TL;DR
This paper introduces a novel "Stratified Land Use Change Simulation Model" (SLUCS) that uses elevation gradients to stratify study areas. This approach significantly improves the accuracy and spatial consistency of land-use change projections compared to traditional methods, offering a more nuanced understanding of regional land-use dynamics.
What Problem Does This Paper Address?
Land use change (LUC) is a critical driver of global environmental processes, but accurately simulating it is challenging due to the complex interplay of factors that vary spatially. Traditional models often apply uniform rules across entire regions, failing to capture the distinct characteristics and dynamics present at different scales, particularly in heterogeneous landscapes like mountainous areas. This lack of scale-awareness leads to uncertainties and inaccuracies in projections, hindering effective land-use planning and decision-making. The authors specifically address the need to account for scale differences, especially within elevation gradients, which significantly influence climate, soil, vegetation, and human activities.
Methods
The core innovation is the Stratified Land Use Change Simulation Model (SLUCS), which integrates an elevation-based stratification strategy. This strategy involves:
- Elevation-Based Stratification Module: This module uses an object-based segmentation algorithm (from eCognition software) applied to elevation data to divide the study area into spatially contiguous "stratifications." The segmentation is optimized by balancing intra-segment homogeneity (similar land-use characteristics within a stratification) and inter-segment heterogeneity (distinct characteristics between stratifications) using metrics like ASDintra and LVinter, and an Optimal Segment Score (OSS). This results in stratifications that reflect elevation gradients and associated land-use patterns.
- Non-Spatial Land-Use Demand Module: This module forecasts overall land-use demands using a Markov chain model to capture historical trends and then designs future scenarios based on different land-use priorities (e.g., economic development vs. ecological protection). The demands are then spatially disaggregated based on the generated stratifications.
- Stratified Suitability Estimation Module: This module employs a modified logistic regression model. Crucially, it incorporates dummy variables representing the stratifications to quantify the relationship between driving factors (climate, soil, topography, socio-economics, etc.) and land use within each specific stratification. This allows for the identification of regionally varying influences.
- Spatial Allocation Module: This module projects the spatially explicit land-use changes by combining the stratified local suitability, conversion resistance (cost of changing land use), and competitive advantage of different land-use types. An iterative process ensures the aggregated allocated land-use quantities match the non-spatial demands.
Datasets
The study utilized several datasets for the Guizhou and Guangxi Karst Mountainous Region in China:
- Land-Use/Cover Datasets: Provided by the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences. Reclassified into seven categories: Paddy field, dry land, forest, grassland, water, built-up land, and bare land.
- Driving Factors:
- Climatic: Mean annual temperature and precipitation (0.1° x 0.1° resolution, 1981-2000).
- Soil: Soil type, ratio of clay/sandy soil, soil organic matter (1 km x 1 km, 1980s).
- Vegetation: Vegetation types (1 km x 1 km, 1980s) and MODIS NDVI (250 m x 250 m, 2001-2005).
- Topographic: Elevation and slope (90 m x 90 m, 2000), Aspect (90 m x 90 m, 2000).
- Land Degradation: Karst rocky desertification (100 m x 100 m, 2005).
- Socio-economic: Population density and GDP (1 km x 1 km, 2000).
- Location: Distance to roads, settlements, and rivers (1 km x 1 km, 2000).
Temporal data for forecasting 2030 included updated temperature/precipitation (1995-2015), NDVI (2011-2015), population density, GDP (2015), and roads (2015). All data were resampled to a 1000 m x 1000 m resolution.
Key Insights and Ideas
- Elevation as a Stratification Driver: The study demonstrates that elevation gradients are a powerful and objective basis for stratifying landscapes, capturing significant variations in environmental conditions and land-use patterns that are missed by uniform models.
- Optimizing Segmentation: The development of quantitative metrics (ASDintra, LVinter, OSS) to determine the optimal scale for segmentation is a key methodological contribution, ensuring stratifications are neither too coarse nor too fine.
- Stratified Relationships: The core insight is that the influence of driving factors on land use is not uniform. By stratifying, the model reveals how factors like population density or climate impact land use differently across elevation zones, leading to more accurate local predictions.
- Improved Spatial Consistency: The stratified approach leads to a more realistic spatial distribution of land-use changes, particularly for complex conversions like urban expansion, compared to traditional methods that can over-disperse or over-generalize.
Key Results
- Accuracy Improvement: The SLUCS model achieved a higher kappa coefficient (0.89) compared to the traditional method (0.83) for simulating land use in 2015.
- Better Simulation of Change: The KSimulation metric, which specifically assesses performance in areas of land-use change, showed a substantial improvement with the stratified model (0.52) versus the traditional model (0.19).
- Enhanced Built-up Land Simulation: The stratified method demonstrated significantly better spatial fitness (higher Modified Lee and Sallee metrics) for simulating built-up land expansion, a critical urban growth process.
- Scenario Projections: Three scenarios (historic-condition, planning, protect) for 2015-2030 projected similar trends (built-up land and forest increase, others decrease) but with varying magnitudes. The "planning" and "protect" scenarios showed sharper increases in built-up land and forest, respectively, highlighting the impact of different policy priorities.
Limitations and Future Work
- Local Area Performance: The authors acknowledge that the model's performance in very small conversion areas, particularly for built-up land, was still lower than desired, with ROC values for built-up land being relatively modest (0.78).
- Data Resolution and Proxies: The use of aggregated data and proxies for complex socio-economic factors (like human demands) at a coarse resolution (1000m) limits the model's ability to fully capture fine-scale dynamics.
- Segmentation Optimization: While the OSS method was used, the authors suggest testing other environmental variables for stratification beyond just elevation.
- Future Work: The authors propose testing the SLUCS model in other regions, acquiring more detailed socio-economic data, improving the stratification process with multiple environmental variables, and updating more driving factors for future simulations to enhance projection accuracy.
Summary updated