Near Real Time Change-Point detection in Optical and Thermal Infrared Time Series Using Bayesian Inference over the Dry Chaco Forest

V. BARRAZA; H. KARSZENBAUM; F. GRINGS; P. PERNA; E. ROITBERG

New Orleans

Congreso; AGU Fall Meeting 2017; 2017

American Geophysical Union

The Dry Chaco region (DCF) has the highest absolute deforestation rates of all Argentinian forests. The most recent report indicates a current deforestation rate of 200,000 Ha year−1. In order to better monitor this process, DCF was chosen to implement an early warning program for illegal deforestation. Although the area is intensively studied using medium resolution imagery (Landsat), the products obtained have a yearly pace and therefore unsuited for an early warning program.In this paper, we evaluated the performance of an online Bayesian change-point detection algorithm for MODIS Enhanced Vegetation Index (EVI) and Land Surface Temperature (LST) datasets. The goal was to to monitor the abrupt changes in vegetation dynamics associated with deforestation events. We tested this model by simulating 16-day EVI and 8-day LST time series with varying amounts of seasonality, noise, length of the time series and by adding abrupt changes with different magnitudes. This model was then tested on real satellite time series available through the Google Earth Engine, over a pilot area in DCF, where deforestation was common in the 2004-2016 period. A comparison with yearly benchmark products based on Landsat images is also presented (REDAF dataset).The results shows the advantages of using an automatic model to detect a changepoint in the time series than using only visual inspection techniques. Simulating time series with varying amounts of seasonality and noise, and by adding abrupt changes at different times and magnitudes, revealed that this model is robust against noise, and is not influenced by changes in amplitude of the seasonal component. Furthermore, the results compared favorably with REDAF dataset (near 65% of agreement). These results show the potential to combine LST and EVI to identify deforestation events.This work is being developed within the frame of the national Forest Law for the protection and sustainable development of Native Forest in Argentina in agreement with international legislation (REDD+).