Datasets:

DeKUT-DSAIL
/

miti360

In the era of artificial intelligence, machine learning combined with remote sensing and ground measurements offers unprecedented opportunities to enhance forest monitoring through faster, more accurate biomass estimation and individual tree analysis. Despite widespread interest, Africa suffers from a shortage of ML-ready forestry datasets, with most major collections—NEON Crowns, Auto Arborist, ReforesTree, and SavannaTreeAI —originating elsewhere. The Miti360 dataset aims to bridge this geographic gap and is tailored to support data-driven decision-making in establishing and monitoring reforested stands across diverse African landscapes. Existing ML-ready datasets from the Global North have limited relevance in Africa.

At a Glance

Dataset:
- Images & Related Files
- Weather Data
Code

The dataset comprises aerial image data (orthophotos and tiles) annotated with bounding boxes for each tree, annotated terrestrial images (single and stereo), tree inventory data (biophysical parameter measurements, GPS coordinates, and species), and historical weather data (precipitation and temperature). These data were collected from a 770-ha reforested section of the Kieni Forest in Kenya between March 2023 and February 2025.

Below is a tabular summary of the dataset contents:

#	Data Category	Data Type	Quantity	Format
1	Drone Images	Orthophoto	3	TIF
		Tiles	1117	TIF
		Tree crown annotations	61866	JSON
		Tree crown species	1208	CSV
		Tree species shapefile	1208	SHP
2	Tree ground measurements	Numeric data	1208 (604 trees twice - 2024 & 2025)	CSV
3	Ground based single images	Images and tree masks	1208 (604 trees twice - 2024 & 2025)	JPEG
4	Tree stereo images	Images and tree masks	1208 (604 trees twice - 2024 & 2025)	JPEG
5	Weather data from 46 stations	Time series data	8 years daily data	API endpoint

For each tree whose data was recorded during the field survey, there is a single image captured using a smartphone and a pair of images captured with a stereo camera. Other attributes recorded are the location, species, height, crown diameter, and basal diameter. These are captured in CSV files with the following column names:

PHONE_IMAGE_FILENAME: Tree's image taken with a smartphone. Saved in JPG format.
LEFT_STEREO_IMAGE_FILENAME: Left image of the stereo pair. Saved in JPG format.
SPECIES: The species of the sampled tree. Given in standard binomial nomenclature.
TH: Height of the tree in cm.
CD: Crown diameter of the tree in cm.
BD: Basal diameter of the tree in cm.
NORTHINGS: How far north or south, in metres, a tree is from the equator based on the CRS EPSG:21037.
EASTINGS: How far east or west, in metres, a tree is from the Greenwich meridian based on the CRS EPSG:21037.
LATITUDE: GPS latitude of the tree in the WGS84 coordinate frame.
LONGITUDE: GPS longitude of the tree in the WGS84 coordinate frame.

Dataset Organisation and Hosting

Directory Structure

The dataset has been organised into two folders - one for aerial images plus associated files and the other terrestrial images plus associated files. Each of the two folders has been zipped and stored in a Google Cloud Storage bucket.

The directory structure of the two folders is as follows:

Aerial/
│   ├──Phase_1/
│   │   ├── Orthophoto/
│   │   │   └── xxx.tif
│   │   │   ├──bounding_boxes/
│   │   │   │   └── 2024_08_annotations.shp
│   │   │   │   └── ...
│   │   │   ├──tree_attributes/
│   │   │   │   └── tree_attributes_ex_1.shp
│   │   │   │   └── ...
│   │   ├── Tiles/
│   │   │   └── xxx.tif
│   │   │   └── ...
│   │   ├── Annotations/
│   │   │   └── xxx.json
│   │   │   └── ...
│   
│   ├──Phase_2/
│   │   ├── Orthophoto/
│   │   │   └── xxx.tif
│   │   │   ├──bounding_boxes/
│   │   │   │   └── 2025_02_annotations.shp
│   │   │   │   └── ...
│   │   │   ├──tree_attributes/
│   │   │   │   └── tree_attributes_ex_2.shp
│   │   │   │   └── ...
│   │   ├── Tiles/
│   │   │   └── xxx.tif
│   │   │   └── ...
│   │   ├── Annotations/
│   │   │   └── xxx.json
│   │   │   └── ...
│   

