Speaker: Dr Yuanxi Yang (The academician of Chinese Academy of Sciences)
Prof. Yuanxi Yang is the academician of Chinese Academy of Sciences (CAS).
He graduated from the Zhengzhou Institute of Surveying and Mapping (1980)
and received master’s degree on Geodesy from the Zhengzhou Institute of
Surveying and Mapping (1987) and Ph.D. on Geodesy from the Institute of Geodesy and Geophysics, CAS (1991). Prof. Yang was the secretary of the forth committee of International Association of Geodesy (IAG), associate editor of the Acta Geodaetica et Cartographica Sinica, vice director of the Science Popularization Committee of China Science and Technology Association, advisory expert of the National Nature Science Foundation, and member of the major special committee of the BeiDou Satellite Navigation System. He has carried out and accomplished the “Data Processing Project of National 2000 GPS Geodetic Control Network” and “Combined Adjustment Project of Nationwide Astro-geodetic and Space-geodetic Networks”. He proposed the “robust estimation theory for dependent observations” and “adaptive navigation and positioning theory”. He was awarded the National Science Fund for Distinguished Young Scholars (1998), Qiushi Science and Technologies Foundation (1999) and the Ho Leung Ho Lee Foundation (2011). His two achievements have won the National Award for Science and Technology Progress and five achievements won provincial science and technology award.
Latest advances in China Beidou satellite navigation system
Speaker: Dr Jianya Gong (The academician of Chinese Academy of Sciences , director of School of Remote Sensing and Information Engineering, Wuhan University)
Prof. Jianya Gong is the academician of Chinese Academy of Sciences (CAS).
He is now the director of School of Remote Sensing and Information
Engineering, Wuhan University. He graduated from the Department of Surveying of East China Geosciences Institute (1982) and received his doctor degree from Wuhan Technical University of Surveying and Mapping (1992). He was the chairman of the sixth committee of the International Society for Photogrammetry and Remote Sensing (ISPRS). Professor Gong has made a number of original innovations in geoinformation theory and geometric remote sensing studies. He invented object-oriented data model, interoperability model, geometric imaging model and precise processing method. Based on his own theory and models, he developed independently GIS basic software (GeoStar) and network service platform (GeoSurf, GeoGlobe) and remote sensing ground processing system. Prof. Gong won National Award for Science and Technology Progress for three times and ISPRS Dolezal Achievement Award.
Theory and Method of High Precision Geometric Processing for Chinese Satellite Imagery
Who can we trust? Making the most of heterogeneous evidence when timely decisions have to be made
Speaker: Dr Lucy Bastin (Senior Lecturer in the School of Engineering and Applied Science, Aston University)
Lucy Bastin initially graduated as a zoologist, and her powerful interest in urban ecology and green corridors then led to a PhD on plant populations and dispersal within urban habitat mosaics and an MSc in GIS. This was followed by a postdoc at the University of Leicester, UK on the FLIERS
(Fuzzy Land Information from Environmental Remote Sensing) project which was led by Pete Fisher. She spent several years working as an industrial GIS software developer, focusing on maintaining map data integrity and currency for field workers with sporadic connectivity, and then returned to academia at the School of Engineering and Applied Science, University of Aston, UK. In this role she engages extensively with schools, industry and community groups and has received Excellence Awards for teaching and reviewing. The majority of her research is interdisciplinary, involving, for example, collaborations with microbiologists, economists and town planners. For the past three years she has been seconded to the Joint Research Centre of the European Commission (JRC) as the lead developer of the Digital Observatory for Protected Areas. In this post she combines her software development and remote sensing experience to generate Web-based data services and decision support tools for use by an international community of policy and decision makers. Recently, she became the East African focal point for the BIOPAMA project where JRC partners with IUCN to build capacity for improved decision-making and information sharing on protected area management and biodiversity conservation.
A key theme throughout Lucy's career has been the quantification and transparent communication of uncertainty information across a range of research domains, using interoperable standards and vocabularies. As an investigator on the INTAMAP and UncertWeb projects, she co-developed the UncertML model for encoding probabilistic uncertainty in web-based systems, and this work was further developed within the FP7 GeoViqua project to support producer and user quality models that underpin the GEO Label. Key themes in her current research include reproducible and reliable workflows for generating Essential Biodoversity Variables from extremely heterogeneous data sources, and (as a an active member of the CSA Data and Metadata Working Group) the design of accessible and user-friendly tools and standards to support appropriate aggregation and more extensive re-use of citizen science data which can be currently difficult to discover and mobilise.
