Measuring the neighbourhood effect to calibrate land use models

doi:10.1016/j.compenvurbsys.2013.03.006

Computers, Environment and Urban Systems

Volume 41, September 2013, Pages 55-64

https://doi.org/10.1016/j.compenvurbsys.2013.03.006 Get rights and content

Highlights

•
We used modified enrichment factors to measure the neighbourhood effect.
•
Measurements confirm the existence of a neighbourhood effect, especially for urban land uses.
•
Modified enrichment factors can be used to improve the calibration of neighbourhood rules in land use models.

Abstract

Many spatially explicit land use models include the neighbourhood effect as a driver of land use changes. The neighbourhood effect includes the inertia of land uses over time, the conversion from one land use to another, and the attraction or repulsion of surrounding land uses. The neighbourhood effect is expressed in the neighbourhood rules, but calibration of the neighbourhood rules is not straightforward. This paper aims to characterise the neighbourhood effect of observed land use changes and use this information to improve the calibration of land use models. We measured the over- and underrepresentation of land uses in the neighbourhood of observed land use changes using a modified version of the enrichment factor. Enrichment factors of observed land use changes in Germany between 1990 and 2000 indicate that the neighbourhood effect exists. This suggests that it is appropriate to use neighbourhood rules to simulate urban land use changes. Observed enrichment factors were used to calibrate a land use model for Germany from 1990 to 2000 and the obtained neighbourhood rules were validated independently from 2000 to 2006. The results show that both the allocation accuracy and the pattern accuracy of the land use model improved for the calibration period, as well as for the independent validation period. This indicates that enrichment factors can be used to improve the calibration of the neighbourhood rules in land use models.

Introduction

Land use models typically include a combination of drivers to simulate land use changes over time (Poelmans and van Rompay, 2010, Wang et al., 2011), often including the interaction between land uses in space and in time (Irwin and Bockstael, 2002, Verburg et al., 2004b). This spatial and temporal interaction between land uses is known as the neighbourhood effect, which is represented in many land use models by the neighbourhood rules (Hagoort, Geertman, & Ottens, 2008). Examples of land use models that include a neighbourhood effect are LUCIA (Hansen, 2007), Dyna-CLUE (Verburg & Overmars, 2009), and LUMOCAP (Van Delden et al., 2010).

These land use models often exist as generic modelling frameworks, which can be calibrated for a specific case study application. This calibration includes the definition of the shape and parameter values of the neighbourhood rules. However, the calibration of neighbourhood rules is not straightforward. Several automated methods have been developed (Arai and Akiyama, 2004, Jenerette and Wu, 2001, Li and Yeh, 2002, Li and Yeh, 2004, Straatman et al., 2004), but, despite these efforts, Hagoort et al. (2008) observe that the current practice of calibrating neighbourhood rules is predominantly manual. This is inherently subjective, not repeatable and highly dependent on the knowledge and skills of the modeller. One limitation of automated calibration methods is that most methods deal with the allocation of one land use type only and cannot handle the interaction between multiple land uses, while many contemporary CA models represent multiple types of land use changes (Arai and Akiyama, 2004, Van Delden and Hurkens, 2011, Wang et al., 2011). Another drawback of these calibration methods is that model parameters are assessed indirectly from the predictive accuracy of the simulation result: such assessment does not indicate directly which parameters should be changed and in what direction.

The research presented in this paper aimed to measure the neighbourhood effect of observed land use changes and use this information to improve the calibration of land use models. To do this, we measured the over- and underrepresentation of land uses in the neighbourhood of observed land use changes using a modified version of the enrichment factor (Verburg, de Nijs, van Ritsema Eck, Visser, & de Jong, 2004a). First, enrichment factors were measured for observed land use changes to test the existence of the neighbourhood effect. The enrichment factors of the observed land use changes were subsequently used to calibrate the neighbourhood rules in a cellular automata land use model. Two methods were employed to calibrate an application for land use changes in Germany between 1990 and 2000: an automated procedure and a manual procedure. Both methods were validated independently by simulating land use changes in Germany between 2000 and 2006. Calibration and validation results for both methods were compared with results from a null calibration to assess their accuracy.

In the next section we discuss the neighbourhood effect in more detail and how this is reflected in the neighbourhood rules in land use models. Section three presents the methodology for this study, including a description of the land use model, the case study application, and the details of both calibration procedures. Section four presents the simulation results and discusses these in relation to the applied calibration methods. In section five we draw conclusions and provide some directions for further research.

