Spatio-Temporal Database Coupled with Spatial Statistics for Urban Land Use Change Analysis Chenglin Xie1, Bo Huang1, Christophe Claramunt2 and Magesh Chandramouli3 1Department of Geomatics Engineering University of Calgary 2The French Navy Academy Research Institute France 3GIS center Feng Chia University
Outline Introduction Spatio-Temporal Data Model and Query Language Rural-Urban Land Conversion Modeling Case Study Summary
Introduction Understanding the driving forces for urbanization is critical for proper planning and management of resources Comprehensive and consistent geographical record of land use and relative information: a prerequisite to understanding land use change Modeling the rural-urban land conversion pattern: critical to predicting urban growth
Introduction (Cont’d) It is necessary to bridge the gap between spatio-temporal database modeling and land use prognostic modeling Automate the process of change-tracking and predictive analysis Makes it possible to look back exploring why the change happened
Spatio-temporal data models Snapshot model Space-time composite model Event-based spatio-temporal data model Spatio-temporal object model in line with the Object Database Management Group (ODMG) standard Huang, B. and Claramunt, C., 2002. STOQL: An ODMG-based spatio-temporal object model and query language. In D. Richardson and P. Oosterom (eds.), Advances in Spatial Data Handling, Sringer-Verlag. Huang, B. and Claramunt, C., 2005. Spatiotemporal data model and query language for tracking land use change. Accepted for publication in Transportation Research Record, Journal of Transportation Research Board, US.
Our spatio-temporal object model Different properties (e.g. owner and shape) may change asynchronously owner: John (1990)–> Frank (1993) –> Martin (2000-now) shape: 1990 1996 2002 Different properties may be of different types (string, integer, struct etc.) owner: string shape: polygon
Our spatio-temporal object model (cont’d) Shape can change in different forms:
Our spatio-temporal object model (cont’d) Designed a parametric type to represent the changes on different properties Parametric type allows a function to work uniformly on a range of types. Temporal<T> (T is a type) {(val1, t1), (val2, t2), (val3, t3), …, (valn, tn)} val: T Class parcel { integer ID; temporal<string> owner; temporal<string> lutype; //land use type temporal<polygon> shape; }
Tracking of complex land use changes 1984 1992 1997 2002
Representing the complex change 345600001’s change: { ([1984, 1991], struct(Land_use_type: “agriculture”, Gextent_ref: “G345600001|1984”)), ([1992, now], struct(Land_use_type: “urban”, Gextent_ref: “G345600001|1992”)) } Temporal<T> is used to represent the changes on different attributes
Spatio-temporal Query Language Spatio-temporal DBMS Query language Data model Interact with the database Spatio-temporal database
Syntactical Constructs STOQL OQL Type [time1, time2] Struct(start: time1, end: time2) TimeInterval e! e.getHistory() List es.val T (any ODMG type and basic spatial types) es.vt es.index e.getStateIndex(ev) (es in e) Unsigned Long
Query Example 1 Query 1. Display all the parcels of land use ‘agricultural’ in 1980. Select p-geo.val From parcels As parcel, parcel.geo! As p-geo, parcel.landuse! As p-landuse Where p-landuse.vt.contains([1980]) and p-geo.vt.contains([1980]) and p-landuse.val = ‘agricultural’
Query Example 2 Query 2. What were the owners of the parcels which intersected the protected area of the river ‘River1’ over the year 1990, while they were away from that protected area over the year 1980. Select parcel.owner From parcels As parcel, parcel.geo! As parcelgeo1 parcelgeo2, protected-areas As p-area, p-area.geo! As p-areageo1 p-areageo2 Where p-area.name = ‘River1’ and p-areageo1.vt.contains([1980]) and parcelgeo1.vt.contains([1980]) p-areageo1.val.disjoint(parcelgeo1.val) and p-areageo2.vt.contains([1990]) and parcelgeo2.vt.contains([1990]) p-areageo2.val.intersects(parcelgeo2.val)
Rural-Urban Land Conversion Modeling Several techniques Cellular automata (CA) Exploratory spatial data analysis Regression analysis Artificial neural networks (ANNs) The general form of logistic regression model:
Case Study New Castle County, Delaware, USA is selected as study area Snapshots of land use and land cover in 1984, 1992, 1997 and 2002 are used Land use classifications Urban areas Residential Commercial Industrial Agricultural Others (not suitable for development) Forest Water Barren
Land use data
GIS-based predictor variables Seven predictor variables were compiled in ArcInfo 9.0 based on 50m×50m cell size Three classes of predictors were employed Site specific characteristics Proximity Neighborhoods Variable name Description Dens_Pop Population density of the cell Dist_Com Distance from the cell to the nearest commercial site Dist_Res Distance from the cell to the nearest residential area Dist_Ind Distance from the cell to the nearest industrial site Dist_Road Distance from the cell to the nearest road Per_Urb Percentage of urban land use in the surrounding area within 200m radius Per_Agr Percentage of rural land use in the surrounding area within 200m radius
Spatial sampling Assumption of econometric model—error terms for each individual observation are uncorrelated Integration of systematic sampling and random sampling methods Land use type Owner shape
Binary logistic regression Variable Model 1984-1992 Model 1992-1997 Model 1997-2002 Coefficient S.E. Dens_Pop -0.0000358 0.0002178 0.0001146 0.0003900 -0.0001553 0.0003338 Dist_Com -0.0001541 0.0000716 -0.0002411 0.0002320 -0.0000207 0.0001753 Dist_Res -0.0000596 0.0001409 0.0001611 0.0005761 0.0005248 0.0004804 Dist_Ind 0.0000589 0.0000280 0.0003375 0.0002128 0.0000389 0.0001639 Dist_Road -0.0044079 0.0010538 -0.0017445 0.0013882 -0.0039010 0.0013603 Per_Urb 0.239770 0.0115755 0.367502 0.0273304 0.394208 0.0288114 Per_Agr -0.0967720 0.0090168 -0.0931497 0.0165395 -0.122817 0.0146438 Constant -0.125040 0.342002 -2.09796 0.550595 -0.654405 0.411060 G.K. Gamma 0.94 0.97 0.96 PCP 92.8% 97.9% 95.7% Note: S.E.: standard error. G.K. Gamma: Goodman-Kruskal Gamma PCP: percentage correctly predicted
Prognostic capacity evaluation The validation process of the model is performed for the span of 1984-2002 The overall 81.9% correct prediction is relative high and the accuracy of correct prediction for urbanized area (62.3%) is relative satisfactory compared to the results of other researches in this field Observed Predicted Total % correct Urban Agriculture 45243 27425 72668 62.3 15775 150351 166126 90.5 Overall 61018 177776 238794 81.9
Prognostic capacity evaluation (Cont’d)
Summary Bridges the gap between spatio-temporal database modeling and land use change analysis Spatial-temporal data model represents complex land parcel changes dynamics over time and parcel Employs spatial land use, population and road network data to derive a predictive model of rural-urban land conversions in New Castle County, Delaware Succeeds largely in revealing the land use change