Research

Correlation between aerosol content and land use changes from remote sensing data - AerosolLand

Remote sensing techniques allow for the monitoring of the spatial and temporal distribution of aerosols, which significantly impact air pollution and health. Hyperspectral data enable a better differentiation of the physical and chemical properties of aerosols in the atmosphere. To this end, an approach for ground-based hyperspectral imaging has been developed, which allows for horizontal spectral recording of aerosols in the air. The method enables the detection of concentrations of mixed aerosol distributions (< 2.5 μm) in urban areas (~ 1 km distance). Since the size-differentiated detection of aerosol concentration from spectral data represents an ill-posed problem, the modeling will take Land Use / Land Cover (LU/LC) into account to enable an unambiguous solution.

Data from different remote sensing systems (ground-based hyperspectral data and satellite data) are collected for the study area in Israel, which is characterized by heterogeneous land use structures. The satellite data are validated by field measurements. Different LU/LC parameters are analyzed to estimate their effects on the size distribution of aerosols. The results are compared with each other to assess how future changes in LU/LC affect aerosol distribution and concentration. The goal is to evaluate which changes influence aerosol distribution and reduce health risks. It is expected that LU/LC information can significantly improve aerosol estimation from hyperspectral data.

Models for agricultural areas and their current stand situation are to be generated from a wide variety of sensor data (e.g., satellite data, cameras on drones (RGB, hyperspectral), lasers (3D, 2D), mobile robot sensor data, stationary sensor data). These models will then be automatically interpreted and evaluated in order to provide the farmer with recommendations for action via a reasoning system.

Use Case 1: Local weed detection for weed control

Plant stands are characterized by inhomogeneities that vary spatially depending on annual weather conditions, crop rotation, etc. Based on multimodal sensor data, digital decision support tools are to be created for the farmer, which can provide recommendations for the timing, location, type, and intensity of weed treatment. In a second step, the resulting sensor data can be analyzed to determine whether they allow conclusions to be drawn about stand anomalies caused by plant stress or diseases. Performance monitoring is carried out using camera images from drone flights.

Use Case 2: Proportion of clover in clover-grass stands

The composition of grassland develops differently from an initially established state depending on the soil and management type. Multimodal sensor data will be used to perform a spatially subdivided estimation of the expected biomass and to differentiate the stand according to the proportions of clover and grass. The use of camera images from drone flights can thus help the farmer to plan and evaluate the correct management measures.

Agri-Gaia: an agricultural AI ecosystem

Sub-project (SP): Agricultural remote sensing and legal aspects of the operator model

The Agri-Gaia project is developing a platform intended to enable comprehensive data exchange in agriculture. In the field of agricultural remote sensing, innovations of the sub-project are assigned to different sub-areas. One partial goal is the characterization of plants through simulated multi- and hyperspectral training data. With synthetic data, forecasting models can be created that allow predictions about the future development of plants in the stand. Another partial goal is plant detection and localization in field crops in order to recognize weeds and initiate appropriate measures. To identify anomalies and perform site characterization, AI-based stand mapping is to be carried out. This site-specific information on soil and plant properties is linked with meteorological data for a spatially differentiated site evaluation. Based on the spatially differentiated evaluation of the site conditions of agricultural land, the farmer can manage their field in a site-specific and thus sustainable manner.

CARPE MEMORIAM

A cross-farm digital plot memory for more efficient agriculture

While in the past, agricultural farms and the operational experience associated with them were often passed down through families for generations, today, farms are frequently leased or sold. Consequently, a change of operator does not occur through a years-long joint transition and learning phase, but rather in short periods and between unrelated entities. The knowledge gained over decades is lost and must be re-acquired and relearned. This results not only in the need for knowledge acquisition but also in the necessity to build cross-farm knowledge stores that allow for the transfer of knowledge learned about specific geographical conditions during transitions of agricultural farms and plots.

The overall goal of this project is to advance digitization in agriculture. This is achieved through the use of new information and communication technologies. In CARPE MEMORIAM, the concept for a cross-farm plot memory is being developed and implemented as a prototype. The plot memory is based on internal and external farm data (e.g., farmer records, aerial photographs, cadastre data), both historical and data from ongoing processes (e.g., machine data, environmental data, manually entered data).

In particular, research is also being conducted into the extent to which new high-resolution satellite imagery, data captured by drones, and in-situ sensors can be integrated. Individual knowledge of the actors involved, gained in the past, and existing data from various sources are thus merged in the plot memory. Information gaps can be filled by satellite data and new data from various sensors. The integration of new data also ensures that the system is continuously up-to-date and thus allows for field monitoring. The developed architecture ensures interoperability with Farm Management Systems through open interfaces. With the plot memory as a digital knowledge base, agricultural processes can be optimized across different vegetation periods.

