INTRODUCTION

With the climate rapidly changing and humans having increasing impact on global ecosystems, today it is not longer sufficient to classify and analyse ecosystems based on their geographical location and historical data. Here, we classify ecosystems based on their wider geoecological parameters. The goal is to create a classifier that allows us to dynamically classify ecosystems. Perspectively, used to predict i.e. expected species richness, species extinction rates, soil status, soil detioration and eventually ecosystem collapse.

Project objective

Minimal viable product (MVP): On EcoVerse we show the development of ecosystems over time.

Goal: Showing a trend of ecosystem development and migration over time

Validation: Highlighting renaturation efforts of the past decade such as Great Green Wall Initiative, Chinese Loess Plateau Rehabilitation and the Netherlands’ Marker Wadden

Disclaimer

This project is designed and pursued in the context of the Data Analytics Consulting Bootcamp 2024-2 from neuefische GmbH. During which a capstone project is to be designed for the timeframe of four weeks, including setup, data retrieval, analysis and integration.

Collaborators:

Member	Role
Heiko Främbs	Communications Project management
Soma Pasumarthy	Web integration Database maintainance
Alexander Schmidt	Data acquisition Data handling
Noah Kürtös	Conceptualization Scripts

About the repository: Collabotors ought to work in independent branches and merge into main need to be approved by at least one other collabotor. For team communications this miro board is used.

DATA

Name	Description	Content	Data URL	Notes
EarthData	Library of accumulated satellite data	Radiation based climate data (detailed here )	GLDAS Noah Land Surface Model L4 monthly 0.25 x 0.25 degree V2.1 (GLDAS_NOAH025_M)	From 2000-2024
			GLDAS Noah Land Surface Model L4 monthly 0.25 x 0.25 degree V2.0 (GLDAS_NOAH025_M)	From 1948-2014
		Elevation data relative	ASTER Advanced Spaceborne Thermal Emission and Reflection Radiometer	From 2009 as .geotif
		Night time illumination data	VNP46A1 - VIIRS/NPP Daily Gridded Day Night Band 500m Linear Lat Lon Grid Night	From 2012 onward as .geotif Black marble
		Vegetation index (NDVI)	MOD13A2.061 Terra Vegetation Indices 16-Day Global 1km	Earth engine snippets as .geotif
Copernicus	Sentinel mission satellite data. 90m global resolution (TanDEM-X mission)	Elevation data absolute	Copernicus DEM - Global and European Digital Elevation Model	Mission: from 2011-2015. Data availble: from 2019-2026 Earth engine snippets as .geotif
DEPRECIATED Meteostat	Library of global weatherstations	Weather data	Data is sourced through the python library meteostat (documentation)	Representation bias towards EU and NA
DEPRECIATED Open elevation	Free API alterntavie to Google	Elevation data absolute	Data is sourced through the API open_elevation_request.py	Open elevation data doesn't feature data from lat > 60°

---

WORKFLOW

1. Literature research

NASA's EarthData features a wide range of global satellite data which can be used to infer climate paramters. These parameters are already widely used for local studies featured in their "News" section. Moreover, there is a "Global Ecosystem Viewer" from the United states geological survey for global ecosystems (27.10.2016).

Hence, the focus of this project was to integrate and automate the classification of ecosystems using dynamically updated satellite data.

2. Data acquisition and storage

All data is acquired through EarthData or through the GoogleEarth engine. The data is stored on an AWS postgreSQL server. Currently migrating.

2.1 Exemplary visualisation of input data

1.) VIIRS night time illumination

---

2.) MODIS vegetation index

---

3.) Copernicus elevation data

3. Modelling

For modelling all parameters were aggregated for each grid cell and a model was trained on 73 different locations of 15 different ecosystems using RandomForestClassifier from sklearn.

label_encoder = LabelEncoder()
scaler = StandardScaler()

#define model data
X = df.drop(columns=["ecosystem", "name"])
y = label_encoder.fit_transform(df["ecosystem"])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

#split data
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

#modelling
model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)

3.1 Training data (2020)

3.2 Classification (2012)

4. Forecasting

For forecasting LinearRegression from sklearn was performed for each parameter per grid cell individually.

# Prepare data for the model
X, y = np.array(df_pixel["year"]).reshape(-1, 1), np.array(df_pixel[parameter])

# Check if there is enough data to fit the model
if len(X) > 1:
    # Fit the model and calculate R-squared
    model.fit(X, y)
    r_squared = model.score(X, y)
    predict_val = model.predict(np.array([[predict_year]]))[0]  # Extract single value

4.1 Classification of forecast for 2030

4.1 Classification of forecast for 2100