Data Preparation

Input Data Format

PhenoNN requires climate data in CSV format with daily time series. Each site should have its own CSV file.

File Naming Convention

Data files should follow the naming pattern: {PFT}_{site}.csv

For example: - GR_bullshoals.csv - DB_harvard.csv - EN_nippon.csv

CSV Structure

The CSV file should contain the following columns:

Required Columns

Column	Description	Units
year	Year of observation	YYYY
doy	Day of year	1-365/366
tmin	Daily minimum temperature	°C
tmax	Daily maximum temperature	°C
daylength	Daily daylength	hours
vpd	Vapor pressure deficit	kPa
swa	Soil water availability	mm
radiation	Shortwave radiation	W/m²
mat	Mean annual temperature	°C
map	Mean annual precipitation	mm
snow	Snow cover	mm
sand	Soil sand content	%
silt	Soil silt content	%
clay	Soil clay content	%
ph	Soil pH	pH units
gcc	Green Chromatic Coordinate	unitless
rcc	Red Chromatic Coordinate	unitless
gcc_lowess	GCC with LOWESS smoothing	unitless
rcc_lowess	RCC with LOWESS smoothing	unitless

Data Requirements

• Time span: Minimum 2 years of data to predict 1 year of GCC

• Missing values: Should be handled before input (interpolation recommended)

• Data range: Should be within realistic bounds for each variable

Directory Structure

For predictions, organize your data as:

your_data/
├── testdata/
│   ├── GR_site1.csv
│   ├── GR_site2.csv
│   └── ...
├── lstm_models/
│   ├── mfull_GR_8f_0
│   ├── mfull_GR_8f_1
│   └── ...
└── gcc_rcc_mins_site_veg.csv

Minimum GCC File

The gcc_rcc_mins_site_veg.csv file should contain minimum GCC values for each site:

,0,1,2,3,4
GR_bullshoals,GR_bullshoals,GR,site,0.287,0.487
DB_harvard,DB_harvard,DB,site,0.295,0.512