Section snippets

Inertia, conversion, and attraction/repulsion

Existing land use patterns influence future land use patterns in three ways: (1) through the inertia of land uses in a location, (2) through the ease of conversion from one land use to another, and (3) through the attraction or repulsion effects exerted by land uses situated in the neighbourhood of a location. The combined influence of inertia, conversion and the attraction/repulsion effects of existing land uses is known as the neighbourhood effect, which therefore includes the effects of land

Characterization of the neighbourhood effect

Verburg et al. (2004a) introduced the enrichment factor to characterize the neighbourhood effect. The enrichment factor is defined as the over- or underrepresentation of a land use in the neighbourhood of a particular location, relative to the average land use distribution: $F_{i, l, d} = \frac{n_{i, l . d} / n_{i, d}}{| N_{l} | / | N |}$ where F_i_,_l_,_d is the enrichment of land use l on location i at distance d, n_i_,_l_,_d is the number of cells of land use type l at distance d of location i, n_i_,_d is the total number of cells at distance d

Enrichment factors of observed land use changes

Fig. 4 shows the over- or underrepresentation of land uses in the neighbourhood of residential land and industrial land that appeared in Germany between 1990 and 2000 and between 2000 and 2006 as a function of the distance to the locations of land use changes. The graphs show two separate effects: the value at distance zero in the graphs indicates the conversion effect, while values at distance greater than zero indicate the over- or underrepresentation of land uses in the neighbourhood of

Conclusions

An analysis of observed land use changes indicates that there is a relation between locations of land use change and the land use in their neighbourhood. This was found for all combinations of land uses, and in both the calibration and the validation period, but predominantly for the allocation of urban land uses. Using enrichment factors to calibrate the neighbourhood rules in cellular automata land use models improved model results considerably. This was found in the calibration as well as in

Acknowledgements

We thank the anonymous reviewers for their helpful comments.

References (40)

T. Arai et al.
Empirical analysis for estimating land use transition potential functions – Case study in the Tokyo metropolitan region
Computers Environment and Urban Systems
(2004)
M. Batty et al.
Modeling and prediction in a complex world
Futures
(2005)
Y. Chen
Derivation of the functional relations between fractal dimension of and shape index of urban form
Computers, Environment and Urban Systems
(2011)
L. Poelmans et al.
Complexity and performance of urban expansion models
Computers, Environment and Urban Systems
(2010)
R.G. Pontius et al.
Detecting important categorical land changes while accounting for persistence
Agriculture, Ecosystems and Environment
(2004)
G.N. Rutherford et al.
Assessing land use statistics to model land cover change in a mountainous landscape in the European Alps
Ecological Modelling
(2008)
N. Schwarz
Urban form revisited – Selecting indicators for characterising European cities
Landscape and Urban Planning
(2010)
B. Straatman et al.
Towards an automatic calibration procedure for constrained cellular automata
Computers, Environment and Urban Systems
(2004)
H. Van Delden et al.
Integrated assessment of agricultural policies with dynamic land use change modelling
Ecological Modelling
(2010)
H. Van Delden et al.
Comparison of scale and scaling issues in integrated land use models for policy support
Agriculture, Ecosystems and Environment
(2011)

J. Van Vliet et al.

Revisiting Kappa to account for change in the accuracy assessment of land use change models

Ecological Modelling

(2011)

P.H. Verburg et al.

A method to analyse neighbourhood characteristics of land use patterns

Computers Environment and Urban Systems

(2004)

F. Wang et al.

Identifying dominant factors for the calibration of a land use cellular automata model using Rough Set Theory

Computers, Environment and Urban Systems

(2011)

Q. Zhang et al.

Simulating and analysis of urban growth scenarios of the Greater Shanghai Area, China

Computers, Environment and Urban Systems

(2011)

W. Alonso

Location and land use

(1964)

A. Anas et al.

Urban spatial structure

Journal of Economic Literature

(1998)

D.G. Brown et al.

Path dependence and the validation of agent based spatial models of land use

International Journal of Geographical Information Science

(2005)

A.C.M. De Nijs et al.

