INTEGRATED SATELLITE IMAGE PROCESSING FOR ENVIRONMENTAL MONITORING AND FOREST FIRE PREDICTION IN ABAY REGION USING GOOGLE EARTH ENGINE

ABSTRACT

The increasing frequency and severity of natural disasters such as wildfires require the development of robust environmental monitoring and disaster forecasting methodologies. This study presents an integrated approach to processing satellite imagery to obtain critical environmental parameters, including daily temperature, humidity, and wildfire incidence, with a particular focus on the Abay region. Using MODIS and ERA5-Land datasets available through Google^* Earth Engine (GEE), the methodology integrates data collection, preprocessing, extraction and visualization. Key steps include cloud masking of MODIS land surface temperature (LST) data, calculation of relative humidity from ERA5-Land dew point data using the Clausius-Clapeyron equation, and fire detection using MODIS active fire data. The results demonstrate the effectiveness of the methodology by providing detailed daily metrics over the study period, which can further improve forecasting models for disaster management.

АННОТАЦИЯ

Растущая частота и серьезность стихийных бедствий, таких как лесные пожары, требуют разработки надежных методов мониторинга окружающей среды и прогнозирования стихийных бедствий. В этом исследовании представлен комплексный подход к обработке спутниковых снимков для получения критических параметров окружающей среды, включая суточную температуру, влажность и частоту лесных пожаров, с особым акцентом на регион Абай. Используя наборы данных MODIS и ERA5-Land, доступные через Google^* Earth Engine (GEE), методология объединяет сбор, предварительную обработку, извлечение и визуализацию данных. Ключевые шаги включают облачную маскировку данных MODIS о температуре поверхности земли (LST), расчет относительной влажности из данных точки росы ERA5-Land с использованием уравнения Клаузиуса-Клапейрона и обнаружение пожаров с использованием данных MODIS об активных пожарах. Результаты демонстрируют эффективность методологии, предоставляя подробные ежедневные показатели за период исследования, что может дополнительно улучшить модели прогнозирования для управления стихийными бедствиями.

Keywords: satellite images, machine learning, wildfires, remote sensing, SVM, spectral indices.

Ключевые слова: спутниковые снимки, машинное обучение, лесные пожары, дистанционное зондирование, SVM, спектральные индексы.

Introduction

Natural disasters such as wildfires are becoming increasingly frequent and severe, necessitating the development of effective forecasting and monitoring systems. Remote sensing technologies play a vital role in assessing environmental risks and supporting disaster management efforts. Automated processing of satellite data enables timely detection and analysis of hazardous conditions. Studies have shown that certain climatic indicators, particularly land surface temperature (LST) and relative humidity, are closely associated with wildfire risks [1, p. 466]. Fires in Central Asia, especially in Kazakhstan, are more frequent compared to neighboring countries [2, p. 507]. One of the most destructive examples is the large-scale fire in the Abay region in June 2023, which burned over 60,000 hectares of forest and caused significant environmental and economic damage [3, p. 12].

In this context, the present study focuses on the Abay region of Kazakhstan as a case study for the development of a satellite data processing algorithm. The region is characterized by a dry continental climate, frequent summer heatwaves, and a high risk of forest and steppe fires. The study uses remote sensing data to extract key parameters—temperature, humidity, and fire incidence—which can later be used for predictive analytics. The purpose of this work is to create a machine learning model capable of forecasting wildfires based on patterns in recent environmental data.

To achieve the goal, it is necessary to complete the following tasks:

1. Define the region of interest (Abay region) and gather relevant satellite data;
2. Preprocess MODIS and ERA5-Land datasets, including cloud masking and unit conversion;
3. Calculate daily average temperature and relative humidity;
4. Detect fire occurrences based on MODIS active fire data;
5. Compile a structured dataset of environmental parameters;
6. Prepare the data for use in supervised machine learning algorithms for wildfire prediction.

Materials and methods

The study relies on satellite datasets obtained from the Google^* Earth Engine (GEE) platform. Specifically, the Moderate Resolution Imaging Spectroradiometer (MODIS) is used to gather land surface temperature and active fire data, while the ERA5-Land dataset from ECMWF provides dew point temperature. The analysis was geographically restricted to the Abay region using a manually defined polygon created within GEE.

Figure 1. General workflow for forest fire prediction using satellite data

Figure 1 demonstrates the general workflow of the algorithm. It poses raw satellite images undergo preprocessing to extract temperature and humidity features, which are then input into a support vector machine (SVM) model to estimate the probability of wildfire occurrence.

Preprocessing of MODIS data involved cloud masking using the quality assurance (QA) bands to exclude pixels affected by cloud cover. The LST data, originally in Kelvin, was converted to Celsius. Dew point temperatures from ERA5-Land were used to calculate relative humidity using the Clausius-Clapeyron-based empirical formula [6, p. 3070]. Fire detection was based on the MODIS MCD14A1 product, which identifies thermal anomalies corresponding to active fire pixels.

