天山山区道路冰雪灾害受复杂地形、局地气象及道路结构共同影响，具有突发性、持续演化性和显著空间差异性。现有数据资源多面向单一图像识别或结构化风险评估，难以支撑从环境感知、风险研判到交通管控和信息发布的连续业务链条。为此，本研究构建了AI-Ready道路冰雪灾害预警多模态大模型微调数据集。\n
文本数据来源于新疆交通公众出行信息服务网2013年3月至2026年1月公开发布的6496条道路气象灾害记录，经标准化处理、语义审核和事件母表构建后，保留5536条有效事件，并据此派生事实抽取、风险研判、管控建议、公众提示、恢复时间研判、应急复盘、道路事件初报生成和跨来源一致性校验8类任务，共形成39909条文本指令微调样本。图像数据经感知去重、算法筛选和人工复核后保留808幅道路图像，覆盖干燥、积雪、结冰、风吹雪、融雪和潮湿6类路面状态，并遵循“可见证据优先”原则构建单图多模态指令样本；其中264幅融雪和潮湿路面图像进一步融合确定性生成的气温、降水量和风速情景参数，并采用四阶多项式模型生成可复算的结冰时间监督标签。最终数据集包含40717条样本，统一采用六字段JSON Lines格式组织。数据质量控制结果表明，事件文本在可追溯性、语义一致性和任务规则约束方面满足微调要求；808组图文样本经人工审校后具有良好的图文一致性，且不同数据子集之间未发现近重复图像。进一步基于Qwen2-VL-2B开展LoRA微调验证，微调模型在道路事件事实抽取任务中的F1值达到0.984，在路面状态识别任务中的准确率和Macro-F1分别达到0.962和0.958，并在应急复盘和道路事件初报生成任务中表现出良好的领域适配性能。该数据集可为高寒山区道路冰雪灾害场景下文本大模型和视觉语言模型的监督微调、任务评测及道路安全预警应用拓展提供可追溯、可复现的多模态数据基础。", "ds_source": "", "ds_process_way": "", "ds_quality": "", "ds_acq_start_time": "2013-03-01 00:00:00", "ds_acq_end_time": "2026-01-01 00:00:00", "ds_acq_place": "新疆,天山山区", "ds_acq_lon_east": null, "ds_acq_lat_south": null, "ds_acq_lon_west": null, "ds_acq_lat_north": null, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "open-access", "ds_total_size": 9026335, "ds_files_count": 0, "ds_format": "*jpg,*jsonl", "ds_space_res": "", "ds_time_res": "2013.03.08-2026.01.03", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "d6307eec-e339-419b-a580-0712757dacf9.jpg", "ds_thumb_from": 0, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "None", "ds_from_station": null, "organization_id": "52b7b79b-860c-49a5-9083-9a70cf8bed5a", "ds_serv_man": null, "ds_serv_phone": null, "ds_serv_mail": null, "doi_value": "", "subject_codes": [ "580" ], "quality_level": 0, "publish_time": "2026-08-20 11:34:24", "last_updated": "2026-08-20 11:34:24", "protected": false, "protected_to": "2027-02-16 00:00:00", "lang": "zh", "cstr": "11738.11.ncdc.llm.db7721.2026", "i18n": { "en": { "title": "An AI-Ready Multimodal Dataset for Fine-Tuning Large Models in Road Ice and Snow Hazard Warning", "ds_format": "*jpg,*jsonl", "ds_source": "", "ds_quality": "", "ds_ref_way": "", "ds_abstract": "
Road ice and snow hazards in the Tianshan Mountains are jointly influenced by complex terrain, localized meteorological conditions, and road infrastructure, and are characterized by sudden onset, continuous evolution, and pronounced spatial variability. Existing datasets are primarily designed for isolated tasks, such as image-based road-surface recognition or structured risk assessment, and therefore provide limited support for the continuous operational chain from environmental perception and risk assessment to traffic control and information dissemination. To address this gap, this study developed an AI-Ready multimodal instruction-tuning dataset for road ice and snow hazard warning in cold mountainous regions. \r\n
The textual data were derived from 6,496 publicly available road weather hazard records released by the Xinjiang Public Travel Information Service Network between March 2013 and January 2026. After data standardization, semantic review, and construction of a master event table, 5,536 valid events were retained and used to derive eight instruction-tuning tasks: fact extraction, risk assessment, traffic-control recommendation, public advisory generation, recovery-time assessment, emergency-response review, initial road-event report generation, and cross-source consistency checking. These tasks yielded 39,909 text instruction-tuning samples. The image data were subjected to perceptual deduplication, algorithm-based screening, and manual review, resulting in 808 road images covering six road-surface conditions: dry, snow-covered, ice-covered, snow-blowing, snow-melting, and wet. Single-image multimodal instruction-tuning samples were then constructed following a visible-evidence-first principle. For 264 snowmelt and wet-road images, deterministically generated scenario parameters for air temperature, precipitation, and wind speed were further incorporated, and reproducible icing-time supervision labels were generated using a fourth-order polynomial model. The final dataset contains 40,717 samples organized in a unified six-field JSON Lines format. Quality-control results showed that the event-text data satisfied the requirements for traceability, semantic consistency, and task-level rule compliance. Manual review of the 808 image-text samples confirmed good image-text consistency, with no near-duplicate images identified across the training, validation, and test subsets. To further evaluate the dataset for domain adaptation, LoRA-based fine-tuning was performed on Qwen2-VL-2B. The fine-tuned model achieved an F1 score of 0.984 for road-event fact extraction and an accuracy of 0.962 and a Macro-F1 of 0.958 for road-surface condition recognition, while also demonstrating good domain adaptation performance in emergency-response review and initial road-event report generation. The dataset provides a traceable and reproducible multimodal data foundation for supervised fine-tuning, task evaluation, and the further development of road safety warning applications using large language models and vision-language models in cold mountainous regions.", "ds_time_res": "", "ds_acq_place": "Xinjiang, Tianshan Mountains", "ds_space_res": "", "ds_projection": "", "ds_process_way": "", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 0, "recommendation_value": 0, "license_type": "https://creativecommons.org/licenses/by/4.0/", "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "belong_to_nieer": false, "allow_update_data": false, "ds_topic_tags": [ "人工智能", "防灾减灾" ], "ds_subject_tags": [ "交通运输工程" ], "ds_class_tags": [], "ds_locus_tags": [ "新疆", "天山" ], "ds_time_tags": [ 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026 ], "ds_contributors": [ { "true_name": "刘景琦", "email": "liujingqi@nieer.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" }, { "true_name": "张智星", "email": "zhangzhixing@sztu.edu.cn", "work_for": "深圳技术大学", "country": "中国" }, { "true_name": "张耀南", "email": "yaonan@lzb.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" }, { "true_name": "任彦润", "email": "renyr@lzb.ac.cn", "work_for": "中科院西北研究院", "country": "中国" }, { "true_name": "康建芳", "email": "kangjf@lzb.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "刘景琦", "email": "liujingqi@nieer.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "ds_managers": [ { "true_name": "李红星", "email": "lihongxing@lzb.ac.cn", "work_for": "中国科学院西北生态环境资源研究院", "country": "中国" } ], "category": "积雪" }