<p>Fires are one of the most destructive natural disasters and have serious long-term effects on the environment, economy, and human health. In Inner Mongolia Autonomous Region, China, frequent fire disturbance occurs due to the intensification of climate change and human activities. It is crucial to understand the fire regime and estimate the probability of regional fire occurrence and reducing fire losses. However, most studies have primarily focused on the dynamic changes, probability of occurrence, and driving mechanisms of wildfires in the grassland and forest land ecosystems in Inner Mongolia, while insufficient research has been conducted on the spatiotemporal variations in active fires and their impact on the wildfire risk in forest land and grassland. Therefore, in this study, we analyzed the active fire regime based on Moderate Resolution Imaging Spectroradiometer (MODIS) thermal anomalies and burned area products from 2000 to 2022. Combined with climate, topographic, landscape, anthropogenic, and vegetation datasets, logistic regression (LR), support vector machine (SVM), random forest (RF), and convolutional neural network (CNN) models were chosen to estimate the probability of active fire occurrence at the seasonal timescale. The results revealed that: (1) a total of 100,343 active fires occurred in Inner Mongolia and the burned area reached 6.59×10⁴ km². The number of ignition point exhibited a significant increasing trend, while the burned area exhibited a nonsignificant decreasing trend; (2) four active fire belts were detected, namely, the Hetao-Tumochuan Plain fire belt, Xiliao River Plain fire belt, Songnen Plain fire belt, and Hailar River Eroded Plain fire belt. The centroid of the active fires has shifted 456.4 km toward the southwest; (3) RF model achieved the highest accuracy in estimating the probability of active fire occurrence, followed by CNN, and LR and SVM models had lower accuracies; and (4) the distribution of the high and extremely high fire risk areas largely aligned with the four fire belts. The probability of active fire occurrence was the highest in spring, followed by that in autumn, and it gradually decreased in summer and winter. Our results revealed active fires migrated to the southwest and ignition sources increased, despite reduction of the burned area was not significant. The RF model outperformed the other models in predicting the probability of active fire occurrence. These findings contribute to future fire prevention and prediction in Inner Mongolia.</p>