In addition to creating a concept for the plot memory, CARPE MEMORIAM aims to record vegetation parameters using multiple sensors with heterogeneous, complementary spatio-temporal resolutions. Today, sensor data can be obtained using available satellites (including Sentinel), flights (including self-developed drones), and in-situ sensors in the field (on commercial vehicles or stationary), which can be fused into a high-precision "situational awareness map". This situational map also serves to populate the plot memory. A self-developed UAV system with direct communication to the server allows for individualization of the recording and evaluation processes. Ultimately, this enables optimized cultivation management and, with decision support tools, the best possible utilization of the respective yield potential.

CognitiveWeeding: Selective weed management using Artificial Intelligence

In the CognitiveWeeding project, sensor systems are to be used to distinguish between weeds and companion plants on agricultural land. A weed is defined as an undesirable and problematic plant in the stand, while a companion plant is a plant that is economically uncritical. The plants are detected using drone- and ground-based sensor systems and subsequently classified as weeds or companion plants. The classification takes into account farm-specific crop production. The goal is to use the collected data, taking into account site and weather conditions, to provide recommendations for selective weed management. This allows for a reduction in the use of plant protection products and enables targeted mechanical weed control. Companion plants can thus serve as a habitat for insects and contribute to increasing biodiversity in intensively used agricultural areas.

For further information on DryLand, please contact  Dr. Thomas Jarmer or  M.Sc. Martin Kanning via email.

Development of methods for mapping urban surfaces, using EnMAP and multisensor data (EnFusionMap)

The main objective of the joint research project is the development of adequate methods for analyzing EnMap data and focus on the enhanced monitoring of urban areas. The project takes place in close cooperation with project partners at the University Bochum. Our subproject aims at the development of adequate methods for (i) (subpixel)-classification of urban areas, using hyperspectral data and (ii) fusion of hypespectral data with TerraSAR-X and RapidEye data. Various approaches for sub-pixel analysis of EnMAP were tested. Moreover, Convolutional Neural Networks (CNN) were adapted for mapping local climate zones.

Project Duration: 11/2014 - 02/2018

Principal Investigator: Prof. Dr. Björn Waske

Projects staff: Johannes Rosentreter

Project partners:

-       Prof. Dr. Carsten Jürgens, Ruhr-Universität Bochum, Geographisches Institut, AG Geomatik

-       Dr. Uta Heiden, Deutsches Fernerkundungsdatenzentrum (DLR-DFD), Team „Angewandte Spektroskopie“

Funding:  German Aerospace Centre (DLR) - Project Management Agency (Gefördert durch die Bundesrepublik Deutschland, Zuwendungsgeber: Raumfahrt-Agentur des Deutschen Zentrums für Luft- und Raumfahrt e.V. mit Mitteln des Bundesministeriums für Wirtschaft und Technologie (BMWi) aufgrund eines Beschlusses des Deutschen Bundestages unter dem Förderkennzeichen 50 EE 1343).

Monitoring Farmland Abandonment by multitemporal and multisensor remote sensing imagery (MOFA)   

The research project studies an area in the border region of Poland and Ukraine. With the fall of the Iron Curtain the region experienced drastic changes in political and socio- economic structures. Large farmland areas become abandoned and gradual processes of forest succession take place on the abandoned land. The aim of the project is the development of adequate strategies to monitor farmland abandonment, using multitemporal SAR and multispectral remote sensing data. Finally enhanced maps should be provided, which enable more detailed analysis of the gradual process of land cover transitions.

Project Duration: 01/2012 - 12/2013 and 1/2014 - 12/2014

Principal Investigator: Prof. Dr. Björn Waske            

Projects staff: Jan Stefanski (University Bonn)

Project partners:

-       Prof. Dr. Tobias Kümmerle, Biogeography and Conservation Biology Group, Humboldt-University of Berlin

-       Dr. Oleh Chaskovskyy, Ukrainian National Forestry University, Lviv (Ukraine)

Funding:  DFG - German Research Foundation (WA 2728/2-1, WA 2728/2-1)

The project aims on the development of methods to map and quantify the biomethane potential of crops. The sub-project deals with the development of adequate classification and regression strategies for an enhanced mapping of energy crops and retrieval of biophysical parameters by combining SAR and hyperspectral images. The specific aims of the projects are (i) development of a one-class-classifier (OCC) for mapping energy crops, (ii) adaption of recent regression methods (e.g., support vector regressions) and further development of ensemble based regression methods, and (iii) development of innovative concepts for sensor fusion to estimation biophysical parameters with higher accuracy.

