Estimation and analysis of missing temperature data in high altitude and snow-dominated regions using various machine...

Area of Science:

• Environmental science and climatology.
• Application of artificial intelligence in meteorology.

Background:

• Accurate temperature data is crucial for managing limited natural resources.
• Northeastern Turkey's mountainous regions present unique climate challenges for data analysis.

Purpose of the Study:

• To evaluate the performance of artificial neural network (ANN), support vector regression (SVR), and regression tree (RT) methods for daily average temperature prediction.
• To identify the most suitable machine learning models for cold, mountainous climates.

Main Methods:

• Analysis of daily average temperature data from 2019-2021 across eight meteorological stations in northeastern Turkey.
• Comparative evaluation of ANN, SVR, and RT models using statistical criteria and Taylor diagrams.
• Focus on model performance at extreme temperatures (high >15°C and low <0°C).

Main Results:

• ANN (specifically ANN6, ANN12) and SVR (medium gaussian, linear) demonstrated high accuracy in temperature estimation, particularly at extreme temperatures.
• All tested methodologies achieved high performance (Nash-Sutcliffe Efficiency-R² > 0.90).
• Snowfall in the -1°C to 5°C range caused minor deviations in estimations due to altered ground heat emission.

Conclusions:

• Machine learning models, particularly ANN and SVR, are highly effective for predicting daily average temperatures in challenging cold, mountainous environments.
• Model architecture, specifically the number of layers in ANNs with high neuron counts, significantly impacts estimation accuracy.
• Further research may refine models to account for snow-induced thermal variations.