Weather Ensemble Forecast
Compare weather forecasts from up to 9 independent global weather models side-by-side.
No setup needed -- just install and run /forecast nyc.
What this skill does
- - Fetches high temperature forecasts from multiple weather models: GFS, ECMWF, JMA, GEM, MeteoFrance, UKMO, ICON, ARPEGE, GraphCast
- Shows each model's individual prediction so you can see where they agree and disagree
- Calculates ensemble standard deviation (how uncertain the forecast is)
- Cross-validates with NWS (National Weather Service) official forecasts
- Covers 16 cities: NYC, Chicago, Miami, Dallas, Atlanta, Seattle, Toronto, London, Paris, Munich, Seoul, Ankara, Wellington, Lucknow, Buenos Aires, Sao Paulo
Quick Start
CODEBLOCK0
That's it. Works immediately with free tier (NYC, Chicago, Miami).
For full access (all 16 cities), set an API key:
CODEBLOCK1
Then set: INLINECODE1
Commands
/forecast <city> - Get multi-model forecast
Compare forecasts from all available weather models for a city.
When the user asks for a weather forecast, temperature prediction, model comparison, or ensemble analysis, run:
CODEBLOCK2
Where <city> is a city key (e.g., nyc, london, seoul). Run /cities to see all options.
And <date> is optional, in YYYY-MM-DD format (defaults to tomorrow).
Display results showing each model's prediction, the ensemble average, and standard deviation.
/cities - List available cities
CODEBLOCK3
Use Cases
- - Weather enthusiasts: See how GFS, ECMWF, and other models compare for your city
- Event planners: Check forecast confidence before outdoor events
- Researchers: Access raw multi-model data for analysis
- Travelers: Get high-confidence forecasts when models agree, warnings when they don't
- Prediction market traders: Use ensemble data to evaluate temperature market positions (see polymarket-weather-scanner skill)
How It Works
- 1. Queries the Weather Ensemble API for forecasts from each weather model
- Returns individual model predictions so you can see the spread
- Calculates ensemble statistics: mean, standard deviation
- Higher standard deviation = more uncertainty = models disagree
- Low standard deviation = high confidence = models agree
External Endpoints
All requests are sent to the Weather Ensemble API server.
| Endpoint | Data Sent | Purpose |
|---|
| INLINECODE7 | City, date, API key (if set) | Get forecast |
| INLINECODE8 |
API key (if set) | List cities |
Security & Privacy
- - Works without an API key (anonymous free tier with IP-based rate limiting)
- All requests are sent to the Weather Ensemble API server (
polymarket-scanner.fly.dev) via HTTPS - Weather data is sourced from Open-Meteo and aggregated server-side
- No personal data collected or stored
天气集合预报
将最多9个独立全球天气模型的预报并排比较。
无需设置——只需安装并运行 /forecast nyc。
该技能的功能
- - 从多个天气模型获取高温预报：GFS、ECMWF、JMA、GEM、MeteoFrance、UKMO、ICON、ARPEGE、GraphCast
- 显示每个模型的独立预测，以便查看它们的一致与分歧之处
- 计算集合标准差（预报的不确定性程度）
- 与美国国家气象局（NWS）官方预报进行交叉验证
- 覆盖16个城市：纽约、芝加哥、迈阿密、达拉斯、亚特兰大、西雅图、多伦多、伦敦、巴黎、慕尼黑、首尔、安卡拉、惠灵顿、勒克瑙、布宜诺斯艾利斯、圣保罗
快速开始
/forecast nyc
就这样。免费层级（纽约、芝加哥、迈阿密）可立即使用。
如需完整访问（全部16个城市），请设置API密钥：
curl -X POST https://polymarket-scanner.fly.dev/keys/free
然后设置：export WEATHERENSEMBLEAPIKEY=pmyourkeyhere
命令
/forecast <城市> - 获取多模型预报
比较某个城市所有可用天气模型的预报。
当用户请求天气预报、温度预测、模型比较或集合分析时，运行：
bash
bash {baseDir}/scripts/forecast.sh <城市> <日期>
其中<城市>是城市键（例如：nyc、london、seoul）。运行/cities查看所有选项。
<日期>为可选参数，格式为YYYY-MM-DD（默认为明天）。
显示每个模型的预测结果、集合平均值和标准差。
/cities - 列出可用城市
bash
bash {baseDir}/scripts/cities.sh
使用场景
- - 天气爱好者：查看GFS、ECMWF等模型在您所在城市的对比情况
- 活动策划者：在户外活动前检查预报置信度
- 研究人员：获取原始多模型数据进行分析
- 旅行者：当模型一致时获得高置信度预报，不一致时获得预警
- 预测市场交易者：使用集合数据评估温度市场头寸（参见polymarket-weather-scanner技能）
工作原理
- 1. 向天气集合API查询每个天气模型的预报
- 返回各个模型的预测，以便查看分布范围
- 计算集合统计量：平均值、标准差
- 标准差越高 = 不确定性越大 = 模型存在分歧
- 标准差越低 = 置信度越高 = 模型趋于一致
外部端点
所有请求均发送至天气集合API服务器。
| 端点 | 发送数据 | 用途 |
|---|
| https://polymarket-scanner.fly.dev/forecast/{城市} | 城市、日期、API密钥（如已设置） | 获取预报 |
| https://polymarket-scanner.fly.dev/cities |
API密钥（如已设置） | 列出城市 |
安全与隐私
- - 无需API密钥即可使用（基于IP限速的匿名免费层级）
- 所有请求均通过HTTPS发送至天气集合API服务器（polymarket-scanner.fly.dev）
- 天气数据来源于Open-Meteo并在服务端聚合
- 不收集或存储任何个人数据
