Models and Tools for an - CiteSeerX

2 downloads 6963 Views 176KB Size Report
models and tools which will be of use in building ... Tabulation and Database SQL, MDA and OLAP ... going on to monitor fire-damage in European forests, to.
Models and Tools for an Integrated European Environmental Management and Decision Support System, (IEEMDSS) Keith Rennolls1, Tim Richards2, Alex Fedorec1, Moh Ibrahim1, Kevin McManus1, Alun Butler1 1 CMS,University of Greenwich, London SE10 9LS 2 Conservation Technology Ltd 1 {rk02,fa01,im01,mk05,[email protected]} 2 [email protected] Abstract This paper proceeds from the assumption that a European Environmental Information and Communication System (EEICS) is already established. In the context of primary users (land-use planners, conservationists, and environmental researchers) we ask what use may be made of the EEICS for building models and tools which will be of use in building Decision Support Systems for the land-use planner. The complex task facing the next generation of environmental and forest modellers is described, and a range of relevant modelling approaches are reviewed. These include Visualization and GIS; Statistical Tabulation and Database SQL, MDA and OLAP methods. The major problem of non-comparability of the definitions and measures of forest area and timber volume is introduced and the possibility of a modelbased solution is considered. The possibility of using an ambitious and challenging biogeochemical modelling approach to understanding and managing European Forests sustainably is discussed. It is emphasised that all modern methodological disciplines must be brought to bear, and a heuristic hybrid modelling approach should be used so as to ensure that the benefits of practical empirical modelling approaches are utilised in addition to the scientifically well-founded and holistic ecosystem and environmental modelling. The data and information system required is likely to end up as a grid-based-framework because of the heavy use of computationally intensive model-based facilities.

1. Introduction Rennolls et.al. (2004c) review the various efforts that have been made over the last fifteen years to build a European Forest Information and Communication System (EFICS). Approaches have varied from that of FIRS(2004) which is squarely based on the use of

remote sensing technology, to that of EFIS(2004) which is based on the use of GIS, but not necessarily EO imagery. Besides the aim of obtaining forest inventory/monitoring information that is comparable across the whole Europe, parallel efforts have been going on to monitor fire-damage in European forests, to monitor damage to European forests from acid-rain, to inventory European Biodiversity and monitor its changes in face of both human activities and climate change driven by the Greenhouse effect. Similar “environmental” activities have been pursued for Water Resources (European Charter on Water Resources (2001)), and Agriculture, (MARS(2004)). Most of these earlier, often quite ambitious European Environmental Inventory/Monitoring projects seem to have ended and left little heritage in their areas of focus. Rennolls et.al. (2004c), Rennolls(2003) consider the best current technologies within which to plan the development of a coherent and integrated extended EFIS(2004) (European Forest Information Service) which will approach the fulfilment of the aims of EFICS(2004) (European Forest Information and Communication System). They suggest that there would seem to be much to be gained from developing a common pan-European integrated information framework, which spanned the various interdependent components of the European Environment, as well as the different States of the EU. In this paper, we concern ourselves with the use of the data and information from such systems for building models, and using tools which will contribute to a useful and effective pan-European Forestry Decision Support System, Rennolls(2003). In doing so it seems almost obligatory to widen the viewpoint from Forests, to the Environment as a whole, including all those components mentioned above.

1.1 Primary users considered. Following Rennolls (2003), we focus issues we consider three primary generic users of pan-European forest information/data/models: 1. EU pan-European land-use planners. 2.Trans-national environmentalists and conservationists. 3. Forest and Environment modellers-researchers: data mining and model building. Many of the questions that may be asked by the planner or the conservationist, require the use of models as developed and used by the modeller. The prime question seems to be: what is the best means of sustainable management of a range of renewable resources while ensuring conservation of irreplaceable natural resource, such as biodiversity. This paper aims to briefly review the full range of relevant modelling approaches that might contribute to answering this question, and hence provide an integrated decision support framework for the management and conservation of pan-European Environmental resources.

