# Canopy height mapping using Google AEF embeddings — Data

来源：https://www.kaggle.com/competitions/canopy-height-mapping-using-google-aef-embeddings/data

采集时间：2026-10-19T13:38:33Z

The hackathon dataset contains ~50.000 quality screened GEDI L2A shots distributed across southern Netherlands and northern Belgium. For each 25m footprint we provide a single response, **rh98, in centimeters (cm)**. 

## Files


### train.csv

The training set contains ~39,500 points collected over southern Netherlands and northern Belgium, where each sample includes a single response variable, spatial coordinates (see below) and predictors.

### test.csv

the test set contains ~17,000 of the whole 56k points overlay and includes points available for internal validation of your models. It will help you fine-tune and assess the accuracy of your predictions during the development phase. In additional, the hidden dataset will be used that represents the final test of your model's effectiveness in real-world scenarios. Your goal is to predict these hidden values accurately. Height values for the hidden set come from trees surveyed in the field with traditional methods. Field-measured dominant heights are typically 1–2 m higher than GEDI RH98. If you apply a bias / linear offset to compensate, please:
- document the method (e.g. constant shift, regression, eco-type specific, …);
- state the exact offset value(s);
- cite the evidence or rationale you used.

Include this note with your submission.

## Columns

*   `rh98` - height difference between the detected ground return and the point at which 98%of the cumulative lidar-waveform energy is reached
*   `lon` - longtitude of the sample point in EPSG:4326
*   `lat` - latitude of the sample point in EPSG:4326
*   `cv_{x}_google.embedding_nl` - value of google embedding vector number *x*, from 1 - 64