A special problem entitled "Ecological Niche Modeling of Diptero-carp Trees Using Maximum Entropy Methodl" prepared and submitted by Adrian Jose B. Sabado in partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science has been reviewed and is recommended for reception. Accepted and approved as partial fulfillment of the requirements for the Bachelor of Science in Computer Science degree.
A program that could predict the likely distribution of a species over a geographical area has always been a sought-after tool in the field of ecology because of the wide variety of applications it could have or, as it is better known, Niche Modeling . Being one of the most important timber sources in Asia, a suitability tool for dipterocarps would have some economic impact on the Philippines' export industry. The environmental variables used were obtained from PAG ASA, but since localities are mainly from central visayas, only the weather stations surrounding the localities are selected. The model produced has an AUC score of 99.64%. However, it is important to note that the occupancy data and climate data are mainly based on Central Visayas.
For future research purposes, the developed software also includes a function to build your own model using your own set of environmental variables. The software uses Java as its interface and R (with Dismo as its main package) for its backend.
Background of the Study
Currently there are many existing modeling methods, but for this paper we are going to talk about the Maximum Entropy approach or as it is more commonly known, MaxEnt. The best approach to approximating an unknown probability distribution is to ensure that the approximation satisfies any constraints on the unknown distribution that we are aware of, and that subject to those constraints, the distribution should have maximum entropy [8].
Statement of the Problem
Objectives of the Study
The best method for approximating an unknown probability distribution is to ensure that the approximation satisfies all constraints on the unknown distribution that we are aware of, and that the distribution should have maximum entropy, subject to these constraints [8]. presence data and environmental variables, the tool will then output a species distribution showing the habitat suitability of the selected species in the Philippines where it is likely and unlikely to thrive. Be able to read the percentage contribution of each environmental variable to the created model. Be able to output a series of distributions in a GIF-like manner, which would be the series of distributions from the current time up to the entered X years from now.
Significance of the Project
Then, when the forest cover is sustainable, this tool can then be used to determine locations where a dipterocarp tree species is likely to be found and use this information to improve timber harvesting and timber export industries, thereby contributing to the overall improvement of the Philippine economy.
Scope and Limitations
Assumptions
Review of Related Literature 6
The ecological niche of a species can be defined as those ecological conditions under which it can maintain populations without immigration [10]. In the sampled region test, Maxent obtained an AUC (area under the curve) score of 0.733 compared to 0.608 for GARP, suggesting that GARP are less predictive models than Maxent [13]. It requires only species occurrence points, uses continuous and categorical data and interactions between them, is not sensitive to collinearity between environmental variables, resulting in probability distributions.
It has also been shown that Maxent does not assign equal weight to environmental variables (essentially saying that each environmental variable contributes different weights to the model's predictive accuracy) [15], which is true for real-world applications. Later in the article, it is noted that the predictive power of the model improves significantly at only 18-20 unique presences compared to more than 5 presences for Maxent in [16]. But due to the efforts of the government along with the private sectors and NGOs, it has increased and reached 7.20 million ha in 2003.
In projects like this, a tool, especially a niche modeling tool, would help determine the habitat suitability of a species, making the planning and decisions of project planners more informed. Ecological level modeling (ENM) is the process of using reference information, particularly locality or presence data, and a set of environmental variables (also called layers) to produce a prediction model that can be used to predict habitat suitability of incoming species over a geographic. area.
Dipterocarp
The models produced by this process are used to predict the invasion of species, to plan where to place reserve areas, to discover the possible locations of nodes of a specific species.
Maximum Entropy Approach
Design and Implementation 13
After the calculation for the prediction model, the tool will now use this model to create a map visualization of the model, using the difference in colors as an indicator of increasing probability of occurrence.
Flowchart Diagram - Researcher
Flowchart Diagram - AI Expert
Results 17
Apart from the single year prediction tab, the user also gets two more tabs with different functionalities, multi-year prediction and rebuilding. If the user is a researcher, the figure below shows how the researcher could use the system's built-in model to predict the distribution of species along with his/her environmental variables. The filter functionality is initially disabled as this feature would be to filter the results (which have not been produced yet) to a specific region.
Another option available to the user is to predict using their own pre-trained model (using R's Dismo package) along with the same environmental variables. As with the one-year model system, the user also has the option to use the model system for multi-year forecasting. As with the one-year forecast, you can also choose your own model for the multi-year forecast.
Added results to the multi-year forecast would be the graph of probability over time over the currently projected region. Filtering in the multi-year forecast tab will not only update the range of images displayed, but also the graph of probability displayed. If the user wants to see the variable contributions of the produced model, the user can click on the show variable contribution button.
Discussions 24
The software produced in this study is a standalone application that aims to provide researchers with a way to create models that can be used to predict the habitat suitability of Dipterocarp trees. Aside from that, the software also gives the user an existing model that can already be used to predict habitat suitability through data provided by PAG ASA and DENR. The software uses a machine learning algorithm, MaxEnt, to predict the likely species distribution of Dipterocarp trees.
However, keep in mind that the data used for the built-in model only comes from Central Visayas, so predictions using the built-in model are only accurate in the vicinity of the region. The proponent suggests that the built-in model could be improved by collecting more Dipterocarp species presence data nationwide, as well as more climate data and, if possible, soil information (elevation and soil type). A way to see what model can be used based on a set of environmental variables would also be useful, because under current limitations you can only predict using the same type of variables that you used when training your model.
Not having to guess which environment variables are used for which models would save a lot of time and avoid confusion. An online implementation of said software would be more accessible to more researchers, which would benefit more people. In addition to being accessible to more people, it would also be available over a larger area since you don't need to have the software installed on your computer.
Boado, "Incentive Policies and Forest Use in the Philippines," Public Policies and the Abuse of Forest Resources, pp. Aragon, "Restoring Forests at the Landscape Level in the Philippines," Forest Landscape Restoration for Forests in Asia-Pacific, p. G Lohmann, et al., "New methods improve prediction of species distributions from occurrence data," Ecography, vol.
FLINT, “Maxent modeling for predicting the potential distribution of goitered gazelle in central Iran: the effect of size and grain size on model performance”, Turkish Journal of Zoology, p. Mu˜ noz, “Effects of abundance on the reliability and stability of Martian species distribution models: the importance of regional niche variation and ecological heterogeneity,” Journal of Vegetation Science, vol. Silander, “A maxent practical guide to modeling species distribution: what it does, and why inputs and settings matter,” Ecography, vol.
Pushin, “Prediction of geographic distribution and habitat suitability due to climate change of selected threatened forest tree species in the Philippines,” Applied Geography, vol.