2. The European Environment and LandUse Planning The physical reality of the European Environment is (approximately) as a sizable 2dimensional subset on the surface of the Earth’s spherical surface, over which there are distributions of various natural resources, which are associated with each other to a greater or lesser extent. It follows that 2-dimensional Euclidian planar maps can never be a completely accurate representation of this reality. Variation in altitude provides topography, alpine zones, etc… The nature of a natural resource (attribute) usually depends on the resolution of the observation process. At very high resolutions, (corresponding to the use of very small pixels), “forests” and “wet-lands” for example, decompose into individual trees and puddles. At a lower resolution, (larger pixel size), we have the impression of continuously distributed natural resources, usually observed though a pixel-based binning procedure. Mapping of a natural resource amounts to allocation of each pixel to one of a discrete number of categories corresponding to the dominant natural resource type in that pixel. This is termed classification or clustering, depending on whether the categories are known or defined independently of the pixel data. Pixels within a category should exhibit low variability of observable attributes, but there should be relatively

high difference in the mean observable attributes between categories. The spatial compactnessfragmentation of the components of categories depends on the pixel size used, a matter of observer definition. Often, category definitions depend on non-local component properties, such as minimal width, or overall area, and this can remove pixel speckle. Hence the primary representational form of the land-use planner, thematic maps, (as mosaics of supposedly homogeneous components), involve many forms of representational approximation. The land-use planner is interested not only in the thematic-map/status at a particular time (i.e. an inventory), but also in the changes in the map that take place over specified periods, (obtained though monitoring). Land-use planners are also interested in growth that takes place in attributes associated with the compartments of a particular mapped category. For example, how does the timber volume stocking/ha increase with time, or how does the species-mix change/drift with the changing climactic conditions? Since such growth is not represented in thematic maps constructed at onetime, the analysis of such growth/evolution has to resort to more primitive data representations (i.e. EO imagery) than thematic maps. The most basic element of data is an observation of a multivariate attribute at a point/pixel at a particular time, from either ground survey or remote sensing. Putting these basic data elements together as a collection, over space, over time, (possibly involving differing instrumentation, and hence pixel size), results in what we may term a “complex-image-object”. It is over such complex image objects that data-fusion models need to be built in order to estimate growth and drift effects, and to eventually capture them in thematic maps. Rennolls (2004a) argues that historical series of remotely sensed images, (most likely EO images), are an essential data component in this whole data-model integration exercise. Rennolls and Wang (2004b) presents some detailed considerations of how to go about this using a range of heuristic data-fusion models. The first stage, considered in some detail is to obtain multi-image co-registration at sub-pixel accuracy so that simultaneous classification and growth modelling can be done from the joint use of Remotely Sensed and Ground Truth data..

3. Visualisation-based and GIS Approaches The power of the “visualisation approach” is that they give the user of them a direct (visual) perception of the data, and of any models that might be fitted to them. This allows the very evolved powers of human pattern

recognition and visual discrimination to be used, both to obtain understanding of complex data objects, and the models fitted to them, and so to direct the model identification and calibration processes. Such considerations have been central in the development of the EFIS(2004) project (1998-2001) with the adoption of the visualisation-based system called CommonGIS, developed by the Department for Spatial Decision Support (SPADE) at Fraunhofer AIS. Considerable DSS functionality, with visualisation support, has also been built into the EFIS prototype, and recent enhancements to EFIS, within the NEFIS(2004) project, have included access to data mining functionality, based on the WEKA(2004) package of machine learning algorithms. Work continues within the NEFIS(2004) on the matching of the EFIS visualisation/analytical engine to the planned evolution of a network of European Forest Data Resources.

4. Analytical Database Methods, (Statistical) Tables, SQL, MDA & OLAP Given a fully functional interoperable distributed database on a particular topic, consider the analytical and modelling tools that are directly available from within the DBMS framework. Any DBMS will have a data-model associated with it. It needs to be clearly realised that this “model” is a characterisation of the structure of the DB. However, such a data-model should not be confused1 with the data-driven modelling process that develops models to characterise and represent previously unknown relationships and patterns in the data. Of the possible Statistical SummariesRepresentations of a dataset/database Statistical Tables are a primary tool. Such tables, usually based upon aggregation or averaging of attributes in the database, provide a concise numerical summary of to data from which patterns and relations are often apparent. In this sense, tabulation is a visualisation tool. Usually such “Statistical Tabulations” are identical to the tables produced within a DBMS using SQL, or even to the pivot tables produced in Excel. If a statistical tabulation is ever distinct from a DBMS table then it will be because there is an associated statistical test of a null hypothesis/model associated with the table. Modern OLAP (On-Line Analytical Processing) techniques of DBMS and data-warehouses have 1