Estimating the influence of the neighborhood in the development of residential areas in the Netherlands

Environment and Planning B

(2010)

S. Geertman et al.

Spatial-temporal specific neighborhood rules for cellular automata land use modeling

International Journal of Geographical Information Science

(2007)

D. Haase et al.

Land use impacts of demographic change – Lessons from eastern German urban regions

Cited by (71)

A novel and dynamic land use/cover change research framework based on an improved PLUS model and a fuzzy multiobjective programming model
2024, Ecological Informatics
Spatial reconstruction and scenario simulation of historical processes and future trends of land use/cover change (LUCC) can help to reveal the historical background of land conversion and the spatial distribution of future land. Moreover, there is a close relationship between the spatiotemporal dynamics of land use/cover and changes in different ecosystem services (ESs). Using this relationship to simulate future land use scenarios is important. In this study, an LUCC dynamic analysis framework (LSTM-PLUS-FMOP) was constructed based on a deep learning time series forecasting model (LSTM), a parallelized urban land use simulation (PLUS) model and a fuzzy multiobjective programming (FMOP) model. The PLUS model was used to analyze the driving mechanism of land expansion and explore the land conversion pattern. In addition, three land conversion scenarios were established: natural land expansion (NLE), economic development priority (EDP) and regional sustainable development (RSD). The FMOP model and the relationship between LUCC and ES were used to perform a spatial simulation of land conversion. The uncertainty parameters in the model were treated by intuitionistic fuzzy numbers (IFSs). This study applied the constructed framework to the Yellow River Basin of Shaanxi Province (YRB-SX). The results showed that (1) from 2000 to 2020, the cropland area of the YRB-SX continuously decreased by 12.67 × 10⁴ ha, while the built-up area continuously increased by 28.25 × 10⁴ ha. The net reduction in woodland and grassland area was 13.90 × 10⁴ ha. (2) The relative error range of land prediction using the LSTM model was 0.0003– 0.0042. This model had better accuracy than the Markov chain prediction model. (3) The cropland area decreased by 0.26% (NLE), 0.85% (EDP) and 1.68% (RSD) under the three scenarios. The built-up area increased by 25.01%, 32.76% and 14.72%, respectively. The RSD scenario followed the principles of ecological protection and spatial constraints, which mitigated the degradation of the ecosystem to some extent. This coupled simulation framework will help to obtain land allocation schemes that meet the requirements of ecological protection and provide solutions for rational land management.
Introducing LandScaleR: A novel method for spatial downscaling of land use projections
2023, Environmental Modelling and Software
Land use and land cover (LULC) projections do not always have sufficient spatial resolution to allow them to be used by environmental models that project how LULC impacts a range of variables, including ecosystem services, biodiversity, and hydrology. We present a downscaling method designed to generate the high resolution LULC projections often required for environmental modelling. LULC change is allocated to a high-resolution reference map based on the density of LULC classes in neighbouring grid cells. Increasing a parameter that controls the likelihood of cells adjacent to existing LULC classes being converted to the same class generated less spatially aggregated landscapes that better represented historic LULC patterns in Colombia between 1960 and 2019. This new downscaling method is available as an R package and will enable the reconciliation of the spatial resolution of LULC projections and key processes that are embedded in a range of environmental models.
Assessing high quality agricultural lands through the ecosystem services lens: Insights from a rapidly urbanizing agricultural region in the western United States
2023, Agriculture, Ecosystems and Environment
Globally, as the human population increases, so does the demand for developable land. More than ever, agricultural land is being developed to meet housing and commercial needs, but at the same time is necessary for food security and other critical ecosystem services (ES). High quality agricultural lands (i.e. those which can support high yields, a diversity of crops, and are resilient to climate change) may play an important role in balancing the demand for housing, food, and other ES. In this paper, we used the Ecosystem Service framework to investigate the quality of agricultural lands beyond that of their value for food production. We conducted our study in the Treasure Valley, Idaho, where the issue of agricultural land loss is particularly relevant due to severe development pressure, few restrictions on land use, and limited arable land. We quantified the relationships between agricultural land quality and ES supply in 2016, and then measured how ES supply is likely to change in the future under a “Business as Usual” land use scenario. We found that higher quality agricultural land was associated with food provision, livestock crop production, and nitrogen retention, whereas lower quality agricultural land was associated with carbon sequestration, habitat quality, and recreation. Urban development is projected to occur disproportionately on higher quality agricultural land. Over 15 times more higher quality agricultural land will be developed between 2016 and 2050 compared to lower quality agricultural land. With the expected loss of agricultural land to development, the landscape in 2050 will have a lower supply of four of the six chosen ES compared to 2016. We found that livestock crop production will decrease the most over the time period (−28.8%) followed by food provision (−22.8%), nitrogen retention (−19.3%), and habitat quality (−7.4%). However, carbon sequestration (+5.5%) and recreation (+1.6%) will increase. Overall, we found strong evidence that certain ES are associated with varying levels of agricultural land quality and higher quality agricultural land was shown to be particularly vulnerable to urban development. Our results suggest the need to consider additional values of agricultural land beyond its use for food production. We recommend the implementation of policies that consider the trade-offs of developing on varying quality of agricultural land to help balance the many competing demands of our lands.
