Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

Session Overview
316R: High resolution remote sensing for understanding dynamics of forest and grassland systems
Thursday, 25/Apr/2019:
10:45am - 12:15pm

Session Chair: Bronwyn Price
Location: UniS-A 022
UniS Building, room A 022, ground floor, 72 seats
Session Topics:
How do we support transformation?

Session Abstract

The rapid advancement of remote sensing technologies and increasing availability of open access data provides many opportunities to better understand land system transformations at a range of spatial and temporal scales and extents. In recent years satellite missions providing worldwide high spatial (<1-20m) and high temporal (up to daily) resolution have been launched, such as Worldview4, Sentinel-2, PlanetLabs CubeSats, etc. Some of these missions (e.g. the Copernicus’s Sentinel missions) also offer the data free of charge through an open-access portal. In addition, there have been continued advancements in the capture of laser scanning data, and very high resolution (spatial and spectral) aerial and terrestrial data including from unmanned vehicles. These new advances and data sources are considered a game changer for the application of remote sensing data to understand land system transition, management, disturbances and their intensities. The high spatial, spectral and temporal resolution of current remote sensing products allows for precise and continuous modelling and monitoring of landscape function such as phenology, primary productivity, 3D vegetation structure and fragmentation at wide extents. There are now excellent opportunities to gather data remotely in areas which have previously been understudied due to their remote or difficult to access locations. Forest and grassland systems are increasingly subject to transformation processes brought about by forces such as shifting population and climate dynamics, with significant implications for land system function.

This session will focus on forest and grassland systems. The aim is to bring together experts from land systems sciences and remote sensing and focus on recent advances in remote sensing and the specific advantages of high resolution (spectral, spatial, temporal, vertical) data for understanding land system transformation and its consequences in multi-functional forest and grassland systems.

Session organizers: Bronwyn Price and Christian Ginzler

Show help for 'Increase or decrease the abstract text size'
Full talk
ID: 520 / 316R: 1
316R High resolution remote sensing for understanding dynamics of forest and grassland systems
Keywords: land use assessment, remote sensing, grassland, time series

Rapid assessment of grassland management intensity with high resolution satellite imagery

Robert Pazur1, Ginzler Christian1, Kolecka Natalia2, Conedera Marco1, Ecker Klaus1, Price Bronwyn1

1Swiss Federal Research Institute WSL, Switzerland; 2Jagiellonian University, Kraków

Grasslands cover approximately 40 % of the world’s land surface and after forests form the world’s largest ecosystems contributing significantly to the planet’s biodiversity. The richness and diversity of grasslands vary considerably across the world and are widely influenced by human management. The heterogeneity and complexity of grassland management systems and their short-term variation make the mapping of grassland use intensity extremely challenging.
Here we propose a semiautomated method (where generation of the grassland occurrence sample set is the only non-automated task) to map grassland management intensity using complex hierarchical rules across NDVI time series derived from satellite imagery. We applied the method in the lowlands of Switzerland across the Canton of Aargau, a largely agricultural area featuring diverse land use management intensities. High-resolution Planetscope imagery (with 3m and ~1day, spatial and temporal resolution, respectively) is the main data source. Planetscope imagery is sourced from a constellation of cube satellites. In order to limit the influence of variability in NDVI across the different satellites in this constellation the Planetscope imagery was fused with imagery from other satellite sensors (Rapideye, Sentinel).
Object-oriented image analysis (OBIA) is used to derive homogenous fields (with respect to RGBN bands signal) from multi-temporal image stacks (selected cloud free scenes from the growing season). Grassland areas were successfully distinguished by delineating open landscapes (areas derived by masking out forest and urban footprints), and then applying a machine learning classification of the NDVI time series, developed on a small sample of known grassland area (presence/absence fields).
The proposed method of evaluating the properties of NDVI curves has proved to be a powerful tool for quantifying frequencies of mowing, grazing and fertilisation activities. Our results show that dense time series, especially within the growing session, are crucial for the accurate mapping of land use intensity.