Speaker: Dr Tom Hengl (Senior Researcher with backgrounds in predictive soil mapping, geostatistics, GIS and remote sensing)
Uncertainty in Machine Learning: modeling and visualizing errors in spatial prediction based on Machine Learning
Dr. Tom Hengl is a senior researcher with backgrounds in predictive soil mapping, geostatistics, GIS and remote sensing. The lecturer is an experienced
R developer and an active developer of Machine Learning Algorithms for processing soil and environmental data. Tom has over 20 years of experience in research for mapping and modelling environmental data at regional and global scales, and has published over 60 journal articles and several textbooks in the field of geo-information science, soil mapping and spatial statistics (Google Scholar h-index of 30 with >500 citations per year). He has been elected vice chair of the International Society for Geomorphometry (geomorphometry.org) in the period 2011–2015; he also initiated the GEOSTAT Summer schools (training courses using Open Source Software tools) that has been running for already 12 years at various places from Europe to North America and Australia. Tom's current special interests are developing Machine Learning methods for spatial and spatiotemporal data primarily for the purpose of automated mapping / interpolation.
Official affiliation of Tom Hengl is bellow, everyone who was interested to him can view his official website.
T. (Tom) Hengl
Senior Researcher @ Envirometrix Ltd
Consultancy, Research and Innovation
Mail: Envirometrix, Roghorst 206, 6708KT Wageningen, NL
Tel: +31 317 427537
Machine Learning has been used for decades also to generate maps from point data. But uncertainty of such models has often been not fully worked out as also the statistical theory for such models needed to be developed. We have recently discovered that Machine Learning is efficient in generating spatial and spatiotemporal predictions and can be universally used to improve or even replace traditional model-based geostatistics. Our focus has been especially on using scalable Random Forest algorithms that can be used also with large data sets. We have developed for this purpose a generic framework for spatil and spatiotemporal interpolation of enviromental data, and in this process discovered that various data-driven methods such as Jackknifing and Quantile Regression Forests can be used to provide spatially explicit measure of prediction uncertainty for Machine Learning-based models. It appears that maps of prediction errors are possibly more informative than kriging variance maps from geostatistics: the RK prediction error standard deviation map is much more homogeneous, mainly because of the stationarity assumption. The RF prediction error map, on the other hand, could potentially be used to depict local areas that are significantly more heterogeneous and complex and that require, either, denser sampling networks or covariates that better represent local processes in these areas. This potentially opens a whole new area of applications.
Taking advantage of system dynamics to improve spatial characterization
Characterizing the spatial variability of a given parameter from limited information and quantifying its uncertainty is at the realm of spatial statistics. When this parameter is used as input to a process model to make predictions about the state of a system and system state observations are available at some locations in time, it is possible to take advantage of these indirect observations of the parameter itself through some kind of inverse modeling. The ensemble Kalman filter is one of such inverse modeling techniques that has proven to be very powerful for the assimilation of dynamic information about the state of the system to improve the characterization of the parameter controlling the process. In this talk, the ensemble Kalman filter will be introduced and an example application in the field of hydrogeology will be presented with hydraulic conductivity being the parameter to characterize and piezometric heads, solute concentrations and/or temperatures the observed states of the system. The talk with end with a review of the recent developments around the Kalman filter, including other fields than hydrogeology, and with an outlook of potential continuing research.
Speaker: Dr Jaime Gómez-Hernández (Professor of Hydrogeology at the Universitat Politècnica de València (UPV) in Spain)
Jaime Gómez-Hernández (Requena, Spain, 1960) is full professor of Hydrogeology at the Universitat Politècnica de València (UPV) in Spain. He received a Civil Engineering degree by UPV (1983), an M.Sc. on Applied Hydrogeology (1987) and a Ph.D. on Geostatistics (1990), the latter two by
Stanford University. He joined UPV in 1994, where he has developed his research career in the fields of applied geostatistics, inverse modeling, stochastic groundwater modeling, and nuclear waste disposal. He has authored more than 100 publications in peer-reviewed journals. He has served as member of the editorial board of several journals, currently Advances in Water Resources and Mathematical Geosciences. He has organized two international conferences on Geostatistics for Environmental Applications (Valencia, geoENV98 and geoENV2012) and the last quadrennial International Geostatistics Congress (Valencia GEOSTATS 2016). He is currently an elected member of the International Association of Mathematical Geosciences Council and the president of the Geostatistics for Environment International Association (geoENVia).