The timeframe for analysis was from May 1 to September 30, 2023, corresponding to the peak fire season in the Abay region. The average daily values of temperature and humidity were calculated using the region reducer function in GEE. Fire occurrence was determined by identifying the presence of thermal hotspots in the region polygon on each day of the study period.

Results and discussions

A total of 151 daily images were analyzed for LST and fire occurrence, while 3624 records were processed for relative humidity calculations. The extracted data includes average temperature, relative humidity index, and binary fire presence indicators. This structured dataset serves as a foundation for training machine learning models.

The results confirm known dependencies between fire risks and climatic factors. For instance, days with high temperatures and low humidity often coincided with fire occurrences. This supports the use of LST and humidity as key predictors, consistent with previous research emphasizing the significance of temperature anomalies and moisture indices in fire forecasting [4, p. 402; 5, p. 2].

Visualization of the data in GEE helped illustrate temporal patterns and geographical clusters of fire activity. The use of Google* Earth Engine enabled high-performance processing of large datasets and accurate extraction of region-specific metrics. This method aligns with the findings of Gorelick et al., who highlight GEE’s utility in planetary-scale environmental analysis [7, p. 19].

Machine learning models, particularly Random Forest classifiers, have been shown to perform well in similar applications. Uddin et al. report that Random Forest achieved the highest prediction accuracy among tested models in the field of medical forecasting, which shares methodological similarities with environmental modeling [8, p. 3]. This suggests that the processed data in this study is well-suited for use in supervised learning frameworks. It was also possible to visualize the obtained data, one example is shown in Figure 2.

Figure 2. Collected and processed data in table format

Conclusion. The study developed a comprehensive workflow for the extraction of key environmental parameters from satellite data, focused on the Abay region of Kazakhstan. By using MODIS and ERA5-Land datasets available through Google* Earth Engine, the research effectively processed and visualized daily values of temperature, humidity, and fire activity for the fire season of 2023.

Preprocessing included cloud masking, unit conversion, and relative humidity calculation using dew point data. Fire detection was successfully implemented using the MCD14A1 dataset. The resulting dataset, comprising temperature, humidity, and fire presence for each day, will be used in further studies to train machine learning models for wildfire prediction. The methodology has strong potential for application across Kazakhstan and other wildfire-prone regions. The scalability and adaptability of the GEE-based approach make it a valuable tool for environmental monitoring and disaster preparedness in diverse climatic zones.

References:

1. Roy, P. S., Behera, M. D., & Srivastav, S. K. (2017). Satellite Remote Sensing: Sensors, Applications and Techniques. Proceedings of the National Academy of Sciences, India Section A: Physical Sciences, 87(4), 465–472. https://doi.org/10.1007/s40010-017-0428-8
2. Yin, H., Jiapaer, G., Jiang, L., Yu, T., Umuhoza, J., & Li, X. (2021). Monitoring fire regimes and assessing their driving factors in Central Asia. Journal of Arid Land, 13(5), 500–515. https://doi.org/10.1007/s40333-021-0008-2
3. Shogelova, N., & Sartin, S. (n.d.). Problems of forest management in Abay Region. CHEMISTRY SCIENCES, 12.
4. Maffei, C., Lindenbergh, R., & Menenti, M. (2021). Combining multi-spectral and thermal remote sensing to predict forest fire characteristics. ISPRS Journal of Photogrammetry and Remote Sensing, 181, 400–412. https://doi.org/10.1016/j.isprsjprs.2021.09.016
5. Duan, S.-B., Han, X.-J., Huang, C., Li, Z.-L., Wu, H., Qian, Y., Gao, M., & Leng, P. (2020). Land Surface Temperature Retrieval from Passive Microwave Satellite Observations: State-of-the-Art and Future Directions. Remote Sensing, 12(16), 2573. https://doi.org/10.3390/rs12162573
6. Ivancic, T. J., & Shaw, S. B. (2016). A US-based analysis of the ability of the Clausius-Clapeyron relationship to explain changes in extreme rainfall with changing temperature. Journal of Geophysical Research: Atmospheres, 121(7), 3066–3078.
7. Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google* Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031
8. Uddin, S., Khan, A., Hossain, M. E., & Moni, M. A. (2019). Comparing different supervised machine learning algorithms for disease prediction. BMC Medical Informatics and Decision Making, 19(1), 281. https://doi.org/10.1186/s12911-019-1004-8

*(По требованию Роскомнадзора информируем, что иностранное лицо, владеющее информационными ресурсами Google является нарушителем законодательства Российской Федерации – прим. ред.)