Full talk
ID: 550 / 316R: 2
316R High resolution remote sensing for understanding dynamics of forest and grassland systems
Keywords: Greater Mekong Subregion, forest change, imagery composition, annual change detection

Forest change detection using time-series of high resolution remote sensing dataset

Yong Pang1, Chungan Li2, Shili Meng1, Shengyuan Hu1, Zengyuan Li1, Huabing Dai3

1Chinese Academy of Forestry, China, People's Republic of; 2Guangxi University, Nanning 530004, China; 3Guangxi Forest Inventory and Planning Institute, Nanning 530011, China

With rapid social and economic development in the Greater Mekong Subregion (GMS), forests in the region face an increasing number of threats and pressures. The capability of remote sensing techniques for forest changes monitoring and evaluation in large scale has long been recognized. With constant updating and developments of earth observation technologies, more and more high spatial resolution remote sensing data are available free of charge or at a reasonably low cost.

In this study, time series of optical remote sensing data from multiple satellites were used, including Landsat-5/7/8, Sentinel-2A/B, GaoFen-1/2/6 (GF-1/2/6 ) and Beijing-2 (BJ-2). These satellites provide repeat observations on a week to month cycle with spatial resolutions ranging from 4 m to 30 m. We developed an algorithm for composition of cloud-free optical remote sensing imagery using weights for collection date, cloud contaminated extent and opaque degree calculated at pixel level. The weighted mean value was calculated for each pixel in the composition image, then time-series of forest cover indices were calculated to detect changes. Results from test sites in Guangxi (China) and Laos were analyzed. Field GPS points and screen interpreted changes were used for evaluation. Overall, the annual change accuracy using the annual composition images is about 82%. The annual change accuracy using quarterly composition images increased to approximately 92%. As some short rotation plantations (e.g., eucalyptus, acacia) are regenerated within several months, sub-annual images are needed to detect these quick change-recover forest patches.

Full talk
ID: 465 / 316R: 3
316R High resolution remote sensing for understanding dynamics of forest and grassland systems
Keywords: unmixing, UAV, Sentinel-2, Pléiades, marsh grassland

UAVs and spaceborne optical images data fusion for grassy wetlands habitats classification

Emilien Alvarez-Vanhard1, Thomas Houet1, Thomas Corpetti1, Cendrine Mony2

1UMR 6554 CNRS LETG-Rennes, France; 2UMR 6553 CNRS ECOBIO, France

Recent airborne and spaceborne remote sensing data sources (UAVs vs. Pléiades or Sentinel2…) provide very promising datasets with respective advantages. The former can be used independently from clouds coverage with 10-20 cm multispectral images while the latter provide interesting spatial / temporal resolutions that can both be very useful for biodiversity habitat mapping. However these remotely sensed data are rarely used jointly to benefit from each other. In this study, we propose a methodology for optical data fusion of satellite (Sentinel-2 and Pléiades) and drone multispectral images for classifying grassland diversity in an intensively used marsh (Sougéal marsh of 177 ha and located upstream of the Mont-Saint-Michel bay, France). Such kind of habitat maps are of importance for managers that have to deal with biodiversity issues (invasive species, species richness preservation…) and agricultural practices (grazing, mowing…) in order to implement efficient spatial land management strategies. One of the main difficulties relies on the fact that grassy wetlands (marsh) exhibit important spatial and temporal vegetation dynamics related to hydrology (i.e. flooding periods) and land practices. This spatiotemporal variability introduces complexity for habitats mapping, making difficult the identification of clear boundaries.

The aim of the study is to identify these spatiotemporal dynamics and enhance habitats classification to the – spaceborne – sub-pixel level, through a spectral mixture analysis of a satellite temporal series with the help of UAV data and non-linear unmixing methods. Validation is made using a reference map produced by supervised machine learning (random forest) applied on a UAV time series dataset and floristic surveys. Preliminary results demonstrate the capacity of remote sensing to classify mesophilic, meso-hydrophilic and hydrophilic habitats. Moreover, although they highlight the importance of accounting for high temporal resolution time series in the classification process, UAV data reveal fine spatial patterns reliable with satellite mixed spectral signatures.

