Effects of Time-Duration on the Performance of the Spatial-Markov Model for Land use Change Forecasting

Abstract

Markov chain is one of the most widely used methods for land use change forecasting, however, it's a nonspatial model and few papers have discussed the effects of timeduration on its performance. In this paper, we first present the primary methodologies of the Spatial-Markov model, which endows the ordinary Markov chain with spatial dimension using spatial analysis techniques, and then explore the effects of forecasting time-duration on the model's performance. By taking Shandong province, China as a case study area, land use maps in 1990, 1995, 2000, 2005 and 2010 were created using on Landsat images and then the Spatial-Markov model was developed at 1 km spatial scale. In detail, we repeatedly run the model by choosing different initial time points and the same time step (five year interval) to simulate the spatial-temporal dynamics of land use change from 1990 to 2010. The forecasting results of a single run included a stack of ratio scale images and a derived nominal scale image, χ 2 test and Kappa coefficient were adopted to evaluate their accuracy respectively. It turned out that the Spatial-Markov model could achieve very good performance for short period forecasting. For the case study, it was quite qualified for the prediction of three time steps (up to 15 years) or more within which the results had much high reliability, however, time-duration of forecasting had much significant impact on the model's performance, the longer the forecasting duration, the lower the model's accuracy.

Multi‐scenario simulation of land use change based on MCR‐SD‐FLUS model: A case study of Nanchang, China

“…The study of LUCC simulation at different temporal and spatial scales plays an important role in revealing the impact of human activities on regional ecological environment changes (Fan et al, 2022; Foley et al, 2005; Verburg et al, 2002). Currently, LUCC simulation is mainly implemented through the following models: (1) Quantitative prediction models, such as system dynamics (SD) model (Geng et al, 2017; Portela & Rademacher, 2001), Markov models (Hou et al, 2015; Wu et al, 2006), etc. These models predict the changes and change rates of various land use types on a macro‐scale, but their spatial distribution is not considered (Cao et al, 2019).…”

Forecasts of Urban Construction Land Scale based on Driving Force Analysis

“…The basic idea of direct prediction is assuming that land use scale growth in the future in line with the historical trend, combined with the previous land use data and variation characteristics, to predict the future scale of land use through a variety of parameters in the model fitting, such as exponential growth model [1], grey system model [2], Markov Model [3]. The basic idea of indirect prediction is to using driving factors of construction land expansion to predict the future scale, such as per capita multivariate statistical regression model [4], neural network model and genetic support vector machine [5].…”