It is not clear if the Canadian NFIS (2004) entirely avoids this confusion when it says ““Generalized data models allow the seamless representation of information by mapping (or translating) attribute information to a common representation.” See Section 5 of this paper on ‘Comparability’, the issue to which this quotation relates.

extended data-exploratory functionality that is applicable to multidimensional tables, (though a range of techniques, from drill-down, slice and dice, etc…) and these methods may also be seen as datavisualisation and exploration techniques. Part of OLAP, also, are various methods for exploring the association structures of multivariate data tables (and in particular the “cube”), and other pattern-revealing techniques are coming on-line as data mining techniques are scaled into DBMSs and warehouses. Use of validation methodology and/or significance testing of models obtained in the DBMS/warehouse context would represent a coming together of the OLAP, Data Mining and Statistical modelling approaches. One of the costs of building a distributed information system using web services, or asynchronous protocols, as described in Rennolls et.al (2004c) is that such tabulation related facilities would not be automatically available. Such facilities are one of the primary tool-sets of the generic land-use planner (e.g. Eurostat) and they would need to be implemented in some say to satisfy the full data-analytical needs of the primary land-use planner.

5. Data Compatibility/Comparability. Even though forest and tree attributes may have the same name and units (for example forest area in hectares, tree basal area in cm2 and tree volume in m3), what is being defined and measured in each European country radically differs from what is being defined and measured in many other European countries. Even though a forest or environmental DDBMS might have a nice degree of interoperability, in terms of universal ability to access, comparisons between the forest information of different countries in Europe, using any tools (visualisation or tabulation) would be meaningless since the data are incommensurate. This has long been well-known to all concerned with pan-European planning and is thee main reason for long standing “harmonisation” efforts in the EU. For example, EFICS(1997) pointed out that various European definitions of forest area depended on minimal values of :- the width of forest (9m to 50 m), the % crown cover (10% to 100%), the forest area (0.05ha to 2ha). Conversion between these different definitions (i.e. “standardisation”) depends on, among other things, the degree of fragmentation of the forested areas considered, and such conversion is not a simple matter. Similarly, it was found that for individual tree volume measurement, the minimum dbh varied between 0cms and d 12cm, and minimal top stem diameters varied between 0cm and 7.5 cm. Such

variations in definition were found to make as much as a 13% difference in the estimation of the standing timber volume of the same stand. The EFICS report concluded that a complete harmonisation of forest mensurational standards throughout Europe would require extra recording of enlarged tree histories, means of recording in-growth, removals and residues etc…. That is, the cost of harmonisation/standardization would be great. However, many of the variations in definitions depend on latitude, and the different nature of the forests found in Northern and Mediterranean countries of Europe. Päivinen, R. and M. Köhl (1998) ask if harmonisation is actually needed, given that forestry in Finland means 100% cover conifer plantations, but in Portugal it includes sparsely spaced cork trees. Köhl, Traub and Päivinen (2000) ask if harmonisation or standardisation is possible. These issues and questions are probably the reason that nothing further seems to have been attempted in the area of standardization, since 1997, in spite of efforts to develop EFIS and NEFIS have continued. Can anything be done in terms of standardization, using modelling methods, without spending vast budgets? Given that the definitions and measures of forest area depend largely on fragmentation measures, and latitude, it should in principle be possible to develop a model which recalibrates the stated forest area in any particular region by using the latitude and local fragmentation measures/models (probably derived from EO imagery). Stand volumes (per ha.) depend on dbh-distributions, on stem profile models and the chosen minimal dbh and minimal top stem diameter. With the use of appropriate diameter distribution models, and some plausible knowledge/assumptions on the way these are affected by management practices, together with stem profile models, reasonable standardization procedures for stand timber volume should be feasible. If such models for standardisation of definitions and measurements were built, it would be possible to put them together into a tool-kit of such models, possibly as a web service, which could then be used reasonably automatically to compare forestry inventory data from different countries on a common standardised basis.

