Complete and high-resolution temperature observation data are important input parameters for agrometeorological disaster monitoring and ecosystem modelling. Due to the limitation of field meteorological observation conditions, observation data are commonly missing, and an appropriate data imputation method is necessary in meteorological data applications. In this paper, we focus on filling long gaps in meteorological observation data at field sites. A deep learning-based model, BiLSTM-I, is proposed to impute missing half-hourly temperature observations with high accuracy by considering temperature observations obtained manually at a low frequency. An encoder-decoder structure is adopted by BiLSTM-I, which is conducive to fully learning the potential distribution pattern of data. In addition, the BiLSTM-I model error function incorporates the difference between the final estimates and true observations. Therefore, the error function evaluates the imputation results more directly, and the model convergence error and the imputation accuracy are directly related, thus ensuring that the imputation error can be minimized at the time the model converges. The experimental analysis results show that the BiLSTM-I model designed in this paper is superior to other methods. For a test set with a time interval gap of 30 days, or a time interval gap of 60 days, the root mean square errors (RMSEs) remain stable, indicating the model’s excellent generalization ability for different missing value gaps. Although the model is only applied to temperature data imputation in this study, it also has the potential to be applied to other meteorological dataset-filling scenarios.
With the development of cities, the relationship between cities is becoming closer, and the study of urban livability based on a single city can no longer meet the guidelines and suggestions for urban agglomerations. A scientific evaluation of livability in urban agglomerations can better help cities to recognize the advantages and disadvantages. However, most studies on urban livability focus on its connotation and history and neglect simulations and analyses of the future. Based on the Yangtze River Delta agglomeration, this paper establishes an index system using data from 2011 to 2019 to simulate urban livability from 2020 to 2025 through the ARIMA model and analyzes the historical and future data by using GIS methods. The results show the following: (1) The ARIMA model has good simulation accuracy when applied to urban livability analysis and can provide a reference for future urban livability development. (2) The urban livability of the Yangtze River Delta agglomeration has obviously changed both on the whole and in subsystems. Cities in the upper ranking of livability have developed rapidly, and the difference in urban livability has increased. (3) The spatial autocorrelation of urban livability in the Yangtze River Delta agglomeration is obvious both on the whole and in subsystems. (4) The influencing factors of urban livability development are diverse. The general public budget expenditure for social security and employment, fixed assets investment in municipal public facilities, total retail sales of consumer goods, and education and medical expenditures have positive effects on the development of urban livability, while industrial emissions have a negative effect. The results show that cities should strengthen inter-city relationships, promote the coordinated development of inter-regional cities, and formulate relevant policies to improve the level of urban environmental governance in the region.
Challenges in enhancing the multiclass segmentation of remotely sensed data include expensive and scarce labeled samples, complex geo-surface scenes, and resulting biases. The intricate nature of geographical surfaces, comprising varying elements and features, introduces significant complexity to the task of segmentation. The limited label data used to train segmentation models may exhibit biases due to imbalances or the inadequate representation of certain surface types or features. For applications like land use/cover monitoring, the assumption of evenly distributed simple random sampling may be not satisfied due to spatial stratified heterogeneity, introducing biases that can adversely impact the model’s ability to generalize effectively across diverse geographical areas. We introduced two statistical indicators to encode the complexity of geo-features under multiclass scenes and designed a corresponding optimal sampling scheme to select representative samples to reduce sampling bias during machine learning model training, especially that of deep learning models. The results of the complexity scores showed that the entropy-based and gray-based indicators effectively detected the complexity from geo-surface scenes: the entropy-based indicator was sensitive to the boundaries of different classes and the contours of geographical objects, while the Moran’s I indicator had a better performance in identifying the spatial structure information of geographical objects in remote sensing images. According to the complexity scores, the optimal sampling methods appropriately adapted the distribution of the training samples to the geo-context and enhanced their representativeness relative to the population. The single-score optimal sampling method achieved the highest improvement in DeepLab-V3 (increasing pixel accuracy by 0.3% and MIoU by 5.5%), and the multi-score optimal sampling method achieved the highest improvement in SegFormer (increasing ACC by 0.2% and MIoU by 2.4%). These findings carry significant implications for quantifying the complexity of geo-surface scenes and hence can enhance the semantic segmentation of high-resolution remote sensing images with less sampling bias.
Deep learning has exhibited potential in land use and land cover (LULC) classification applications. However, the effectiveness of deep learning remains constrained by the availability and quality of annotated training data. The persistent scarcity of labeled samples and spatial heterogeneity of remote sensing imagery hinder the robustness and generalization of trained models. Meta-learning, commonly referred to as “learning to learn”, is a paradigm that trains models over a distribution of tasks to acquire transferable knowledge, enabling rapid adaptation to new tasks with only a few labeled samples. This cross-task learning capability makes meta-learning a promising solution to data scarcity and spatial heterogeneity in the remote sensing context. This paper provides a systematic review of meta-learning applications in LULC classification, identifying a total of 70 relevant studies between 2018 and 2025. Three mainstream meta-learning paradigms (memory-augmented, optimization-based, and metric-based) are reviewed, and the applications are analyzed across four core challenges in LULC remote sensing: label scarcity, cross-region and cross-domain distribution shifts, temporal dynamics modeling, and multimodal data integration. The review reveals that optimization-based and metric-based methods dominate current research, with MAML and its variants being the most widely adopted due to the model-agnostic property, while memory-augmented methods remain underexplored. A consistent finding is that meta-learning outperforms conventional pre-training followed by fine-tuning under significant domain shifts across multiple data modalities. Current limitations, including computational overhead, episodic training constraints, and the lack of standardized evaluation protocols, are discussed. Future directions in cross-domain generalization, integration with foundation models, novel architectures, and standardized benchmarks are identified.
The Huang-Huai-Hai Plain is one of the typical agri-ecosystems in China, which suffers from cold damage frequently resulting in substantial economic losses. In order to monitor the changes in the occurrence of cold damage in an effective and large-scale manner, and to determine their meteorological causes, this paper collected low temperature data from the agricultural meteorological stations and remote sensing data of MODIS from 2005 to 2015, and constructed a monitoring model of cold damage to winter wheat in Huang-Huai-Hai Plain based on the Logistic regression model. This model was used to analyze the spatio-temporal changes of cold damage of winter wheat in Huang-Huai-Hai Plain from 2011 to 2020, and correlation analysis was performed with the spatio-temporal changes of meteorological factors to ascertain how they affect cold damage. The results show that the harm from cold damage in winter wheat has been gradually decreasing from 2011 to 2020, and the cold damage areas with high probability and high frequency are moving from north to south. The meteorological elements with the greatest impacts on the degree of cold damage from stronger to weaker are heat, precipitation and sunshine duration, whose influence has spatial variability.