TL;DR
This paper introduces a novel approach for monitoring land degradation by combining expert knowledge with a Backpropagation Artificial Neural Network (BP-ANN) and remote sensing data (NDVI and albedo). This method achieves high accuracy (91.2%) and offers a more efficient and reliable way to track land degradation dynamics, particularly in regions like the Ordos Plateau, China.
What Problem Does This Paper Address?
Land degradation is a critical global environmental challenge, yet there's a lack of consensus on effective monitoring methods. Existing approaches, like image classification and vegetation index-based methods, have limitations. Image classification is often labor-intensive and subjective, while vegetation indices (like NDVI) can be influenced by factors other than degradation, requiring extensive fieldwork for validation. This paper aims to bridge this gap by developing a more robust, efficient, and accurate method for detecting and monitoring land degradation, especially in arid and semi-arid regions.
Methods
The proposed approach integrates three key components:
- Expert Knowledge Extraction: Domain experts (co-authors with extensive experience) visually interpret Landsat TM images to identify land degradation degrees and their relationship with predisposing factors like NDVI and albedo. This process involves repeated interpretations and blind tests to minimize subjectivity.
- Land Degradation Detecting Model: A BP-ANN algorithm is employed to simulate the experts' decision-making logic. This network learns the complex relationship between the extracted expert knowledge (represented by a Land Degradation Index, LDI, derived from NDVI and albedo) and the actual land degradation degrees. The BP-ANN is trained to establish a predictive model.
- Dynamic Analysis: A degradation dynamic (DD) index is used to quantify changes in land degradation over time, allowing for the analysis of temporal trends and spatial shifts.
The core novelty lies in the synergistic use of expert knowledge to guide the BP-ANN, overcoming the limitations of purely data-driven or purely expert-driven approaches.
Datasets
- Landsat TM Images: Acquired in the summers of 1990, 2000, and 2010, with a resolution of 30m. These were preprocessed (radiometric calibration, atmospheric correction, mosaicking, clipping).
- Field Survey Data (2011): Approximately 500 sample points were collected, measuring NDVI, vegetation coverage, and land use type. This data was used for validating the land degradation detection accuracy.
The paper does not introduce a new benchmark dataset but rather uses existing remote sensing data and field measurements for validation.
Key Insights and Ideas
- Expert Knowledge is Crucial: The study highlights that while remote sensing indices are valuable, they are insufficient on their own. Integrating expert knowledge, which captures nuanced understanding and practical experience, is essential for accurate land degradation detection.
- NDVI and Albedo as Feasible Indicators: The research confirms that NDVI (reflecting vegetation cover) and albedo (related to surface soil moisture and roughness) are indeed useful indicators for land degradation, especially when combined.
- BP-ANN as a Powerful Tool: The BP-ANN effectively bridges the gap between expert knowledge and remote sensing data, creating a robust model that can generalize and predict land degradation with high accuracy, even with limited initial samples.
- Reversal of Degradation Trend: A significant finding is the observed reversal of land degradation trends in the Ordos Plateau from 1990-2010, suggesting the positive impact of ecological restoration projects and land use optimization.
Key Results
- High Accuracy: The proposed expert knowledge and BP-ANN-based approach achieved an overall accuracy of 91.2% in land degradation detection.
- Temporal Trend Reversal: From 1990 to 2010, the Ordos Plateau showed a significant reduction in high and severe land degradation, with an increase in non-degraded and low-degraded areas.
- Spatial Dynamics: Regions with high land degradation dynamics were primarily located in the northeast of the Ordos Plateau. However, most areas transitioned from "hot spots" of degradation to less dynamic regions.
- Reduced Workload: The method significantly reduces the workload associated with traditional methods, particularly by minimizing the need for extensive fieldwork.
Limitations and Future Work
- Subjectivity and Regional Dependency: The authors acknowledge that expert knowledge, despite efforts to mitigate subjectivity through repeated testing, can still have regional dependencies.
- Indicator Limitations: The study suggests that relying solely on NDVI and albedo might not be sufficient in all cases, as distinguishing between lightly degraded and non-degraded lands, or between healthy vegetation and degraded land with high cover, can be challenging.
- Future Work Suggestions: The paper proposes incorporating additional indicators, such as human appropriation of net primary productivity (HANPP), to further enhance the model's ability to reflect ecosystem productivity changes caused by land degradation. They also suggest that the approach could be adapted for other degraded areas with the involvement of local experts.
Summary updated