Forest Type Classification (Field Level @30m) – Native vs Planted

[kapildadheech]

Ticket Contents

Description

Goals

• Differentiate between native and planted forests within Area of Interest (AoI) and Micro Watershed (MWS) boundaries.
• Use datasets from planted forest records and other sources (e.g., Forest Survey of India, remote sensing data) to train classification models.
• Compute forest type rasters on Google Earth Engine (GEE) at 30m resolution.
• Vectorize raster outputs to generate field-level or MWS-level polygons representing forest types.
• Publish raster and vector outputs as Earth Engine assets with proper metadata.
• Enable visualization and temporal monitoring of native vs planted forests.

Expected Outcome

• Raster datasets (30m resolution) showing forest type classification: native vs planted.
• Vectorized polygons for AoI/MWS with attributes:
  • Forest type (native/planted)
  • Area (ha)
  • Confidence score or classification probability (if available)
• Published Earth Engine assets (raster + vector) with metadata (source datasets, processing date, classification method).
• Visualization in GEE with color-coded forest types (e.g., native = dark green, planted = light green).
• Validation report confirming coverage, classification accuracy, and alignment with known planted forest datasets.

Acceptance Criteria

Data Acquisition

• Input datasets (planted forest records, FSI, satellite imagery) must be preprocessed and clipped to AoI/MWS boundaries.
• Resolution standardized to 30m.

Raster Classification

• Raster outputs must classify each pixel as either native or planted forest.
• Entire AoI/MWS must be covered with no gaps.
• Optional: include classification confidence or probability per pixel.

Vectorization

• Raster outputs converted to vector polygons using reduceToVectors() in GEE.
• Each polygon must include:
  • Forest type
  • Area (ha)
  • Confidence score (if available)
• Polygons must align with field/MWS boundaries.

Asset Publishing

• Raster and vector datasets must be published as Earth Engine assets.
• Metadata must include data sources, classification schema, resolution, and processing date.

Quality & Validation

• Coverage check: 100% of AoI/MWS classified.
• Accuracy check: ≥85% agreement with reference datasets (FSI, planted forest records).
• Resolution check: raster outputs at 30m.
• Attribute check: all polygons contain forest type, area, and confidence (if available).
• Visualization in GEE shows correct color-coded classification.

Implementation Details

Data Sources

• Planted forest datasets (official records, plantations).
• Native forest datasets (FSI or other surveys).
• Satellite imagery: Sentinel-2, Landsat-8/9.

Processing

• Train classification model (e.g., Random Forest or Decision Tree) using labeled pixels from datasets.
• Apply model in GEE to classify forest type across AoI/MWS.
• Compute raster outputs at 30m resolution.

Vectorization & Publishing

• Vectorize raster outputs into polygons.
• Include attributes: forest type, area, confidence score.
• Upload raster and vector layers as Earth Engine assets with metadata.

Visualization

• GEE visualization with native forests (dark green), planted forests (light green).
• Overlay with AoI/MWS boundaries for field-level inspection.

Validation

• Compare outputs with ground truth or reference datasets.
• Spot-check polygons for correct classification.
• Document coverage, accuracy, and attribute completeness in validation report.

Mockups/Wireframes

No response

Product Name

KYL

Organisation Name

C4GT

Domain

No response

Tech Skills Needed

Python

Organizational Mentor

@amanodt @ankit-work7 @kapildadheech

Angel Mentor

No response

Complexity

High

Category

Backend

[ravjot07]

Hi @kapildadheech ,

I’m Ravjot Singh, a final-year B.Tech student at IIITM Gwalior and currently an SDE Intern at Capillary Technologies, with prior experience as an SDE Intern at Finexos. I was also an LFX Mentee with CNCF (Kmesh), where I worked on Kubernetes-based systems and e2e test automation.

Alongside CNCF work, I actively contribute to open-source projects in communities such as Veraison (Confidential Computing Consortium) and Sugar Labs, focusing on feature development, CI pipelines, and backend enhancements.

This forest classification project looks exciting — especially the use of GEE, raster/vector workflows, remote sensing datasets, and Python-based classification models. I’m very interested in contributing to this issue.

