Modelling of land-use change in Thailand using binary logistic regression and multinomial logistic regression

Modelling of land-use change over time gives useful information relating to land development. Logistic regression is arguably one of the most used in the land-use study when a binary outcome variable. Multinomial logistic regression used when multinomial outcome variable. This study aims to model of land-use change using logistic regression and multinomial logistic regression. These methods were applied to data from a survey on the land-use of Krabi Province, Thailand, in 2000, 2009 and 2018. The data is based on an analogue-to-digital conversion which replaces polygonal shapes by coded grid points. The land-use observations were categorized into four, namely developed, rubber plantation, agricultural and undeveloped lands. Our findings show that developed and agricultural lands increased by two periods of study 2000–2009 and 2009–2018. Whereas rubber plantation was decreased on two periods observes. The logistic regression models for developed land, agricultural land and rubber plantation were used for modelling land-use change, consequently as the multinomial logistic regression model with the undeveloped land being the reference category. The results showed that the modelling of land-use in 2009 and 2018 with logistic regression and multinomial logistic regression does not differ by the area under the curves (AUC) and prediction accuracy. Therefore, logistic regression can be used for the modelling of the land-use change using land-use in the previous year as predictors instead of the multinomial logistic regression, since it is not a complicated formula for adjusting spatial correlations using the variance inflation factors (VIF) and displaying the modelling result on 95% confidence intervals (CI) based weighted sum contrasts.