Project Duration: 07/2010 - 06/2013

Principal Investigator: Prof. Dr. Björn Waske

Projects staff: Ron Hagensieker

Project partners:

-       Prof. Dr. Thomas Udelhoven, University Trier

Funding: German Aerospace Centre (DLR) - Project Management Agency (Gefördert durch die Bundesrepublik Deutschland, Zuwendungsgeber: Raumfahrt-Agentur des Deutschen Zentrums für Luft- und Raumfahrt e.V. mit Mitteln des Bundesministeriums für Wirtschaft und Technologie (BMWi) aufgrund eines Beschlusses des Deutschen Bundestages unter dem Förderkennzeichen 50 EE 1011).

This research project aims on the development of a Self-taught Learning framework for the land cover classification of remote sensing data. The approach enables the use of labeled pixels (i.e., with reference information) and unlabeled pixels from arbitrary scenes and different acquisitions dates. In contrast to semi-supervised frameworks, the unlabeled data can contain unknown and irrelevant classes. Moreover, the classes need not to be explicitly modeled. The developed framework will be used for classifying remote sensing data from different study sites and sensors.

Project Duration: 08/2013 - 07/2016

Principal Investigator: Prof. Dr. Björn Waske

Projects staff: Dr. Ribana Roscher

Project partners:

-       University of Bonn, Institute of Geodesy and Geoinformation

Funding:  DFG - German Research Foundation (WA 2728/3-1)

Sentinels supporting carbon estimates and REDD+ (SenseCarbon)

SenseCarbon develops methods to improve the mapping of REDD+ relevant land use and land cover change processes. In preparation of the upcoming Sentinel missions, SenseCarbon uses existing optical and SAR remote sensing data archives of different spatial and temporal resolutions. Our research preliminary focus on the development of advanced image processing techniques including state-of-the art (i) classification methods, (ii) fusion of X-band + C-Band as well as multispectral + SAR data, and (iii) analysis of SAR time series. Besides the methodological development, the projects aims on the generation of recent and historic land cover for study sites in the Brazilian Amazon

Project Duration: 05/2013 - 04/2016

Principal Investigator: Prof. Dr. Björn Waske

Projects staff: Ron Hagensieker

Project partners:

-       Prof. Dr. Patrick Hostert, Humboldt Universität zu Berlin, Geomatics Lab

Funding: German Aerospace Centre (DLR) - Project Management Agency (Gefördert durch die Bundesrepublik Deutschland, Zuwendungsgeber: Raumfahrt-Agentur des Deutschen Zentrums für Luft- und Raumfahrt e.V. mit Mitteln des Bundesministeriums für Wirtschaft und Technologie (BMWi) aufgrund eines Beschlusses des Deutschen Bundestages unter dem Förderkennzeichen 50 EE 1255).

Data sources for SOIL-DE include available basic data for Germany (e.g., digital elevation models, soil maps, climate and weather data), data from the European LUCAS surveys to be evaluated, historical satellite data from the LANDSAT archive, as well as current satellite data from the European Copernicus program. The derived information is intended to expand existing assessment systems and serve as a decision-making aid in sustainable and long-term spatial development. The analysis of time series of high-resolution satellite imagery (10 m to 30 m pixel resolution) using entirely new methods also represents an innovative way to detail existing soil information. The threat to soil, whose fertility and functions are impaired by changes in land use, is to be recorded nationwide and at the state level by analyzing time series of satellite images in combination with official soil information of different spatial and thematic resolutions. The goal of this survey is to detect land loss over the past ten years with high spatial precision, to determine soil yield potential for the first time, and to identify risk areas, i.e., regions with particularly high loss rates and high yield potential.

The joined research project (Planungsbüro Zumbroich, Bonn) aimed on the development of robust and (semi-)automated methods that can be used for pre-mapping in structure-ecological surveys of river courses. Additionally, we traget change detection methods that will be specifically adapted in order to verify if proposed renaturation actions have been carried out without dedicated field surveys. The project is based on a combined analysis of SAR and multispectral remote sensing data from the high-resolution satellite systems TerraSAR-X and RapidEye. Therefore the project targets fundamental mapping requirements addressed by the EU water framework directive.

Project Duration: 06/2009 - 12/2012

Principal Investigator: Prof. Dr. Björn Waske (previously Prof. Dr. M. Braun – University Erlangen-Nürnberg)

Projects staff: Sascha Klemenjak

Project partners:

-       Planungsbüro Zumbroich, Bonn

Funding: German Aerospace Centre (DLR) - Project Management Agency (Gefördert durch die Bundesrepublik Deutschland, Zuwendungsgeber: Raumfahrt-Agentur des Deutschen Zentrums für Luft- und Raumfahrt e.V. mit Mitteln des Bundesministeriums für Wirtschaft und Technologie (BMWi) aufgrund eines Beschlusses des Deutschen Bundestages unter dem Förderkennzeichen 50 EE 1011).