7 Phenology
This dataset provides subnational estimates of key seasonal metrics for crop and vegetation systems, derived from remote sensing and climate data. It captures temporal patterns in vegetation growth to support agricultural planning and climate risk assessment.
7.1 Start of Season (SOS) and End of Season (EOS)
This dataset provides subnational estimates of seasonal timing derived from GLASS 8-day NDVI data using the Phenofit framework and R package. ahhhh
7.1.1 Dataset Overview
Start-of-season (SOS) and end-of-season (EOS) are derived primarily from the DER method (maximum positive/negative derivative) and threshold-based approaches (TRS2: 20% seasonal amplitude, TRS5: 50% seasonal amplitude). These are the most robust indicators of green-up and senescence, while other metrics such as logistic inflection points (UD, SD, DD, RD, Greenup, Maturity, Senescence, Dormancy) are less reliable in regions with multiple or noisy growing seasons. Users are advised to rely on DER values for seasonal analysis and timing, but TRS2 and TRS5 may also be useful.
7.1.2 Format
All SOS and EOS data are stored in Parquet format, with one file per country. The definitions for the columns are shown below.
7.1.2.1 Column definitions
| Variable | Description | Unit |
|---|---|---|
| DER.sos | Start-of-season: date of maximum positive derivative (fastest green-up) | Date |
| TRS2.sos | Start-of-season: date when curve crosses 20% of seasonal amplitude | Date |
| TRS5.sos | Start-of-season: date when curve crosses 50% of seasonal amplitude | Date |
| DER.eos | End-of-season: date of maximum negative derivative (fastest senescence) | Date |
| TRS2.eos | End-of-season: date when curve falls below 20% of seasonal amplitude | Date |
| TRS5.eos | End-of-season: date when curve falls below 50% of seasonal amplitude | Date |
| meth | Curve-fitting method selected as ‘best’ | Text |
| R2 | Coefficient of determination for selected model | Numeric |
| NSE | Nash–Sutcliffe Efficiency | Numeric |
| RMSE | Root-mean-square error of fit | Numeric |
| R | Pearson correlation between fitted and observed NDVI | Numeric |
| pvalue | p-value of the correlation (R) | Numeric |
| pixel | Raster cell index in original NDVI stack | Integer |
| admin1_name | Name of admin1 region | Text |
| flag | Flag to indicate year and season number (format: year_season#) | Text |
7.2 Seasonal Rainfall
This dataset provides estimates of seasonal rainfall aligned with the start and end of the growing season (SOS/EOS) derived from phenology data. Rainfall is based on CHIRPS v3 and is currently only processed for Kenya. It captures the total precipitation occurring within each identified season, allowing assessment of seasonal water availability for agriculture and climate risk analysis.
Rainfall metrics are computed using the same seasonal delineation as the phenology dataset, and the recommended variable for robust seasonal analysis is rain_der. This use the DER method (maximum positive/negative derivative) for SOS/EOS to produce more stable season lengths. Other thresholds (rain_trs2, rain_trs5, rain_greenup_senescence) are also provided but are less robust.
The kenya file is found on the Atlas S3. The file name is “KEN_seasonal-phenology_plus-rain.parquet”.
7.2.1 Units Table
| Variable | Description | Unit |
|---|---|---|
| rain_der | Rainfall total for the season based on DER SOS/EOS | mm |
| rain_trs2 | Rainfall total for the season using 20% amplitude thresholds | mm |
| rain_trs5 | Rainfall total for the season using 50% amplitude thresholds | mm |
| rain_greenup_senescence | Rainfall between green-up and senescence phases | mm |
7.2.2 How to begin?
To start using the phenology dataset, first identify the country or region of interest. Each country’s data is stored as a separate Parquet file, containing per-pixel seasonal metrics. For analysis at broader scales, it is recommended to aggregate the data to administrative level 1 (ADM1), using the admin1_name column in the dataset. Filtering and subsetting can be done based on the provided quality statistics if needed (see below). Aggregation can be performed by computing the mean, median, max, min, and other summary statistics of SOS and EOS dates within each Admin 1 area. These admin 1 names match with the admin 1 names in the Chapter 4 admin boundaries dataset. This allows for easier comparison across regions and integration with other spatial datasets, such as crop or livestock VoP and population exposure.