Spatio-Temporal Land Cover Change Detection using Machine Learning and Google Earth Engine

The combination of Google Earth Engine
ARIMA time-series exploration with Random Forest, a
powerful machine learning technique, to offer a novel hybrid
solution for spatio-temporal land cover change detection. The
proposed method aims to address the challenges of identifying
variations in land cover accurately and efficiently over
extensive periods. To produce accurate categorization results,
the Random Forest model is conditioned on past land cover
data during the first step of the hybrid approach. The second
section examines the temporal patterns of different land cover
classes using Google Earth Engine ARIMA time-series analysis
to determine trends and patterns in land cover changes. The
new approach, which marries Random Forest with ARIMA
time-series analysis, leverages both methods' strengths to
provide a more accurate and robust spatio-temporal detection
of changes. The combined approach can be employed for large
regional and also global surface area monitoring programs
because of its ability to be expanded and its efficacy. This
approach has real-world applications in environmental
management, conservation, city planning based on an intimate
understanding of land cover change and their consequences. In
studies of spatiotemporal LST variations, the MODIS land
surface temperature (MOD11A1) component has been used
widely. However, it has a limited suite of uses, partly due to the
low resolution of space. To transcend this, the paper uses the
proposed ARIMA-RF approach. It achieves accuracy of
approximately 99.5%.
Keywords— AutoRegressive Integrated Moving Average,
Google Earth Engine; Spatio temporal changes in land cover;
Random Forest; Time series investigation