Analysis of Uncertainty and its Propagation in Spatial Information System – A General Approach
Prof. Yee Leung has done novel research in the probabilistic approach to uncertainty analysis in general and uncertainty propagation in geographical
Speaker: Dr Yee Leung (Professor of department of Geography and Resource Management at The Chinese University of Hong Kong)
information systems in particular. He pioneers research in geographical analysis under fuzziness, and generalizes uncertainty analysis to various types of uncertainties using statistics, rough set theory, possibility theory, and theory of evidence. He has also done significant research on spatial associations and regressions, and the scaling behavior of geographical processes in space and time, particularly on trends and inter-relations. He also engages in novel theoretical and applied research in artificial intelligence, intelligent spatial decision support systems, spatial data mining and knowledge discovery, climate variability, air and water pollution, as well as urban and regional analysis. He has published 6 research monographs and over 180 international papers.
He was the former Chair of the Commission on Modeling Geographical Systems of the International Geographical Union; the Chair of the Commission on Mathematical and Computational Geography of the Chinese Geographical Society; Director of the Institute of Future Cities, former Director of the Institute of Environment, Energy and Sustainability, Leader of the Climate Change Program of the Centre of Big Data Decision Analytics, and former Dean of Faculty of Social Science, The Chinese University of Hong Kong.
Uncertainty in geographical information can come in various forms. It might be due to randomness or imprecision of the data. Under randomness, the error-laden data might not be normally distributed as assumed by most error models. Under imprecision, data might only be captured as interval-valued variables with vague judgment or fuzzy connotation. Compounding the complexity of the uncertainty problem is the nonlinearity of the transfer function, which is generally assumed to be linear, that we use to propagate uncertainty from inputs to outputs. To formulate a formal approach to uncertainty analysis and propagation, we need a general paradigm for uncertainty analysis with which accuracy can be assessed. This talk introduces a general and rigorous framework within which such uncertainties can be analyzed and propagated through operations in spatial information system. It will start with a general error model and then build into the framework novel and powerful methods for uncertainty propagation when the transfer function is non-linear, with linearity as a special case, and the inputs and outputs are multi-dimensional, with single input and single output as a special case. The role of Taylor series expansion, Monte Carlo simulation and moment design method in the propagation of various types of uncertainty will be critically examined. Numerical and real-life applications, including error propagation in GIS and space-time geography, will be made to show the efficacy and applicability of the framework and the related methods. With the burgeoning growth of interest in volunteered geographic information (VGI), error and quality assurance is of utmost importance for reliable spatial decision making, some thoughts on VGI error analysis and propagation will also be given in this talk. Our discussion will conclude with error analysis and propagation in data fusion, an important means in data analysis, where multi-source and multi-scale data are fused into the final data for decision making. The talk will shed light on the formulation of a general framework for error propagation in the fusion of multi-source and multi-scale data, and the development of optimization methods to minimize the uncertainty of the fused products so that we can solve problems with richer and more accurate information.
Handling uncertainties in spatial big data
Speaker: Dr Wenzhong Shi (Head of Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University)
Prof Shi is Head of Department and Chair Professor in Geographic Information
Science (GISci) and remote sensing, for Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University. He obtained his
doctoral degree from University of Osnabruck in Vechta, Germany in 1994.
Prof Shi's current research interests are in the areas of GISci and remote sensing, with focusing on uncertainty modelling and quality control for spatial data, object extraction and change detection from satellite images and laser scanning (LiDAR) data, 3D and dynamic modelling and spatial analysis in GISci.
Prof Shi served as President of Commission on Theory and Concepts of Spatial Information Science, International Society for Photogrammetry and Remote Sensing (ISPRS) (2008-2012), President for Hong Kong Geographic Information System Association (2001-2003). He has published more than 160 SCI journal articles and over 10 books.
He received a number of international and national awards, including the Wang Zhizhuo Award from ISPRS in 2012 and State Natural Science Award from the State Council, China in 2017.
Spatial statistics for monitoring land use changes: towards optimal use of big data
Speaker: Dr Alfred Stein (Pofessor in Spatial Statistics and Image Analysis, department of Earth Observation Science at ITC )
Prof. dr. ir. Alfred Stein (1958) is professor in Spatial Statistics and Image Analysis. He received his MSc in mathematics and information science, with a specialization in applied statistics from Eindhoven University of Technology.
