KICS—ZertAI × ECOLOGY × CITIZEN SCIENCE

THE KICS-ZERT RESEARCH DEMONSTRATOR

Nature is complex.
Look closer.

A leaf within a meadow. A filament within a network. A community within a reef. We bring fine ecological structure into view with AI.

FROM OBSERVATION
TO ECOLOGICAL INSIGHT
A closer view.
A broader understanding.
A NEW VIEW OF THE SAME WORLDEXAMPLE 1 OF 5
Original portrait photograph 1781861008867 with yellow flowers, grass and moss5% crop shared PCA feature colors for the same vegetation photographRGBour technology - 5%
Dense vegetationPhoto 1781861008867
↔ Drag to reveal

Five ecological worlds to exploreSelect an example ↓

COMPLEXITY, MADE VISIBLE
5 ecological image domains
50 comparative examples
4 agricultural field sites

Forced Focused Attention improves
how AI sees the world.

DINOv3 takes in the scene in a single pass. Our forced focused attention approach directs the same encoder to enlarged, overlapping local crops. See what changes in the visible structure.

Inspect the full analysis

EXAMPLE 01 / 05

Dense vegetation

Photo 1781861008867

LOCAL FOCUS
THE OBSERVATIONOriginal photographThe same RGB image for all methods
DINOv3The latest AI: DINOv3One pass over the working image
OUR VISION TECHNOLOGYOur technology:
Forced focused attention
5% crops · enlarged · four flips · 75% overlap

Slideshow · one example every 8 seconds

This vegetation example uses shared centered PCA with 16 retained axes, displaying the first three axes with the same color limits in both methods. FFA uses 5% crops. The full photograph is retained; colors are appearance features, not species labels.

How the technology works

Dense ecological scenes ask us to look beyond the whole image. We examine small, overlapping regions, enlarge them for the vision model, and average aligned views. The result is a detailed feature map that can be explored through color and clustering.

How the technology works
01

Focus locally

5% and 10% crop scales bring fine structures into view.

02

Build a continuous view

Four-flip averaging and 75% overlap combine neighboring observations.

03

Make the features inspectable

PCA and PCA–varimax color views; automatic K-means and GMM clustering, with λ=4.

MOIN vegetation photograph with overlapping crop positions illustrated
Overlapping local windows
5% or 10% of the working image’s short side

Enlarge to 384 × 384

Bicubic interpolation
DINOv3 · 24 × 24 patch tokens

Actual focused PCA–varimax feature map for the same vegetation photograph
A continuous feature field
Four flips · 75% overlap · weighted blending

Use Cases

Go beyond the highlight images. Inspect the observations, compare methods, and follow the analysis into its details.

THE BIODIVERSITY MONITOR

Your next observation starts here.

Open the working application to sign in, submit photographs, and review analysis results.

Open monitor

TEN MATCHED EXAMPLES

How our vision differs

SAM3, DINOv3, and forced focused attention on the same ecological photographs.

Compare the technologies
Aerial ecosystemsOriginal photograph · SAM3 · DINOv3 · forced focused attention with 10% crops
Aerial example neon_02: original tree-canopy photograph, 40 accepted SAM3 masks, whole-image DINOv3 PCA–varimax features, and finer local structure from forced focused attention using 10% crops, four flips, and 75% overlap.
Fungal networksOriginal photograph · SAM3 · DINOv3 · forced focused attention with 10% crops
Fungal-network example: the original grayscale photograph, two accepted SAM3 masks, whole-image DINOv3 PCA–varimax features, and branching filaments revealed by forced focused attention using 10% crops, four flips, and 75% overlap.

Select an image to open the full-size comparison.

Scroll across each image to inspect all four panels.

The underlying concept:
virtual ecologist

Monitoring, education, and reporting—connecting images, sound, and knowledge across ecological levels. The vision extends from individual organisms to whole ecosystems.

The project’s Virtual Ecologist concept: multiple purposes, multiple observation types, and connected ecological tasks. It illustrates the research vision rather than a claim that every capability is already operational.
01 / INDIVIDUAL

Who is here?

Species identification, traits, and organism descriptions.

02 / POPULATION

How many?

Abundance, density, and observations through time.

03 / COMMUNITY

How do they coexist?

Diversity, interactions, and community structure.

04 / ECOSYSTEM

What is the wider context?

Habitat properties, ecological processes, and knowledge.

About the project

KICS-Zert develops AI- and citizen science-supported monitoring for certified biodiversity conservation projects. The goal is to make biodiversity assessments more scalable and connect observations with the needs of conservation, land management, and certification.

Led by Prof. Dr. Masahiro Ryo at the Leibniz Centre for Agricultural Landscape Research (ZALF), the project works with the Global Nature Fund and the wider biodiversity community. Smartphone observations and accessible AI tools are part of the path toward broader participation.

Agricultural biodiversity monitoring is the project focus. The reef, fungal, aerial, and microscopy examples extend the demonstrator’s research exploration across image domains.

RESEARCH & PROJECT ECOSYSTEMZALFGlobal Nature FundAgoraNaturaNABU naturguckerNABU MOIN

Original photograph · RGB
Forced focused attention - local 5% - shared PCA

The Virtual Ecologist vision

Original Virtual Ecologist concept figure

Monitoring, education, and reporting across ecological levels. Original project concept figure.

DINOv3 and our forced focused attention

Original photograph · RGB
The latest AI: DINOv3 - shared PCA
Forced focused attention · local 5% · PCA–varimax

The same photograph and DINOv3 encoder. This vegetation example uses shared PCA16 and common color limits; colors are appearance features.