Validation of Federated Learning and Remote Sensing Integration for Forest Multifunctionality Mapping in Poland (MoniFun-PL)

Project description

The project focuses on validating the MoniFun methodological framework for mapping forest multifunctionality indicators in Poland, with particular emphasis on integrating multi-source remote sensing data (Landsat, Sentinel-1, and Sentinel-2) with data from the National Forest Inventory (NFI) of Poland. The project addresses key challenges related to data fragmentation and the confidentiality of sample plot coordinates by employing federated learning on a local server infrastructure.

Project goals

The primary objective of this project is to validate and test the methodology developed within the MoniFun project for mapping forest multifunctionality indicators in Poland. The specific objectives are:

1)  Convert and harmonize data from the Polish National Forest Inventory (NFI) (2016-2025 inventory cycle) into a format compliant with the nFIESTA standard and the MoniFun code lists;

2) Acquire annual satellite mosaics (Landsat, Sentinel-1, and Sentinel-2) and automatically extract spectral time series for NFI sample plots using a local Linux-based infrastructure, ensuring complete confidentiality of plot coordinates;

3) Train a deep neural network model (temporal Convolutional Neural Network, CNN) using the Flower federated learning framework to model growing stock volume (m³/ha), without transferring raw field inventory data beyond the local server;

4) Disseminate validated datasets, scripts, and modeling results in accordance with the Open Access and FAIR principles to support the European forest monitoring system and evidence-based forest policy development.

The project consists of four work packages:

WP1: Harmonization of Polish NFI data and code list mapping
WP2: Remote sensing data acquisition and local feature extraction for NFI sample plots
WP3: Federated learning and distributed CNN modeling
WP4: Evaluation, reporting, and open-access data dissemination

Characteristics of the project

The project validates the MoniFun methodological framework for mapping forest multifunctionality indicators in Poland. The study focuses on integrating ground-based data from the Polish National Forest Inventory (NFI) with remote sensing data (Landsat, Sentinel-1, and Sentinel-2) to model growing stock volume (m³/ha).

Scope of IBL participation

The Forest Research Institute (IBL) is responsible for the implementation of all four work packages.

Project contractors

Department of Geomatics
Yousef Erfanifard, Krzysztof Stereńczak, Bartłomiej Kraszewski

Partners