气候趋势分析 | Climate Trend Analysis

Performs pixel-wise trend analysis on multi-temporal temperature / precipitation time series (multi-band GeoTIFF cubes or composited series) to identify regions with significant warming / cooling or wetting / drying trends. Suitable for climate change monitoring, regional warming-rate assessment, and long-term precipitation trend screening.

Three core algorithms:

• Mann-Kendall trend test: a non-parametric rank test with statistic S, standardized Z, and two-tailed p-value, including a tie correction. It does not require normality and is robust to outliers, making it the standard method for climate trend analysis.
• Sen slope: the median of all pairwise slopes (x_j − x_i)/(j − i), providing a robust estimate of trend magnitude (units / time step) that resists outliers.
• OLS linear regression slope: provided as a comparative reference; the correlation between the two slope estimates is reported to assess robustness differences.

The built-in --synthetic mode generates simulated series with spatially varying warming trends, mild seasonality, and red noise, allowing the full workflow to be validated without network access or real data.

Dependencies / 依赖

pip install numpy rasterio scipy

Usage / 使用方法

Basic usage (synthetic data, offline)

python geoskill-climate-trend-analysis.py --bbox 116.0 39.0 117.0 40.0 --output-dir ./output

Example 1: temperature warming trend (synthetic)

python geoskill-climate-trend-analysis.py --bbox 116 39 117 40 --variable temperature --n-dates 24 --output-dir ./temp_trend

Example 2: precipitation trend (synthetic)

python geoskill-climate-trend-analysis.py --bbox 121 31 122 32 --variable precipitation --n-dates 30 --output-dir ./precip_trend

Example 3: real multi-temporal rasters

python geoskill-climate-trend-analysis.py --input annual_temp_stack.tif --alpha 0.01 --output-dir ./real_trend

Example 4: stricter significance level

python geoskill-climate-trend-analysis.py --bbox 116 39 117 40 --variable temperature --alpha 0.01 --output-dir ./strict --quiet

Example 5: bbox-only auto-synthesis + quiet mode

python geoskill-climate-trend-analysis.py --bbox 110 30 111 31 --n-dates 20 --output-dir ./auto --quiet

Output / 输出

Data Source / 数据源 / Source

• Input mode: local multi-temporal GeoTIFF (each band = one time step).
• Synthetic mode: generated locally, no external data source.

Privacy / 隐私声明 / Privacy

• Runs fully offline by default; no network requests are made.
• --synthetic mode reads no external data.
• All computation is performed locally; user data is never uploaded.

License / License

MIT

name: geoskill-climate-trend-analysis description: '对温度/降水时序执行 Mann-Kendall 趋势检验与 Sen 斜率（Sen slope）估计，输出趋势斜率栅格、显著性（p 值）栅格与时序统计 JSON。Mann-Kendall trend test and Sen slope estimator for temperature/precipitation time series, outputting slope raster, significance (p-value) raster, and time-series JSON.'

气候趋势分析 | Climate Trend Analysis

对温度 / 降水的多期时间序列（多波段 GeoTIFF 立方体或合成序列）做逐像元 趋势分析，识别显著变暖 / 变冷、变湿 / 变干区域。适用于气候变化监测、 区域增温速率评估、降水长期趋势筛查等场景。

核心算法三件套：

• Mann-Kendall 趋势检验：非参数秩检验，统计量 S、标准化 Z、双尾 p 值， 含结（tie）修正。不要求正态分布、对异常值稳健，是气候趋势分析的标准方法。
• Sen 斜率（Sen slope）：所有点对斜率 (x_j − x_i)/(j − i) 的中位数， 稳健估计趋势幅度（单位 / 时间步），抵抗离群点。
• OLS 线性回归斜率：作为对比参考，输出二者相关系数评估稳健性差异。

内置 --synthetic 模式生成含空间变化增温趋势 + 温和季节性 + 红噪声的模拟 序列，无需网络与真实数据即可验证流程。

依赖

pip install numpy rasterio scipy

使用方法

基本用法（合成数据，离线）

python geoskill-climate-trend-analysis.py --bbox 116.0 39.0 117.0 40.0 --output-dir ./output

示例 1：温度增温趋势（合成）

python geoskill-climate-trend-analysis.py --bbox 116 39 117 40 --variable temperature --n-dates 24 --output-dir ./temp_trend

示例 2：降水趋势（合成）

python geoskill-climate-trend-analysis.py --bbox 121 31 122 32 --variable precipitation --n-dates 30 --output-dir ./precip_trend

示例 3：真实多期栅格

python geoskill-climate-trend-analysis.py --input annual_temp_stack.tif --alpha 0.01 --output-dir ./real_trend

示例 4：更严格的显著性水平

python geoskill-climate-trend-analysis.py --bbox 116 39 117 40 --variable temperature --alpha 0.01 --output-dir ./strict --quiet

示例 5：仅 bbox 自动合成 + 静默

python geoskill-climate-trend-analysis.py --bbox 110 30 111 31 --n-dates 20 --output-dir ./auto --quiet

输出

数据源 / Source

• 输入模式：本地多期 GeoTIFF（每波段 = 一个时间步）。
• 合成模式：本地生成，无外部数据源。

隐私声明 / Privacy

• 默认完全离线运行，不发起任何网络请求。
• --synthetic 模式不读取任何外部数据。
• 所有计算在本地完成，不上传用户数据。

License

MIT