Forest fires harm forests and endanger lives. Susceptibility assessment mapping using remote sensing and geographic information systems can help prevent forest fires and allocate fire-fighting resources rationally. This study combines machine learning (ML) and multi-criteria decision analysis to propose LOGEAHP – an integrated method based on logistic regression and the analytic hierarchy process (AHP). This method is used to calculate the relative weights of forest fire influencing factors and generate a forest fire susceptibility map for the Greater Khingan Range. The results show that nearly one-third of the Greater Khingan Range faced severe high forest fire risks between 2013–2024, with higher fire risks in the northern and southern regions than those in the central region. “Distance from Road” is the most critical factor affecting forest fire risks in the Greater Khingan Range, with a feature value as high as 1.1287. The LOGEAHP exhibits excellent fire assessment performance and stability, with an average receiver operating characteristic (ROC) of 0.9311 and 95% confidence interval for ROC ranging from 0.91 to 0.95. Bootstrap validation indicates that LOGEAHP achieves a statistically significant improvement over ML and AHP, with p-value significantly less than 0.01. This approach provides a robust basis for local fire prevention strategies.

ABSTRACT Monitoring riparian vegetation with up-to-date technologies, such as remote sensing and machine learning, is emphasized by their extensive and dynamic nature. This study evaluates the spatiotemporal dynamics of vegetation in the Slovak Danube riparian zone from 2004 to 2022 using a multi-sensor approach (Landsat-5/8 and Sentinel-2). We integrated seasonal vegetation indices, textural, and morphological patterns to enhance Land Cover (LC) classification accuracy. Using Random Forest within a spectral–spatial framework, we achieved notable improvement in classification accuracy. We evaluated ecological processes, spatial errors, misclassification, and the spatial distribution of LC changes within the riparian zone. Our core-area analysis revealed that only 52% of the riparian zone remains stable over time. Crucially, we quantified the impact of spatial misclassifications at class boundaries, finding that edge-related errors account for 0.24% to 2.85% of the detected change. The classification distinguished 1,076 ha of managed plantations and 1,236 ha of natural forest. Overall, the results demonstrated the effectiveness of various feature combinations for classifying different satellite datasets and assessing the temporal evolution of riparian zones in regulated river systems, while highlighting the critical importance of incorporating uncertainty analysis to ensure robust interpretation of detected LC changes.

This paper aims to study the spatiotemporal variations in PM2.5 concentrations between 2015 and 2020 in the Fen–Wei Plain. The AOD products of MCD19A2 based on MAIAC were utilized. Besides of AOD, some of its arithmetic values were also introduced to determine the best variable. AOD was determined as the variable for retrieval because of its highest correlation coefficient with PM2.5, reaching up to 0.526**. Several statistical models and SVR have been used and compared for establishing the best retrieval model. Results showed, the determination coefficient r 2 of the cubic function was the best (r 2 = 0.294**), and the correlation between the predicted and measured values was strong (k = 1.1478, r 2 = 0.532**). Both monthly and annual variations of PM2.5 concentrations were researched, which indicated the changing trend was fluctuating. The mean value in 2015 was the largest, reaching up to 38.02 μg/m3, the smallest value of 28.28 μg/m3 appeared in 2019. It can be concluded the influencing mechanism of PM2.5 was complicated, as proven by correlation analysis and gray relational analysis. The influencing factors should be introduced as auxiliary data to promote the retrieval accuracy. With advancements in remote sensing technology, the monitoring efficiency can be effectively improved, which will greatly promote sustainable development.

ABSTRACT China’s new urbanization and rapid social economic development have led to a significant increase in carbon emissions, while their precise monitoring plays a crucial role in carbon emission reduction and assessment of sustainable development goals (SDGs). For the shortcomings of the temporal lag and insufficient spatial scale of statistical data on CO2 emissions, as well as the low estimation accuracy, this study considers the differences in CO2 emissions among different industries and sectors, designs a spatiotemporal fine monitoring method for Beijing’s CO2 emissions from 2015 to 2020, combining with NPP-VIIRS NTL images, point of interesting (POI) data, grid population and residential land data. The distribution characteristics of energy consumption CO2 emissions in lit/unlit areas of 35 industries and residential areas, such as agriculture, forestry, mining and metal ore, are analyzed. The CO2 emissions in lit areas account for a larger proportion, up to 99%, while in unlit areas is relatively small at 1%. The density of population and residential areas in central urban areas will produce more CO2 emissions. This study can help to formulate appropriate emission reduction policies, optimize the energy structure, and build a green and low-CO2 economic development pattern.