6. Modelling: from ecosystem to landscape In modelling a given natural resource, such as forests, the models operate at their own levels of resolution. Hence we have individual-tree models and forest-stand-level models either of which may be regarded as either empirical/statistical or process-base-

physiological, and may be distributional, distancedependent, or spatially-explicit. There are various possible hierarchies of forest models that may be defined, with high-resolution leaf models “aggregating up” to models at higher levels; from leaf to tree, tree to stand, and stand to landscape. There are parallel hierarchies of data/measurements to which the models have to be fitted/calibrated. Development of hybridmodels which have components at different levels of a model hierarchy, with appropriate inter-level scaling properties, and the associated is a challenging current research area in forest modelling, (Monserud, 2003; Landsberg, 2003). Most practical forest management models are of the empirical stand-level type. There is still considerable scope for the combination of such “traditional” forest modelling approaches with data which comes from widely varying environments. Model parameters will be made to depend on environmental covariates, and hence pan-European forest growth and yield models (p-EFGYM) could be obtained. If p-EFGYMs are to be then a unification of the Growth and Yield sample plot databases across Europe would be necessary. A harmonized generic family of growth and yield of models would need to be adopted, and fitted in ways which respected variations in regional practice. Such models could be built offline, but if adaptive model building were required, taking into account new regional data, statistical modelling facilities would need to be made available as part of the framework, possibly with expert systems or intelligent agents to guide their (probably Bayesian) usage. Ecosystem and environmental modelling, which logically contains forest modelling, has a rich and long history with similar classes of model as mentioned above for forest models. The Scientific Committee on Problems of the Environment, SCOPE(2004), focuses its projects on the concepts and practices of sustainability: maintaining the life-support system of humankind by safeguarding the natural resources over time. Its aim is to contribute to designing processes and practices which reduce the depletion rate of nonrenewable resources, identify substitute resources, and assure a sustainable supply of renewable resources. SCOPE(2004) has a rich open-access library resource in ecosystem and environmental modelling A full hierarchy of ecosystem and environmental models, with appropriate scaling behaviour, and the parallel data structures, linked together by appropriate methods of model identification, calibration and validation is an extremely challenging long term goal. It is such a modelling framework which is alluded to in Mårell et.al.(2003) when biogeochemical models are mentioned in their discussion of the EU COST Action

E25 on scientific issues relating to sustainable forest management in an ecosystem and landscape perspective. Similar thinking underpinned recent considerations at the ENFOR-ECOFOR-EFI-IUFRO (2004) symposium: “Towards the sustainable use of Europe's forests: forest ecosystem and landscape research: scientific challenges and opportunities”. Such multi-scale ecosystem conceptual approach goes right back to the International Biological Programme, (IBP(1964-74), REM(2004). Future ecosystem modelling work must avoid the pitfalls of the IBP. Attempting to build a single integrated model from stomata up to landscape is not possible. Hybrid modelling techniques which integrate different forms of models at different scales need to be used: models at one scale (for example, p-EFGYMs) will be entered as constraints of models at lower levels rather than necessarily being discovered emergent behaviour. All the methodological disciplines must be involved in their full and modern form if there is be a chance of building practical solutions which have such wide and deep scientific foundations. A IUFRO 4.11/4.01 initiative of 2001, called the Forest Model Archive (FMA), (Rennolls, et.al. (2001)) proposes to develop an living archive of forest models in a form that they can be easily compared, re-used in differing circumstances from their original use, and easily combined into hybrid models.

7. Forest Decision Support Systems; Financial and Economic modelling; Grids. A Forest Decision Support System is a set of tools arranged into a coherent and integrated system which assists the land-use planner, or forest manager, to make decisions between alternative actions. They are needed when the situation is complex, there are multiple criteria that need to be satisfied simultaneously, and financial and economic factors need to be taken into account, as well as multi-factorial issues of sustainability, biodiversity conservation, pollution avoidance, etc… Such a DSS will make use of pEFGYMs or biogeochemical models, or their hybrids, as suitable and appropriate. They will be heuristically based, and supportive, and this might mean that they involve knowledge-based expert systems, and possibly a Grids.