Analysing urban growth using machine learning and open data: An artificial neural network modelled case study of five Greek cities
2023, Sustainable Cities and Society
Urban development if not planned and managed adequately can be unsustainable. Urban growth models have been a powerful toolkit to help tackling this challenge. In this paper, we use a machine learning approach, to apply an urban growth model to five of the largest cities in Greece. Specifically, we first develop a methodology to collect, organise, handle and transform historical open spatial data, concerning various impact factors, into machine learning data. Such factors involve social, economic, biophysical, neighbouring-related and political driving forces, which must be transformed into tabular data. We also provide an artificial neural network (ANN) model and the methodology to train and evaluate it using goodness-of-fit metrics, which in turn provide the best weights of impact factors. Finally, we execute a prediction for 2030, presenting the results and output maps for each of the five case study cities. As our study is based on pan-European datasets, our model can be used for any area within Europe, using the open-source utility developed to support it. In this sense, our work provides local policy-makers and urban planners with an instrument that could help them analyse various future development scenarios and take the right decisions going forward.
Re-considering the status quo: Improving calibration of land use change models through validation of transition potential predictions
2023, Environmental Modelling and Software
The increasing complexity of the dynamics captured in Land Use and Land Cover (LULC) change modelling has made model behaviour less transparent and calibration more extensive. For cellular automata models in particular, this is compounded by the fact that validation is typically performed indirectly, using final simulated change maps; rather than directly considering the probabilistic predictions of transition potential. This study demonstrates that evaluating transition potential predictions provides detail into model behaviour and performance that cannot be obtained from simulated map comparison alone. This is illustrated by modelling LULC transitions in Switzerland using both Logistic Regression and Random Forests. The results emphasize the need for LULC modellers to explicitly consider the performance of individual transition models independently to ensure robust predictions. Additionally, this study highlights the potential for predictor variable selection as a means to improve transition model generalizability and parsimony, which is beneficial for simulating future LULC change.
A size-adaptive strategy to characterize spatially heterogeneous neighborhood effects in cellular automata simulation of urban growth
2023, Landscape and Urban Planning
As an important component of the cellular automata (CA) model, the neighborhood defines the geographical interaction domain (GID) of cells and their interaction mechanisms. However, the invocation of “spatial neighborhood stationarity” leads to the CA neighborhood being spatially homogeneous, which apparently violates the spatial heterogeneity of geographical dynamics. This study proposes a size-adaptive strategy for the definition of the CA neighborhood, which makes its size vary with the cell location. The CA model with size-adaptive neighborhood (SAN) has been constructed by coupling a defined size sequence with the distribution of cell GID approximated by the local accessibility distribution. Taking the urban growth of Wuhan from 2000 to 2020 as an example, the CA models with original neighborhood (ORN), dual-scale neighborhood (DSN), and SAN have been compared to investigate the performance of the SAN. The results show that the SAN is superior to the ORN and DSN in the characterization of the local interactions of urban growth, and the SAN-CA model can produce a more regular urban landscape and more accurate simulation results. Besides, the applicability of serial and parallel algorithms to the size-adaptive strategy has been detected by comparing their computational efficiency. Although the serial algorithm is easier to perform the size-adaptive strategy, the high computational efficiency of the parallel algorithm makes it more suitable for calculating the SAN effect. This study contributes to superseding the hypothesis of “spatial neighborhood stationarity” and improving the ability of CA models to simulate urban growth with spatial heterogeneity.

View all citing articles on Scopus

View full text

Measuring the neighbourhood effect to calibrate land use models

Highlights

Abstract

Introduction

Section snippets

Inertia, conversion, and attraction/repulsion

Characterization of the neighbourhood effect

Enrichment factors of observed land use changes

Conclusions

Acknowledgements

Computers Environment and Urban Systems

Futures

Computers, Environment and Urban Systems

Computers, Environment and Urban Systems

Agriculture, Ecosystems and Environment

Ecological Modelling

Landscape and Urban Planning

Computers, Environment and Urban Systems

Ecological Modelling

Agriculture, Ecosystems and Environment

Ecological Modelling

Computers Environment and Urban Systems

Computers, Environment and Urban Systems

Computers, Environment and Urban Systems

Location and land use

Urban spatial structure

Journal of Economic Literature

Path dependence and the validation of agent based spatial models of land use

International Journal of Geographical Information Science

Estimating the influence of the neighborhood in the development of residential areas in the Netherlands

Environment and Planning B

Spatial-temporal specific neighborhood rules for cellular automata land use modeling

International Journal of Geographical Information Science

Land use impacts of demographic change – Lessons from eastern German urban regions