Terrestrial/
├── Phase_1/
│   ├── config/
│   │   └── single_1.yml
│   │   └── stereo_1.yml
│
│   ├── images/
│   │   ├── phone/
│   │   │   └── xxx.jpg
│   │   │   └── ...
│
│   │   ├── stereo/
│   │   │   └── xxx_left.jpg
│   │   │   └── xxx_right.jpg
│   │   │   └── ...
│
│   ├── masks/
│   │   ├── phone/
│   │   │   └── xxx_mask.jpg
│   │   │   └── ...
│
│   │   ├── stereo/
│   │   │   └── xxx_left_mask.jpg
│   │   │   └── xxx_right_mask.jpg
│   │   │   └── ...
│
│   ├── metadata/
│   │   └── tree_attributes_phase_1.csv
│
├── Phase_2/
│   ├── ...

Naming Convention

Tiles: year_month_{tile_id}.tif
Single images: year_month_{plot_id}phone{tree_id}.jpg
Single image masks: year_month_{plot_id}phone{tree_id}_mask.jpg
Stereo images: year_month_{plot_id}stereo{tree_id}_{left/right}.jpg
Stereo image masks: year_month_{plot_id}stereo{tree_id}_left_mask.jpg
Camera configuration files: {single/stereo}_{phase_id}.yml
Tree attribute files: tree_attributes_phase_{phase_id}.csv

Meaning of fields:

year: Year when image was captured or file created
month: Month when image was captured or file created
tile_id: A unique three-digit integer indicating the position of the tile in the sequence
plot_id: A unique two-digit integer identifying the specific plot in the sampling design where the picture was taken
tree_id: A unique three-digit integer indicating the position of the tree in the sequence of all trees sampled during one phase of the data collection
left/right: Identifies whether a stereo image is the left or right one in the pair
single/stereo: Identifies whether the camera calibration file is for a single camera or a stereo camera
phase_id: Identifies the camera calibration file is for the stereo camera during which phase of data collection e.g., 1 or 2.

Weather Data

For instructions on how to access the weather data, we refer users to our GitHub repository.

NOTE on Masks

For every image inside the images/phone or images/stereo folders, the corresponding mask is located in the masks/phone or masks/stereo folder. The only difference between the filenames is in the suffix "_mask" appended to the image's filename before the extention. For example, the mask of image 2024_07_01_phone_001.jpg is 2024_07_01_phone_001_mask.jpg, and so on.

Dissemination

The Miti360 dataset has a website. You can read more about it there.

Technical Report

For more details on the methods used to develop the dataset and usage guidelines, please refer to the technical report.

Code

For the source code used to analyse the data, please look at the notebooks found in our GitHub repository

Appropriate Usage of Miti360

Miti360 can be used in varied ways to train and assess machine learning models. One useful research angle we have pursued in the past is that of automating tree inventory using stereoscopic photogrammetry. With recent advances in deep learning and 3D computer vision, the stereoscopic images in Miti360 would be invaluable in developing better techniques for achieving the same goals. Regardless of the ways in which dataset may be used, we believe that all efforts directed towards developing novel techniques for forest monitoring tailored towards our African context will produce the greatest impact.

Licensing

The miti360 dataset is licensed under the CC-BY 4.0 License, Version 4.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at https://creativecommons.org/licenses/by/4.0/legalcode.txt.

Citation & Attribution

If you use Miti360 for research purposes, please consider citing:

@techreport{kiplimo2024miti360,
  title = {Miti360: An Integrated Dataset Combining Remote Sensing, Ground Measurements and Weather Data for Improved Reforestation Monitoring},
  author = {Kiplimo, Cedric and Mbatia, Samuel and wa Maina, Ciira and Sichangi, Arthur and Gitundu, Denis},
  institution = {Centre for Data Science and Artificial Intelligence (DSAIL), Dedan Kimathi University of Technology and Institute of Geomatics, GIS and Remote Sensing (IGGReS), Dedan Kimathi University of Technology},
  publisher={Hugging Face}
  year = {2026},
  howpublished = {\url{[https://huggingface.co/datasets/DeKUT-DSAIL/miti360](https://huggingface.co/datasets/DeKUT-DSAIL/miti360)}}
  address = {Nyeri, Kenya}
}

Miti360: An integrated dataset combining remote sensing, ground measurements and weather data for improved reforestation monitoring