Please let me know how I can get started or if there’s any recommended reading before diving in.

Thanks!

[aaditeshwar]

Hi @ravjot07 , A recent paper has done this classification pan-India https://www.nature.com/articles/s41597-025-06097-z. We can potentially use its results to classify tree pixels in the CoRE stack LULC into whether they belong to natural forests or not. We have been having weekly calls to discuss methodologies for such projects. Please do go over the recordings https://core-stack.org/core-stack-innovation-challenge-1st-edition/ and join the Google group https://groups.google.com/u/2/g/core-stack-dev and discord channel https://discord.gg/FpMArraM2X. We'll meet next Friday at 3pm. Please let us know if you can join.

[aaditeshwar]

@ravjot07 , can check #229 which has a discussion on estimating the age of forests and plantations.

[RAGINI-06]

hii @aaditeshwar
I’m very interested in contributing to this project and wanted to check if it is still active or if I may take part in it. Please let me know if it’s available or if there’s any other way I can contribute.

[aaditeshwar]

Yes @RAGINI-06 , this is still open. Please join in the weekly call on Friday and we can discuss.

[RAGINI-06]

@aaditeshwar yes sir ,i m gonna attend tody's meeting

[ssz2605]

Hi @aaditeshwar , I am interested in working on this issue , is this still active? I’m Shreya Saxena, a sophomore at IIIT Jabalpur. I have previously contributed to the PictoPy project under GSoC and was among the top 15 contributors during the contribution period. Through this, I gained hands-on experience working with structured codebases, collaborative development, and feature implementation in an open-source environment.

I am interested in contributing to impactful governance-tech projects under C4GT and aim to work on well-defined, scalable solutions where I can contribute consistently and meaningfully.

[amit-spatial]

Hi all,

We'll have our dev call at 3pm on Friday as usual. There has also been a lot of interest on some issues we had created on github to add new datasets and pipelines to the CoRE stack. So we wanted to put out some guidelines to make this experience enriching and productive for all of us.

Many of the issues we had created and a wishlist of datasets we had published some time back are very domain specific and need to be technically correct before they can be included in the stack. For context, any of the layers currently on the CoRE stack that required some kind of geospatial algorithms or machine learning went through several months of testing, sometimes years, to examine the outputs carefully, understand when and where the models are performing well or not, how to handle missing data (satellite data can have cloud cover and needs specific methods to address this), and so on.

So if you are keen to pick up such issues then please share your intermediate progress with us during the dev calls. Help us understand how you are approaching the issue, show us the results, discuss problems, and it'll become much easier then to absorb your contributions. If you can inform us in advance about anything specific you'd like to discuss then we can also arrange for somebody who understands that area really well to also come on the call. Amongst so many research scholars and students who have contributed to the CoRE stack, there is a lot of expertise on LULC, computations on DEM layers, tree cover analysis, rainfall, drought, indexes like NDVI, census data, commodity pricing data, agriculture marketing, etc.

We are working hard to put up a developer website asap with all required documentation in one place, but here are two MUST read links before you begin to take up any substantial tasks.

Contribute datasets, pipelines and tools

Core-Stack Innovation Challenge

@ssz2605 @ravjot07 @RAGINI-06 See you in today's developer call!

Zoom Link
Meeting ID: 856 5729 0380
Passcode: 317250

[ssz2605]

@amit-spatial Can you please reschedule the call like I am busy till 5:00 pm ?

[ssz2605]

@amit-spatial Soon, I will be sharing my progress on the current issue along with the implementation strategy.

[ssz2605]

@aaditeshwar @kapildadheech @amit-spatial
https://docs.google.com/document/d/1S6DcYERmwFMTb9fWL5Z7VAIxI79dMuBEKfGyqPOan6I/edit?usp=sharing

Unfortunately due to my classes I cannot attend friday's meet but sharing how I will be approaching this issue , further can even share the demo (if you agree).

[aaditeshwar]

Hi @ssz2605 , if you scroll through the comments you will see a paper that has already done this classification of natural forests. So you need not do a classification. This task can be simple, to intersect the CoRE stack LULC of tree cover with the natural forests map, and produce the required statistics. I request that you attend the weekly dev calls and we can discuss this in detail.