He obtained a PhD in 1991 at Wageningen University on spatial statistics. He started his career at the soil science and geology department of Wageningen university. In 1995 he was appointed a visiting professor at the Faculty ITC, in the soils department. In 1999 this changed to the department of spatial data acquisition.
In 2000 he was appointed a professor at the chair of mathematical and statistical models in Wageningen university (0.2) and in 2002 he became a 0.8 professor at the new department of Earth Observation Science at ITC, which he has headed for more than 10 years. In 2008 he became vice-rector research of the institute, a position that he had for four years. This was followed in 2012 by a position as portfolio holder education of the management team of the faculty.
His research interests focus on statistical aspects of spatial and spatio-temporal data, like monitoring data, in the widest sense. Optimal sampling, image analysis, spatial statistics, use of prior information, but also issues of data quality, fuzzy techniques, random sets, all in a Bayesian setting.
Speaker: Dr Xiaohua Tong (College of Surveying and Geo-informatics, Tongji University)
Trust in Spatial Data: Concept and Framework
In this presentation, the methodological framework of “trust in spatial data” is proposed, which consists of trust measurement, trust controlling, and trust
assessment. The associated results and findings are applied in the newly-made laser-imaging system based hazardous detection in soft landing of the Chinese Chang'E-3 lunar spacecraft, the jitter detection and compensation of the Chinese civilian stereo surveying and mapping satellite, and the accuracy validation of 30-m global land cover product of China. Finally, the conclusions are drawn.
Spatial Stratified Heterogeneity, Biased Sample, and BLUE Estimators
Speaker: Dr Jinfeng Wang (Professor at LREIS, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences)
Jinfeng Wang is Professor of State Key Laboratory of Resources & Environmental Information System, Institute of Geographic Science and Resources , Chinese Academy of Sciences, Beijing, PRC. He graduated from
Department of Geography at Shaanxi Normal University with bachelor’s degree, then took master’s degree in Thermodynamics from Institute of Glaciology and Geocryology, Chinese Academy of Sciences, Lanzhou, PRC and PHD in Geographical Information Science from Institute of Geography, Chinese Academy of Sciences, Beijing, PRC; and PostDr research in spatial statistics and epidemiology in Vienna, Sheffield, Cambridge and Canberra. His areas of specialization are spatial sampling and statistical inference, with applications in environment science and population health. Prof Wang coined the concept of spatial stratified heterogeneity (SSH) and developed a series of tools (www.geodetector.org) to measure, attribute, and statistics for SSH. He with his team stipulated the PRC State Standard of Spatial Sampling and Statistical Inference (www.sssampling.org). Prof Wang is currently board members of Spatial Statistics (Elsevier), International Journal of Geographical Information Science (Taylor), Stochastic Environmental Research and Risk Assessment (Springer), Journal of Geographical Systems (Taylor), Acta Geographica Sinica (in Chinese) and Journal of Remote Sensing (in Chinese). He has published 16 books in Chinese and English, and more than 150 ISI list papers.
Spatially stratified heterogeneity (SSH), referring to the within-strata variance less than the between strata-variance, is ubiquitous in the nature, such as cological zones and many ecological variables. If a sample is too small to cover all of the strata, the histogram of the sample must be different to that of a population, we call the sample is biased. A biased sample increase the error of inferences when conventional estimators are applied to it. The geodetector q-statistic was proposed to measure SSH and to test its significance. The q value is within [0,1]: 0 if SSH is not significant, and 1 if there is a perfect SSH. If a population is SSH, a series of estimators are developed to make BLUE estimations under different sample conditions: MSN (mean of spatial nonhomogeneity) if the sample is stratified, i.e. each of strata has a sample; BSHADE (biased sentinel hospital area disease estimator) if the sample is biased; and SPA (single point area estimator) if only a single point sample available.
Uncertainties in sea ice remote sensing and the propagation to polar climate modeling
Speaker: Dr Xi Zhao (James Smith 2018 Medal Winner, Associate Professor at Chinese Antarctic Center of Surveying and Mapping, Wuhan University, China)
Xi Zhao is currently an associate Professor of Geo-Information Science at the Wuhan University, China. She defended her PhD thesis ‘Random sets to model uncertainty in remotely sensed objects’, supervised by prof. Alfred
Stein, in 2012 at the University of Twente, the Netherlands. In it she addressed spatial and spatio-temporal uncertainty by random sets, with applications in vegetation and flood level monitoring at various scale levels. After her PhD research she worked on the dynamics and uncertainty of the Antarctic sea ice extent. Her current research interest includes uncertainty modeling, image quality analysis, random sets and sea ice remote sensing. She acts as the guest editor for a special issue in Spatial Statistics, and gave a keynote speak in the Spatial Statistics conference in 2015, France. So far, she published 42 peer-reviewed papers, among which 21 were SCI indexed and most of them focused on the uncertainty issue in remote sensing application.
