Bayesian Change Point Detection of Vegetation Cover Dynamics of Akure Forest Reserve, Ondo State in Southwestern, Nigeria

P.A Ukoha,S.J Okonkwo,A.R Adewoye

doi:10.4314/jasem.v25i8.25

Abstract

This study uses satellite acquired vegetation index data to monitor changes in Akure forest reserve. Enhanced Vegetation Index (EVI) time series datasets were extracted from Landsat images; extraction was performed on the Google Earth Engine (GEE) platform. The datasets were analyzed using Bayesian Change Point (BCP) to monitor the abrupt changes in vegetation dynamics associated with deforestation. The BCP shows the magnitude of changes over the years, from the posterior data obtained. BCP focuses on changes in the long‐range using Markov Chain Monte Carlo (MCMC) methods, this returns posterior probability at > 0.5% of a change point occurring at each time index in the time series. Three decades of Landsat data were classified using the random forest algorithm to assess the rate of deforestation within the study area. The results shows forest in 2000 (97.7%), 2010 (89.4%), 2020 (84.7%) and non-forest increase 2000 (2.0%), 2010 (10.6%), 2020 (15.3%). Kappa coefficient was also used to determine the accuracy of the classification.

Highlights

Forest degradation has become a global concern especially forest conversion to non-forest
Enhance Vegetation Index (EVI): Enhance Vegetation Index of time series was extracted in Google Earth Engine (GEE) using Landsat collection of images which covers 30m resolution for every 16 days
1 chosen so that this method is very effective in situations when there aren’t too many changes (p0 is small) and when the changes are of reasonable size (w0 is small) (Barry and Hartigan, 1993)

Summary

MATERIALS AND METHODS

Study Area: Akure forest reserve is located in Akure South Local government area of Ondo State in Southwestern, Nigeria (Figure 1). Enhance Vegetation Index (EVI): Enhance Vegetation Index of time series was extracted in GEE using Landsat collection of images which covers 30m resolution for every 16 days. EVI was used because of its sensitivity to changes in areas with high biomass and dense vegetation It reduces the influence of atmospheric conditions on vegetation index and correct for canopy background noise. The EVI time series extracted was used to detect changes and breakpoints using Bayesian Change and Change Point packages in R-studio. The transition probability, p, for the conditional probability of a change at the position i + 1, is obtained from the simplified ratio presented in Barry and Hartigan, (1993): Pi = P(Ui=1⎢ X,Uj,j ≠i ). 1 chosen so that this method is very effective in situations when there aren’t too many changes (p0 is small) and when the changes are of reasonable size (w0 is small) (Barry and Hartigan, 1993)

AND DISCUSSION

Conclusion