Species Distribution Modeling in Scikit Learn

Species Distribution Modeling (SDM) is a crucial tool in conservation biology, ecology, and related fields. It involves predicting the geographic distribution of species based on environmental variables and species occurrence data. This article explores how to implement SDM using Scikit-Learn, a popular machine learning library in Python.

Introduction to Species Distribution Modeling

Species Distribution Models (SDMs) predict the spatial distribution of species by correlating species occurrence data with environmental variables. This correlation enables scientists to infer where species are likely to be found based on the environmental characteristics of a given area.

These models are essential for understanding species habitats, planning conservation efforts, and studying the impacts of climate change on biodiversity.

• Species Distribution Modeling (SDM) is a pivotal tool in ecology and conservation biology, allowing researchers to anticipate and map the spatial distribution of species across landscapes.
• By integrating species occurrence data with environmental variables such as temperature, precipitation, elevation, and land cover, SDMs unveil the ecological niches and habitat preferences of organisms.

Why Use Scikit-Learn for SDM?

Scikit-Learn offers a robust set of tools for machine learning, including various algorithms that can be applied to SDM. Its ease of use, extensive documentation, and active community make it an excellent choice for implementing SDMs.

Workflow for Species Distribution Modeling

The typical workflow for SDM in Scikit-Learn involves several steps:

1. Data Collection: Gather species occurrence data and environmental variables.
2. Data Preprocessing: Clean and prepare the data for modeling.
3. Model Training: Train a machine learning model using the prepared data.
4. Model Evaluation: Assess the model's performance using appropriate metrics.
5. Prediction and Mapping: Use the model to predict species distribution and visualize the results.

Step-by-Step Guide for Building an Species Distribution Model

Let's create a Species Distribution Model (SDM) using a dataset from Kaggle, we need to select a dataset that is relatively small in size (in kilobytes). Based on the provided search results, the "Bird Sightings Dataset" from Kaggle seems to be a suitable choice as it includes information on different bird species, their locations, dates, and times of sighting, as well as descriptions of the birds.

Step 1: Load Necessary Libraries

import pandas as pd
from sklearn.preprocessing import StandardScaler, LabelEncoder, OneHotEncoder
from sklearn.impute import SimpleImputer
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.svm import OneClassSVM
from sklearn.metrics import roc_auc_score
from sklearn.preprocessing import label_binarize
import matplotlib.pyplot as plt

Step 2: Load and inspect the dataset 

data = pd.read_csv('birdsoftheworld-unprocessed.csv')
print(data.columns)

Output:

Index(['species', 'location', 'time', 'description of bird', 'sex',
       'feather color', 'Unnamed: 6', 'Unnamed: 7', 'Unnamed: 8', 'Unnamed: 9',
       'Unnamed: 10', 'Unnamed: 11'],
      dtype='object')

Step 3: Data Preprocessing

We'll use the 'location' feature and other relevant features. We will need to encode categorical features and handle any missing values.

 Select relevant columns
features = data[['location', 'sex', 'feather color']]
labels = data['species']

# Handle missing values and encode categorical features
preprocessor = ColumnTransformer(
    transformers=[
        ('cat', Pipeline([
            ('imputer', SimpleImputer(strategy='most_frequent')),
            ('encoder', OneHotEncoder(handle_unknown='ignore'))
        ]), ['location', 'sex', 'feather color'])
    ])

# Standardize the features
pipeline = Pipeline(steps=[
    ('preprocessor', preprocessor),
    ('scaler', StandardScaler(with_mean=False))
])

features_processed = pipeline.fit_transform(features)

# Encode labels
label_encoder = LabelEncoder()
labels_encoded = label_encoder.fit_transform(labels)

Step 4: Model Training

Train a One-Class SVM model to predict species distribution

model = OneClassSVM(kernel='rbf', gamma=0.1, nu=0.1)
model.fit(features_processed)

Output:

OneClassSVM
OneClassSVM(gamma=0.1, nu=0.1)

Step 5: Model Evaluation

Evaluate the model using the Area Under the ROC Curve (AUC) metric for multi-class classification.

labels_binarized = label_binarize(labels_encoded, classes=range(len(label_encoder.classes_)))

# Predict the species distribution
predictions = model.decision_function(features_processed)

# Reshape predictions to match the shape of labels_binarized
predictions_reshaped = predictions.reshape(-1, 1)

auc_score = roc_auc_score(labels_binarized, predictions_reshaped, average='macro', multi_class='ovr')
print(f'Area under the ROC curve: {auc_score:.4f}')

Output:

Area under the ROC curve: 0.0038

Step 6: Prediction and Mapping

Since we don't have geographic coordinates, we will visualize the predictions using a simple scatter plot.

# Predict the species distribution
predictions = model.predict(features_processed)

plt.figure(figsize=(10, 6))
plt.scatter(range(len(predictions)), predictions, c=predictions, cmap='coolwarm', alpha=0.5)
plt.title('Bird Species Distribution Predictions')
plt.xlabel('Sample Index')
plt.ylabel('Prediction')
plt.show()

Output:

The scatter plot provides a clear visualization of the model's binary predictions for bird species distribution. The distinct separation between the two clusters of points indicates that the model is making confident predictions. This visualization is valuable for understanding species distribution patterns and informing conservation efforts.

• The prediction values are binary, with -1.0 indicating one class (likely absence or a negative prediction) and 1.0 indicating another class (likely presence or a positive prediction).
• Each point on the x-axis corresponds to a different sample or observation in the dataset.
• The plot shows two distinct clusters of points: one at y = -1.0 (blue) and another at y = 1.0 (red). This indicates that the model has made clear binary predictions for each sample, classifying them into two distinct groups.

Conclusion

Species Distribution Modeling is a powerful tool for understanding and conserving biodiversity. Scikit-Learn provides a flexible and efficient framework for implementing SDMs. By following the workflow outlined in this article, you can leverage Scikit-Learn's machine learning capabilities to predict and visualize species distributions.