M.Sc thesis (+ a two-month internship):
Agent-based modeling to understand Mediterranean wetland (former saltworks) dynamic based on multiple remote sensing data
UAV imagery over a portion of the study site. Image courtesy Cyril Fleurant (Uni Angers)
The Camargue’s former saltworks is a 6500-ha site located at the Mediterranean coast in southern France. The site has been recently purchased by the Conservatoire du Littoral, a public organization created in 1975 to ensure the protection of outstanding natural areas along the coast. The ongoing management of the area has been entrusted to the natural regional parc (PNR camargue), the national reserve of Camargue and the Tour du Valat. The site comprises a wide range of habitats. It has traditionally been home to the single colony of Flamingos nesting in France and is used by thousands of shorebirds during breeding and migration. Various construction works such as embankments (to control circulation of pumped sea-water through lagoons) and sea-front dike (to prevent uncontrolled flooding by the sea) together with salt exploitation and sea-level rise led to profound changes in the landscape that in turn call for the restoration of natural processes of coastal lagoon ecosystems. However, the conservation and management measures are restricted to be timely done as a result of difficult access for ground survey. Very high resolution remote sensing can introduce alternatives to this by providing continuous and objective surface coverage.
In this context, this M.Sc project aims at developing predictive tools on the basis of remote sensing data to follow habitat dynamics in order to help adaptive ecosystem management. The objective is to develop a method to understand the fast changes of the habitats using very high resolution remote sensing data. To this aim, LiDAR and very high resolution optical data (WorldView 2) and other GIS layers will be analyzed to produce spatially-continuous input for a state-of-the-art agent-based model. Few studies have applied this modeling approach to image analysis but the first results are promising .
Agent-based modeling will allow considering multiple non parametric factors that characterize the landscape dynamics. This approach will allow taking complex spatial and temporal processes as well as changing factors into account. The GAMA agent-based simulation platform (Taillandier et al. 2014, http://gama-platform.org/) was initially developed to integrate GIS data in the simulation. Within the envisaged M.Sc work this platform will be used for prediction based on the layers created from remote sensing data.
The M.Sc thesis is planned to be ideally started with a preliminary phase of two-month internship at the LETG, University of Angers . During the internship the M.Sc student will encompass a NetLogo and GAMA learning phase and gets to know the area and data. A site visit at Tour du Valat research centre may help to understand the management objective of the area. The second phase would be the M.Sc thesis, during which the candidate will spend time at both Universities of Würzburg (4 months) and Angers (2 months). The stay in Angers is supported by an existing ERASMUS agreement between the two universities.
Interested candidates are wellcome to send an Email to Dr. Hooman Latifi.
Dr. Aurélie Davranche (University of Angers, France)
Dr. Hooman Latifi (University of Würzburg)
Dr. Brigitte Poulin (Tour du Valat, France)
In a recently-published paper in Forestry featuring Hooman Latifi, Steven Hill and Stefan Dech from the Dept. of Remote Sensing, further advancements have been reported in developing unbiased statistical models for area-based estimation of forest understorey layers using LiDAR point cloud information. The study leveraged an original high-density LiDAR point cloud, which was further processed to simulate two lower-density datasets by applying a thining approach. The data were then combnined with three statistical modeling approaches to estimate the proportions of shrub, herb and moss layers in temperate forest stands in southeastern Germany.
Despite the differences between our simulated data and the real-world LiDAR point clouds
of different point densities, the results of this study are thought to mostly reflect how LiDAR and forest habitat data can be combined for deriving ecologically relevant information on temperate forest understorey vegetation layers. This, in turn, increases the applicability of prediction results for overarching aims such as forest and wildlife management.
Further informaiton on the published paper can be retrieved here.
