the state of the environment. It would be possible for dissertation students to undertake projects on classification methods, change detection, multi-sensor fusion, field evaluation of MODIS products, inversion of the satellite signal to biophysical parameters and hyperspectral remote sensing. Students choosing projects in this area will benefit from considerable on-site technical expertise in processing satellite data, will gain a broad experience of remote sensing methods, and become proficient in some advanced techniques. One possible project could involve collecting a field reference dataset using GPS and then making and validating a classifier for temporally coincident moderate resolution multispectral data from Landsat 5, Landsat 7, ALI and ASTER probably using a maximum likelihood approach, but it would be possible to investigate tree-based, Bayesian or object oriented classifiers too. Another project would be to use our existing collections of processed images to evaluate the strengths and weaknesses of a range of change detection methods, including univariate differencing, delta classification and change indices such as the TC disturbance index. In this project, it would also be possible to investigate the effects of scale by contrasting results from Landsat-like sensors with coarser-resolution sensors such as MODIS. This project could make recommendations about the best way to operationally monitor the forest using satellite data in the long term. A third project could take the latest fusion algorithms and use them to develop an effective tool (such as a macro or script) that would allow environmental managers in developing countries to be able to combine the latest 500m MODIS reflectance composite with recent Landsat-like data to frequently produced high-resolution images for monitoring purposes. A lot of derived data products are routinely produced from MODIS data for environmental monitoring purposes, in particular MOD13 vegetation indices, MOD14 fire products and MOD12 change products. However, before these datasets can be routinely used for monitoring in a particular site, it is important to check that the automatic products accurately capture local scale processes. Mahamavo is an ideal test system to investigate the accuracy of these products in relation to field survey data and higher resolution images since local people frequently burn the forest and savannah and the forest extent is highly dynamic, exhibiting complex patterns of forest loss and regeneration.
MN345 Developing monitoring protocols for REDD+ in Madagascar (Weeks 1 - 6) Climate change mitigation initiatives such as REDD+ have focused attention on the need to be able to monitor carbon stocks and flows in tropical forests in order to administer schemes which compensate local people for avoided deforestation. Field surveys of forest structure, as conducted by Opwall, are an effective means of gathering this data for small areas. However mapping forest biomass over large areas requires the development of methods for inversion satellite signals to estimate biophysical parameters, such as canopy height or basal area. There are a variety of ways doing this from extremely complex physical models of light scattering to artificial neural networks to very simple regression methods. A possible project would integrate satellite data (either Landsat, MODIS or MISR) with field data from forest plots and also measurements taken from interpretation of high-resolution commercial imagery (e.g. Quickbird) to develop, refine and validate a statistical model (or a neural network) to take satellite data and generate maps of forest properties such as above-ground woody biomass.
MN346 Evaluation of biodiversity monitoring methods in Madagascar (Weeks 1 - 6) It is only worthwhile to commit resources to a biodiversity monitoring programme if the methods being used would permit a significant trend of a certain magnitude in an indicator (say, 10% decline) over a specified period (say, over 3 years) to be detected with a desired level of confidence (e.g. 95%). Whether a particular method (e.g. forest bird point counts) will satisfy these criteria depends on the number of sample units, their spatial distribution, the number of sampling occasions on each unit each year, the prevalence in the landscape and detectability of the species of interest, the underlying variance in the state parameter estimated (i.e. density, occupancy, relative abundance), and the level of disaggregation of the indicator. Students with strong statistics skills could use our monitoring data from 2009, 2010, 2011 and 2012 and undertake a power analysis focused on a particular group, e.g. birds, lemurs, reptiles and amphibians, and contrast the indicators derived from GLMM, distance sampling and occupancy models. It would be extremely useful to confirm by power analysis or simulation that our monitoring and data assimilation methods are effective in meeting the aims of the monitoring programme, or to identify ways in which the allocation of effort to occasions or sampling units could be refined to make our monitoring more powerful.
|
|
.jpg)
.jpg)
.JPG)
