Animal migration poses risks to human health and economic stability, highlighting the need for effective monitoring. Weather radars are essential tools for monitoring migratory insects and birds. While S-band radars can accurately distinguish insect and bird echoes, X-band radar, offering higher resolution, has not been sufficiently explored, limiting its use in aerial ecological monitoring. In this paper, joint observational experiments were conducted to evaluate insect and bird echoes from S-band and X-band weather radars. The results show significant overlap in the polarization features on X-band radar, making existing algorithms unsuitable for X-band data. To address this issue, a point-surface feature fusion method is proposed. This approach extracts polarization variables to construct point-scale features for initial classification with statistical models. A residual network captures surface-scale morphological features, which are integrated with the point-scale recognition results. Finally, a feature fusion module generates the final classification. The method achieves a mean intersection-over-union (mIoU) of 84.56% and demonstrates high accuracy and robustness in historical data tests. This study enhances X-band radar’s ability to differentiate between insect and bird echoes, providing a new solution for aerial ecological monitoring.

Predicting aerial animal migration is of great significance for biological research, ecological conservation, and agricultural production. The mechanism of animal migration is deeply coupled with spatiotemporal and meteorological factors. However, the existing large-scale prediction models using weather radar isolate the spatiotemporal characteristics and the meteorological factors. Additionally, their long-term prediction capabilities are limited, posing challenges in accurately forecasting long-term migration patterns to support applications, such as ecological warnings. This article introduces an aerial migration prediction neural network model combining multiple meteorological factors with weather radar data while expanding the horizon of the migration forecast to the scale of 7 days. Differentiated feature extraction methods are applied to different meteorological factors in the network. The transfer characteristics of the wind field in 2-D space are used to construct a dynamic migration model. The scalar meteorological data are encoded by entity embedding to perform feature fusion with the dynamic branch, collectively forming the forecast model that outputs future migration intensity. We validate the effectiveness of our model China weather radar network real data and reanalysis data, accurately forecasting migratory biomass within China for a horizon of up to 7 days. Moreover, our model is compared with two existing prediction models, demonstrating a maximum improvement of 14.00% in the coefficient of determination ($R^{2}$) in long-term forecast, and the visualized results highlight the predictive effectiveness for the spring and autumn seasons. In future applications, more meteorological factors should be considered and radar data from more stations should be collected to enhance the dataset.

Accurately extracting biological echoes is a fundamental prerequisite for weather radar aeroecology monitoring. However, the concurrent presence of meteorological echoes and biological echoes greatly restricts the extraction accuracy. Traditional neural network-based echo extraction algorithms rely on the spatial continuity feature of the echoes. But, the concurrent presence of multiple types of echoes will lead to the invalidation of the spatial feature and the error of boundary identification of biological echoes. To address this challenge, this study proposes a weak biological echo extraction algorithm using a superpixel technique, aimed at preserving richer biological details in adverse weather conditions. To amplify the imaging distinctions between biological and meteorological components, we design 8-D differential features for each superpixel patch on the CIELAB color space. The gradient boosting tree model is trained for biology classification in handling complex data scenarios. Trained trees exhibit strong generalization capabilities and imbalanced testing data that reflect real weather conditions. To mitigate the limitations posed by the lack of publicly available datasets, we establish a trainable weather radar image dataset encompassing typical weather conditions across national weather radar stations. Experimental results validated that the algorithm retains over 98% of biological data under adverse weather conditions.

Weather radar serves as a crucial tool for monitoring aeroecology by enabling the observation of migrating birds and insects. Although dual-polarization weather radar offers the possibility of classifying echoes, extracting migration echoes of birds and insects from historical single-polarization weather radar data remains challenging. The current deep-learning methods have been successfully extracting aerial migrations from single-polarization weather radar data. However, it still faces challenges in distinguishing between birds and insects at the pixel level, primarily due to the absence of distinct semantic features for each. To tackle this challenge, we propose a semi-supervised radar data processing framework, which generates a large number of single polarization training datasets from a small amount of dual polarization truth data and trains the image segmentation network of single polarization data to distinguish between bird and insect echoes. The framework comprises three components: an image classifier, an image generator, and an image segmentation model. Specifically, the image classifier and image generator leverage a small set of manually annotated dual-polarization radar data to generate the pixel-level single-polarization dataset for training the image segmentation model. The well-trained image segmentation model extracts migration echoes of birds and insects from radar images. Experimental results demonstrate that the proposed method achieves a mean intersection over union (IoU) of 97% for segmenting precipitation, bird, and insect targets. The proposed framework can utilize historical archived single-polarization weather radar data to provide large-scale, long-term, and repeatable monitoring data for birds and insects.

Communal roosts are essential stopover sites for migratory birds. Monitoring the diurnal activity patterns of birds in communal roosts (DAPBCRs) is crucial for understanding their migratory behavior and ecological needs. This information is crucial for guiding habitat conservation and management strategies and assessing the impact of environmental changes on bird populations. Traditional methods for extracting DAPBCR often rely on detecting high reflectivity factor arc features generated by birds collectively leaving or returning to communal roosts using weather radar data and deep learning target detection algorithms. However, many bird echoes do not produce these high reflectivity factor arc features, making pixel-level extraction of DAPBCR challenging. To address this, we propose a method for DAPBCR extraction based on the differences in the probability distribution function (pdf) of differential backscattering phase between birds and insects. This method first removes nonbiological echoes from polarimetric weather radar data, retaining only biological echoes. By calculating the differential backscattering phase using the differential phase and system differential phase, we obtain the pdf of the differential backscattering phase for biological echoes. We fit this pdf to a mixed von Mises distribution to obtain the PDFs for birds and insects. Using posterior probabilities for birds and insects, we estimate the bird-insect mixing ratio and further estimate the number of birds by combining the reflectivity factor and mean radar cross section (RCS) of birds. Applying the proposed method, we extracted DAPBCR data in the midsection of the Huai River Basin near Fuyang City from June to October 2021. We found that bird activity peaked in August and September. Based on normalized cumulative bird activity, we estimated the start, peak, and end times of DAPBCR to be July 6, August 19, and September 30, respectively.

Monitoring and forecasting aerial animal migration benefit biological conservation, aviation safety, and agricultural production. Due to the lack of large-scale observation data and quantitative knowledge of aerial animal migration mechanisms, it is difficult to build a numerical simulation system for migration prediction. However, the extensive deployment of weather radars makes it possible to obtain large-scale aerial migration information. Meanwhile, artificial intelligence technologies provide new insights into the modeling of complex system. In this article, we develop a deep-learning model to predict aerial migration from the perspective of spatio-temporal evolution. Specifically, an undirected graph is applied to describe the geographic structure of the weather radar network, and then graph convolution and gated recurrent unit (GRU) are combined to extract spatio-temporal features of migration information. In addition, a multi-head self-attention mechanism is applied to enhance long-term dependence. Experiments are conducted to validate the effectiveness of the proposed model on the data from the Chinese weather radar network. The results show that our model can achieve state-of-the-art performance among the competing methods. Moreover, improvements from graph convolution and multi-head self-attention are also analyzed. In future applications, more weather radar data will be collected to enrich the dataset and build an aerial migration monitoring and prediction system.