Predicting dominant terrestrial biomes at a global scale using machine learning algorithms, climate variable indices,...

Area of Science:

• Ecology and Environmental Science
• Computational Biology
• Climate Science

Background:

• Global biome distribution is critical for biodiversity conservation, climate modeling, and land-use planning.
• Traditional biome models often simplify climate data, while newer approaches explore extreme climate events.
• The impact of different machine learning techniques and climate data representations on biome modeling accuracy and robustness remains unclear.

Purpose of the Study:

• To evaluate how machine learning algorithm choice, climate data summarization, and extreme climate indices influence global biome model accuracy and robustness.
• To compare the performance of Random Forest (RF), Support Vector Machine (SVM), Naive Bayes (NV), and LeNet Convolutional Neural Network (CNN) for biome modeling.
• To assess the trade-offs between accuracy gains and robustness reductions when incorporating extreme climate data.

Main Methods:

• Four machine learning algorithms (RF, SVM, NV, CNN) were applied to global biome modeling.
• Climate data was analyzed in summarized forms and with the addition of extreme climate indices.
• Model accuracy was evaluated, and robustness was assessed by measuring prediction consistency against observed climate values.

Main Results:

• Random Forest (RF) and Convolutional Neural Network (CNN) achieved the highest accuracy, with CNN showing less overfitting.
• Summarizing climate data reduced accuracy by 1-2%; adding extreme indices increased accuracy by less than 2% (except for NV).
• Extreme climate data led to significant mismatches, reducing prediction consistency and overall model robustness.

Conclusions:

• Convolutional Neural Networks (CNNs) and Random Forests (RFs) are effective algorithms for global biome modeling.
• The inclusion of extreme climate indices provides marginal accuracy benefits but substantially compromises model robustness.
• Caution is advised when integrating extreme climate data into biome prediction models due to potential negative impacts on robustness.