University of Twente Student Theses


Species Distribution Modelling : A Multimodal Learning Approach

Velmurugan, Pranesh (2023) Species Distribution Modelling : A Multimodal Learning Approach.

[img] PDF
Abstract:A species distribution model (SDM) makes use of environmental factors at a location to predict whether one species, or potentially several, will be present there. Some SDMs could even predict the count of the species present there. This work aims to predict the count of Anura (frogs) present in a location by building a SDM based on multiple modalities of data. Eventually, these will provide us valuable information about the condition of the environment. While there has been many methods of building SDMs, this work specifically aims to build a SDM based on Multimodal Learning which takes as input environmental features not just from one data type but from multiple modalities to predict the presence of the species. This work describes the proposed architecture and evaluates the results obtained from the model. Moreover, this paper compares the results obtained from the proposed model to the existing State-of-the-art methods. The fusion architecture proposed in this model makes use of both tabular data and satellite image data. The results evaluated are compared with the winner of the frog counting challenge [6]. According to the leader board of the challenge, the proposed work achieved a F1-score of 0.36, which is placed second, and the winner of the challenge achieved a score of 0.42. Apart from the task of counting frogs, this work also performs the classification of a location as presence / absence. The best performing model achieved an accuracy of 89.19% and an AUC score of 0.96. Though there are no direct comparison available for this task, still the results are on par with the existing classification SDMs. For the task of classifying the location as presence/absence, a novel method of generating pseudo-absence dataset has been presented and is compared with some of the existing methods. The proposed method performed better than the distance criteria method by almost 4% better accuracy and by 19% better accuracy than the random selection method. Overall, this work provides ways to use multiple modalities of data in building a SDM and suggests ways to improve the performance further.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:31 mathematics, 43 environmental science, 54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page