Full talk
ID: 770 / 316R: 4
316R High resolution remote sensing for understanding dynamics of forest and grassland systems
Keywords: Grassland, Livestock production, Soil organic matter, Phosphorus, Animal feed

Estimation of yield, soil carbon and fertilization needs using remote sensing for understanding and improving animal production in sown biodiverse pastures in Portugal

Ricardo FM Teixeira1, Tiago G Morais1, Marjan Jongen1, Nuno R Rodrigues2, Ivo Gama2, Tiago Domingos1

1Instituto Superior Técnico - University of Lisbon, Portugal; 2Terraprima – Serviços Ambientais Lda

Sown biodiverse pastures (SBP) rich in legumes is a nature-based, innovative and economically competitive animal production system for climate-smart pastures in the Mediterranean. SBP are a mixture of up to 20 species or varieties of high-yield grasses and legumes. They use biodiversity to promote pasture productivity, supporting a more than doubling in sustainable stocking rate, with several potential environmental co-benefits such as carbon sequestration in soils. Farmers in Portugal were paid for this environmental service between 2009-2014 by the Portuguese Carbon Fund.

Over the past years, research has focused on understanding the connection between the agronomic features of the system and its environmental benefits, and on improving both simultaneously through management. Here, we present results from three ongoing projects on SBP. Operational Group “GO Solo” is producing an expedited and low-cost method for soil organic carbon (SOC) mapping and assessment of carbon sequestration in sown biodiverse pastures. The method uses visible and near-infrared spectroscopy (VNIR) using field sensors and satellite data. Another Operational Group, “GO Fósforo”, is optimizing the use of phosphorus fertilizers in SBP by using remote data sensing for evaluating pasture nutrient needs and using Variable Rate Technology for fertilizer distribution. Finally, Project ModelMeat is using these and other features, such as indirect measurement of pasture yield, to develop a decision support service (DSS) for integrated sustainability management in SBP, starting with a sample of 100 farms in Portugal.

We correlated satellite or drone flight data simultaneously with SOC, yield, and phosphorus fertilizer needs. Satellite data is automatically loaded into the software tool developed for the DSS, and with the correlations established for the sampled farms, the three variables are automatically provided to farmers. Results show that a move towards automating monitoring and the key management procedures can help maximize farm economic and environmental sustainability.

Full talk
ID: 443 / 316R: 5
316R High resolution remote sensing for understanding dynamics of forest and grassland systems
Keywords: land-use, intensity, sentinel-2, classification, remote sensing

Mapping land-use intensities for grasslands in Germany

Maximilian Lange, Daniel Doktor, Sebastian Preidl

Helmholtz-Centre for Environmental Research - UFZ, Germany

Information on land-use is crucial for earth system science and environmental monitoring to support decision making and reporting of climate-relevant processes. Land-use intensities in particular are linked to numerous environmental processes and indicators, such as biodiversity, primary production, nitrogen deposition and resilience to climate extremes, and are thus increasingly used in ecological studies. However, there are still large knowledge gaps in the understanding of the relationship between land-use intensity and the environment. Thus, the quantification and analysis of land-use intensity of grasslands is a timely endevour. New satellite generations, such as ESA Sentinel-2, enable the detection of the mainly subtle changes induced by land-use intensification by their fine spatial and temporal resolution. We developed a methodology mapping land-use intensity of grassland areas in Germany using Sentinel-2 satellite data with 20m spatial resolution. Grassland areas were classified into low, intermediate and high land-use intensity areas. A supervised classification with the random-forest algorithm was compared to an unsupervised approach using clustering algorithms. Resulting classes of the latter were assigned to the respective land-use intensity classes by the use of time series analysis of Sentinel-2 multispectral bands. Habitat mappings of Germany's federal states served as training and validation data. Both methodologies, the supervised and unsupervised approach, were compared to find the most suitable method for land-use intensity classification. Here, we present our methodology and first results.