I, Amin Naboureh, am thrilled to be the first author of this paper (Figure 1) and eager to share the story behind our publication in the Scientific Data journal. Our research, led by Professor Ainong Li, is the result of a collaborative effort from all the authors involved. With the generous support of the National Natural Science Foundation of China and the Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, we have dedicated nearly three years to creating a dataset that not only provides insights into the land cover of the China Central-Asia West-Asia Economic Corridor region but also has the potential to promote sustainable development and environmental planning.
Unveiling the Land Cover Puzzle
Land cover maps are a treasure trove of knowledge, offering insights into various fields, from sustainable development to climate change. These maps hold the key to understanding the ever-evolving Earth, where land cover changes wield an incredible influence on natural processes like the hydrological cycle, ecological balance, and ecosystem services. Yet, despite their importance, existing global land cover products faced limitations that hindered their effectiveness in long-term land cover change monitoring.
A Region in Need
Our study focused on the China Central-Asia West-Asia Economic Corridor, a crucial component of the Belt and Road Initiative (BRI) program, spanning eight countries in arid and semi-arid climates. This region had been witnessing severe land cover changes, exemplified by the Aral Sea crisis and the shrinkage of Lake Urmia. The Aral Sea, in Central Asia, had lost over 80% of its surface area in recent decades, and Lake Urmia in Iran had similarly shrunk by 80% since 2000. These drastic changes posed a significant threat to both human lives and ecosystems, necessitating a precise long-term land cover dataset for the entire corridor. Unfortunately, aside from China, such a dataset was conspicuously absent for the other countries in the region.
The Birth of Our Solution
The landscape of earth observation data underwent a transformation with Landsat's free data sharing policy in 2008 and the rise of Earth science data cloud computing and analysis tools, such as Google Earth Engine (GEE). It was with this backdrop that our research group embarked on a monumental undertaking. We constructed a historical set of six 30-meter resolution land cover maps between 1993 and 2018 at 5-year intervals. This involved processing nearly 200,000 Landsat scenes, offering a comprehensive view of land cover dynamics for the seven countries in the region.
Our methodology, executed on the GEE cloud processing platform, eliminated the need for extensive data downloads and enabled pixel-wise analysis. The process consisted of six main steps:
1.Study Area Subdivision
2.Satellite Data Acquisition and Pre-processing
3.Input Features Generation
4.Reference Sample Data Collection
5.Supervised Classification (Adaptive Random Forest Scheme)
Validation and Accuracy
We used a combination of statistical and visual assessments to validate our generated land cover maps. Visual interpretation demonstrated that the maps were free from noise and accurately depicted all nine land cover classes. Statistically, we employed metrics like Overall Accuracy (OA), User's Accuracy (UA), Producer's Accuracy (PA), and the F1-score to gauge the performance of our maps. The average OA of the generated six land cover maps was an impressive 91.4%, with the highest OA of 92.7% recorded in 2003 and the lowest at 90.3% in 2018. Different land cover classes displayed varying accuracy levels, with Water and Snow classes showing the highest accuracy values (>95%) and Grassland and Cropland showing slightly lower values (>85%). While our research has demonstrated a high level of accuracy, it's essential to acknowledge the potential for errors in any dataset. Therefore, users should consider possible error/noise when interpreting the LC maps.
The resulting land cover dataset is the first of its kind for one of the BRI’s Corridor, offering a comprehensive view of land cover dynamics for seven countries since 1993. This dataset has a multitude of applications. For instance, in light of the significant shrinkages in surface water area in the region, it can serve as a valuable tool for analyzing surface water changes and identifying their driving factors. By sharing this dataset, we aim to facilitate data-driven decision-making, promote transparency and accountability, and encourage collaboration among the BRI countries.