Sea ice is frozen seawater that floats on the ocean surface. It acts as a thermal conductor and it is a considerable reflector of incoming solar radiation. To a substantial degree it regulates local and global energy balances between the atmosphere and the underlying sea surface. In the Arctic, some sea ice persists year after year, whereas most of Antarctic sea ice is seasonal ice which melts away and reforms annually. The main problem addressed in this keynote is on how to retrieve different parameters of sea ice in the polar areas by remote sensing, including ice extent, thickness, ice lead (opening) and etc.. I will also address the types of uncertainty that exist in parameter retrieval, and how these uncertainties propagate to the further surface heat flux modeling. This presentation consists of three main parts. Firstly, a brief introduction about sea ice and its spatial-temporal dynamics in Arctic and Antarctic will be given in relation to the global change discussion. Secondly, the basic principles of sea ice remote sensing and its related uncertainties will be discussed, from the perspectives of ice edge identification, monthly ice extent average, ice thickness retrieval and ice lead detection. Since the retrieved sea ice parameters are used in successive climate modeling, we use the surface heat flux models over ice leads and polynya as examples to show how the uncertainty in remote sensing outputs continuously propagate to the next modeling step. Quantification of the uncertainty in retrieved sea ice parameters will help us to better understand the climate modeling variability. We consider remote sensing as an advanced technique with its unique advantages in sea ice studies, but its retrieval results are still constrained by issues of scale, and the related uncertainty will further influence the climate modeling.
Speaker: Dr Murray Lark (Professor of Geoinformatics at the University of Nottingham)
Some reflections on prediction error
Dr. Daniel A. Griffith is an Ashbel Smith Professor and a faculty member in the School of Economic, Political and Policy Sciences (previously, Social Sciences) at UT-Dallas, Associate Program Head of the Geography-Geospatial Sciences Program in the Geographic Information Sciences, a past editor of
Geographical Analysis (2008-2014), a previous member of the Steering Committee of the Commission of Modeling Geographical Systems, International Geographical Union (2008-2012), and chair of the Steering Committee of the International Spatial Accuracy Research Association (2014-). He teaches courses about spatial statistics/spatial econometrics, GIScience research design, and urban economics. His primary areas of research are in spatial statistics, quantitative urban and economic geography, and spatial epidemiology. In chronologicaly order, he held faculty positons at Ryerson University, at SUNY/Buffalo, at Syracuse University, and at the University of Miami, before moving to UT-Dallas in 2005. He has been a visiting professor at Oregon State University (under the auspices of the USEPA EMAP Program), Erasmus University/Rotterdam, University of Rome I (La Sapienza), Cambridge University (under the auspices of the Leverhulme Trust), and University of Jyväskylä. He also has been an American Statistical Association Research Fellow to USDA-NASS, a visiting researcher at the Max Planck Institute for Demographic Research/Rostock (Germany), a Fulbright Research Fellow (to the University of Toronto), a Fulbright Senior Specialist (to the University of Alberta), a Guggenheim Fellow, an elected Fellow of the New York Academy of Sciences, a elected founding fellow of the Spatial Econometrics Association, an elected fellow of the Regional Science Association International, an elected fellow of the American Association for the Advancement of Science (AAAS), awarded Distinguished Research Honors by the AAG, and a past president of the North American Regional Science Council.
Prediction error is a familiar concept to spatial accuracy researchers. Although it plays an important role in spatial accuracy and uncertainty research, it needs to play an even more important role in the future. In order to do so, more needs to be know about this concept. This presentation derives properties of prediction error from: the ratio PRESS/ESS (i.e., the prediction residual error sum of squares divided by its corresponding fitted regression equation error sum of squares); the five common sources of error (i.e., calculation, measurement, specification, sampling, and stochastic noise); the aspatial and spatial versions of the E-M algorithm for calculating imputations; and, Moran eigenvector spatial filtering. Illustrations of findings employ Cressie’s Pennsylvania coal ash data, Bailey and Gatrell’s High Peak remotely sensed spectral bands, and Griffith and Haining’s Ohio space-time agricultural production data.