Latifi, H., Hill, S., Schumann, B., Heurich, M., Dech, S. 2017. Multi-model estimation of understorey shrub, herb and moss cover in temperate forest stands by laser scanner data. Forestry, DOI:10.1093/forestry/cpw066
In a newly-published paper featuring Steven Hill and Hooman Latifi from Dept. of Remote Sensing, very high resolution remote sensing (laser scanner data and aerial orthophotos) were used in a full remote sensing-based framework to study post-disturbance tree and stand development, particularly in its early seral stages.
Future stand development on test sites 1–5 simulated for a period of 80 years. A) Number of trees (N) per test site. B) Basal area (BA) of trees per hectare. C) Mean tree height (MH). D) Tree height variation (MAD = mean absolute deviation).
The first step involved extraction of single trees and their allometric attributes form LiDAR-based canopy height models, after which the extracted tree locations were additionally validated by a sample based scheme implemented on aerial photos. The single tree based forest growth simulator SILVA ver. 2.2 was then used to simulate the stand development during a 80 year simulation period. In addition, landscape and spatial point pattern metrics were calculated to assess the structural heterogeneity. The results approve that natural regeneration of post disturbed forest stands reveal structural heterogeneity even at the early-seral stages. Furthermore, the study showed that the structural heterogeneity might already be determined in the early successional stages. following the bark beetle disturbances. This study open up interesting horizons in how remote sensing data and methods can be combined with spatial statistics to investigate early-phase forest dynamics in natural stands.
Further information on the published material can be found here.
Hill, S., Latifi, H., Heurich, M., Müller, J. 2017. Individual-tree- and stand-based development following natural disturbance in a heterogeneously structured forest: a LiDAR-based approach. Ecological Informatics 38, 12-25. DOI: dx.doi.org/10.1016/j.ecoinf.2016.12.004
A M.Sc thesis was written by Bastian Schumann under the supervision of Dr. Hooman Latifi and Prof. Christopher Conrad that focused on a LiDAR-based approach to combine structural metrics and forest habitat informaiton for causal and predictive models of understory canopy cover. The data base used consisted of a bi-temporal LiDAR dataset as well as two field datasets and two habitat maps. The entire data were initially edited, revealing that a bi-temporal treatment is only possible for understory layers. The statistical models used for modelling canopy cover density included random forest, logistic models and zero-and-one inflated beta regression.
The results revealed the most relevant LiDAR metrics which contribute to explain the canopy cover density. Furthermore it indicates that the habitat types have a significant influence on canopy cover density. In addition, it was shown that with the use of a denser point cloud a higher performance can be achieved in almost every vertical stand layer.
Wall-to-wall predictions of understory canopy cover usign high density point cloud, habitat types and a logistic model
A M.Sc thesis by Raja Ram Aryal at the University of Applied Sciences Stuttgart was recently written under the supervision of Dr. Hooman Latifi and Prof. Michael Hahn. The thesis focused on a comparative study on the variations of an adaptive TIN ground filtering algorithm to extract DTM from discrete LiDAR point cloud captured in leaf-off and full wave LiDAR point cloud collected in leaf-on conditions. In addition Analysis of Variance (ANOVA) type II was used to assess the influential factors that are related to DTM random error. The Accuracy assessment of extracted DTMs was done at local and landscape levels in heterogeneous forest stands of Bavarian Forest National Park. The DTM generated using mirror points in leaf-off returned less RMSE (0.844 m) than in leaf-on (0.988 m) conditions. Furthermore RMSE values of 0.916 m (leaf-off) and 1.078 m (leaf-on) were observed the local level analysis when no mirror points were used. However, RMSE value of ca. 0.5 m was observed at the landscape level, with leaf-off DTM showing slightly higher error than leaf-on DTM. The DTM error increased with increasing slope. Deciduous habitat was found to significantly influence DTM error in both leaf-off and leaf-on conditions. Interaction effects were mainly observed between slope and forest habitat type.
DTMs Extracted using denser point cloud LiDAR data (leaf-on condition) using mirror points (left side) and without using mirror (right side)