ABSTRACT As monthly rainfall is a long-term and complex process, this study utilized the characteristics of the sliding window to replace the current month’s data with the multi-month rainfall data set within the sliding window for predictive analysis. At the same time, a CNN-LSTM model was used to assist in the prediction of monthly rainfall, and the results were compared with those obtained using LSTM and GRU. Due to the new division of data through the sliding window method, the input data dimension to the CNN-LSTM model was relatively small. Therefore, data augmentation was employed to enhance the overall predictive ability of the model. This study selected 43 years (1980–2022) of meteorological data from the Yichun Station and Zhangye Station in NOAA. The following evaluation indicators were used to assess the model’s performance: NSE, RSR, RMSE, and R. The results demonstrated that the predictive ability of the CNN-LSTM model was superior to that of LTSM and GRU. Specifically, R and NSE increased significantly (20%–30%), while RSR and RMSE decreased slightly (6%–13%). The CNN-LSTM model proposed in this study, based on the sliding window, enabling more accurate predictions of monthly rainfall.

ABSTRACT Global food security is challenged by population growth and limited agricultural land. Intensifying existing cropland use is crucial for increasing agricultural production. Mapping cropping intensity, defined by annual crop cycles and commonly classified as single, double, or triple cropping, is vital for food production modeling. While most studies focus on regions with higher cropping intenisity, this study targets Vojvodina in northern Serbia, where double cropping is less prevalent. We investigated the impact of different vegetation indices (VIs) on detecting double cropping using machine learning (ML) and Sentinel-2 imagery with collected ground truth data over two years with contrasting weather conditions: one dry and one with above-average rainfall. Our approach improves existing methods by integrating VIs not previously examined in related studies. Alongside NDVI, indices such as CVI, VARI, and ExG significantly improved model performance. In the dry year (2022), overall accuracy reached 95.80%, with an F1-score of 91.19% for classifying double cropping, while in the wet year (2023), the accuracy was 93.56% and the F1-score 84.96%. This approach simplifies data requirements while maintaining high accuracy, making it applicable to regions with similar geographical characteristics. Additionally, this study fills a key knowledge gap on the extent and distribution of this practice in Serbia.

ABSTRACT This research aims to evaluate the performance of machine learning algorithms and multispectral aerial images in estimating wheat grain yield, contributing to the eradication of hunger and food security. Two sampling sites with different cultivation periods were used in this study, with multiple aerial flights conducted throughout the phenological cycle. At the end of the experiment, grain yield (t/ha) was determined. The tested supervised machine learning algorithms included Linear Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN), combined with vegetation indices from the visible spectrum (RGB), multispectral indices, and bands. The Linear Regression algorithm, combined with the RGB indices, showed the best performance in the initial and final phases of the crop, with coefficients of determination (R2) of 0.61 and 0.58, respectively. The most robust performance was observed during the booting-heading phase, where the SVM algorithm, combined with the red band, red edge and green band, achieved an R2 of 0.78 and a root mean squared error (RMSE) of 0.479. t/ha. Therefore, the application of these variables and algorithms to estimate wheat productivity proves to be a viable approach, offering an efficient method to predict grain productivity, especially in the southern region of Brazil.

ABSTRACT Efficient and precise extraction of forest information is crucial for assessing the effectiveness of ecological projects and informing ecological policy adjustments in the Gansu section of the Yellow River Basin, China. Utilizing PALSAR (Phased Array L-band Synthetic Aperture Radar) and Landsat data, this study gathered forest/non-forest samples and formulated classification rules using the backscatter coefficients from PALSAR’s HH, HV, HH-HV, and HH/HV polarizations to generate monthly 30-meter forest data. These results were further refined using Landsat’s NDVImax and B7max indices. Our findings are as follows: The forest/non-forest data extracted from PALSAR and Landsat aligns more closely with publicly available forest data, achieving an overall accuracy of 95.4% and a Kappa coefficient of 0.87, which demonstrates the reliability and accuracy of our extraction method. Driven by ecological engineering projects and climate factors, the forest area has expanded significantly, from 13,200 square kilometers in 2008 to 19,800 square kilometers in 2018, with an average annual growth rate of 5%. Approximately 94.09% of the forest area is located at elevations up to 3,500 meters, with stable and increasing forest types predominantly found on the northern and western slopes, spanning elevations between 3,000 and 3,500 meters and slopes ranging from 20° to 35°. Our study provides valuable insights into regions characterized by rugged terrain and low forest density.