References Jeffers, J.N.R. (1982) Modelling, Chapman and Hall, London. Jeffers, J.N.R.(1988) Practitioner's handbook on the modelling of dynamic change in ecosystems. Wiley, Chichester Jeffries, C. (1989) Mathematical Modelling in Ecology, Birkhauser, Switzerland.

Köhl, M.,B. Traub, and R. Päivinen. 2000. Harmonisation and Standardisation in Multi-National Environmental Statistics Mission Impossible? Environmental Monitoring and Assessment. 63 (2):361-380. Landsberg, Joe (2003) Physiology In Forest Models: History And The Future. FBMIS Volume 1, 2003, 49-63. http://cms1.gre.ac.uk/conferences/iufro/fbmis/A/3_1_LandsbergJ_1 .pdf Mårell, A., N. Kräuchi, Giorgio Matteucci, Folke Andersson, Ernst Leitgeb (2003) Scientific Issues Related to Sustainable Forest Management in an Ecosystem and Landscape Perspective. Technical Report No. 1. COST Action E25, ECOFOR, Paris. Monserut, R. A. (2003) Evaluating Forest Models In A Sustainable Forest Management Context. FBMIS Volume 1, 2003, 35-47 ISSN 1740-5955. http://cms1.gre.ac.uk/conferences/iufro/fbmis/A/3_1_MonserudR_ 1.pdf Päivinen, R., M. Lehikoinen, A. Schuck, T. Häme, S. Väätäinen, P. Kennedy, and S. Folving (2001) Combining Earth Observation Data and Forest Statistics. EFI Research Report 14. 101p. Päivinen, R. and M. Köhl (1998) Systems of Nomenclature in European Forest Resource Assessments - Is Harmonisation Needed? Proceedings of the International Workshop on Application of Remote Sensing in European Forest Monitoring. 14-16 October 1996, Vienna, Austria. pp. 9-23. Rennolls, K., Moh Ibrahim and Peter Smith (2001) A Forest Model Archive ? In Rennolls, K. (ed.); Proceedings, IUFRO 4.11 Conference, Forest Biometry, Modelling and Information Sciences. http://cms1.gre.ac.uk/conferences/iufro/proceedings/RennIbrSmithF MA.pdf Rennolls, K. (2004a) Satellite “Image-Banks” in Forest Information Systems? Prospects for Accurate Estimation of Forest Change and Growth from Data-Fusion Models. Spatial Accuracy’04. In Press. Rennolls, K. & Wang, M. (2004b) Grasping Complex Image Objects. International Workshop on Digital Forestry. In preparation. Rennolls, Keith, Tim Richards, Alex Fedorec, Moh Ibrahim, Kevin McManus and Alun Butler (2004c) Requirements and Design of an Integrated European Environmental Information Communication System, (IEEICS). FEIDSS’04. IEEE Press. Weishampel, J. (2004) Models in Ecology; World Wide Web Sites of Interest. http://reach.ucf.edu/~pcb5485/links.html Web References [All the web references with a date 2004 were accessed on May 3, 2004] European Charter on Water Resources (2001) http://www.nature.coe.int/CO-DBP6/codbp08e_01.doc MARS(2004) http://www.scot-sa.com/mars/mars710.htm CommonGIS(2004) http://commongis.jrc.it/ EFIS(2004) http://www.ec-gis.org/efis/ NEFIS(2004) http://www.efi.fi/projects/nefis/ WEKA(2004) http://www.cs.waikato.ac.nz/ml/weka/ Rennolls(2003) NEFIS Technology Review. http://cms1.gre.ac.uk/conferences/iufro/DEXA04_FEIDSS/FEIDS S%20Technology%20Review_1.doc NFIS (2004) http://nfis.org/overview/overview_e.html SCOPE(2004) http://www.icsu-scope.org/ REM(2004) (The Register of Ecological Models) http://eco.wiz.uni-kassel.de/ecobas.html ENFOR/ECOFOR/EFI/IUFRO(2004) June 2003 (Tours, France). http://www.efi.fi/events/2003/Forest_ecosystem/ FMA(2004) http://www.forestmodelarchive.info/