Intensity and Stationarity Analysis of Land Use Change Based on CART Algorithm

Xiao Sang, Qiaozhen Guo, Xiaoxu Wu, Ying Fu, Tongyao Xie, Chengwei He, Jinlong ZangScientific 2019 · DOI

TL;DR

This paper introduces new models for analyzing land use change intensity and stationarity at multiple levels (interval, category, and transition) using a CART algorithm for classification. The study reveals that Tianjin experienced rapid urban development between 1995 and 2015, with urban land increasing by 7.5% at the expense of agricultural land, offering insights for sustainable urban planning.

What Problem Does This Paper Address?

Understanding how land use changes over time is crucial for sustainable development, as it directly impacts environmental issues like land degradation and climate change. While various remote sensing data and classification methods exist, a comprehensive and systematic approach to quantifying the intensity and stationarity of these changes across different scales (time intervals, land use categories, and transitions between them) has been lacking. This paper aims to fill this gap by developing and applying novel models to analyze land use dynamics in Tianjin, China, providing a more nuanced understanding of the patterns and drivers of change.

Methods

The study employs a multi-step approach:

  1. Image Preprocessing: Standard steps like radiometric calibration, atmospheric correction, band merging, and image mosaicking were performed on remote sensing data.
  2. Land Use Classification: The CART (Classification and Regression Trees) algorithm was used for decision tree classification to extract land use information. This is a well-established machine learning technique for image classification.
  3. Accuracy Assessment: The classification accuracy was evaluated using producer's accuracy, user's accuracy, overall classification accuracy, and the Kappa coefficient, ensuring the reliability of the land use maps.
  4. Intensity and Stationarity Analysis: This is the core methodological contribution. The authors adapted and extended the intensity analysis framework proposed by Pontius et al. They developed new models to quantify:
  5. Interval Level: Overall land use change intensity and uniform intensity to assess the speed and stability of change within specific time periods.
  6. Category Level: Annual gain and loss intensity for each land use category to identify "active" vs. "dormant" changes.
  7. Transition Level: New models were developed to analyze the intensity and contribution of specific transitions (e.g., agricultural land to urban land). This includes analyzing gains and losses from different perspectives.

Datasets

  • Remote Sensing Data:
    • Thematic Mapper (TM) images for the years 1995, 2000, 2005, and 2010.
  • Operational Land Imager (OLI) image for 2015.
  • These images were acquired from the USA Geological Survey (http://glovis.usgs.gov).
  • Ancillary Data:
    • Administrative map of Tianjin for clipping images.
  • Textual data from the Tianjin Statistics Bureau (http://stats.tj.gov.cn).

The paper does not introduce new benchmark datasets; it utilizes publicly available satellite imagery and administrative data.

Key Insights and Ideas

  • Beyond Simple Change Detection: The paper moves beyond simply identifying that land use changed to quantifying how intensely and how stably it changed, and which transitions are most significant.
  • Urbanization as a Dominant Force: The study clearly illustrates the rapid pace of urbanization in Tianjin, with urban land expanding significantly. This highlights the pressure on other land types, particularly agricultural land.
  • The Importance of Transition Analysis: The novel models for transition analysis provide a deeper understanding of the "flow" between land use types. For instance, understanding that agricultural land is the primary source for urban expansion, but only accounts for a portion of its total loss, offers crucial insights for policy.
  • Stationarity as a Predictor: The concept of stationarity (whether changes are evenly distributed or concentrated) is presented as a key indicator for predicting future land use trends and informing sustainable development strategies. Stable transitions are more predictable.

Key Results

  • Overall Land Use Change: Between 1995 and 2015, Tianjin experienced a significant increase in urban land (approx. 7.5%) and a corresponding decrease in agricultural land (approx. 7.5%). Forest, water, and other land saw slight increases.
  • Interval Level Intensity: Land use change was most intense during the 1995-2000 period (4.17% annual intensity), coinciding with policy changes related to housing reform. Subsequent periods showed slower change rates.
  • Category Level Intensity: "Other land" and "water" showed the most active changes, while "forest" and "agricultural land" were more dormant. Urban land also showed active change.
  • Transition Level Insights:
    • Agricultural land was the primary source for urban land expansion, accounting for a significant portion of the land transformed into urban areas.
  • However, the transition from agricultural land to urban land represented only about 45% of the total loss of agricultural land, indicating other significant transition pathways.
  • Transitions from agricultural land to water were also notable.
  • Classification Accuracy: The land use classification maps achieved good accuracy, with overall classification accuracy ranging from 88.40% to 96.40% and Kappa coefficients indicating "great" to "excellent" classification quality.

Limitations and Future Work

  • Limitations:
    • The study acknowledges that while land use intensity analysis provides detailed insights into change, its impact on the human society and natural ecological environment needs further investigation.
  • The remote sensing data for different years were not acquired on the exact same date, and precipitation variations could introduce errors, particularly for water area estimations.
  • Future Work:
    • Further research is needed to explore the societal and environmental impacts of the observed land use changes.
  • Methods to mitigate the impact of varying acquisition dates and precipitation on remote sensing data quality, especially for water bodies, should be investigated.

Summary updated

Abstract

Land use directly reflects degree of human development and utilization of land. Intensity analysis of land use is a quantitative method to analyze land use changes. In this paper, land use changes in Tianjin were studied using Thematic Mapper (TM) remote sensing images in 1995, 2000, 2005, 2010 and Operational Land Imager (OLI) remote sensing image in 2015. Land use information was extracted using decision tree classification method based on CART (Classification and Regression Trees) algorithm. This paper introduced land use intensity analysis to analyze its change intensity and stationarity, respectively at interval, category and transition levels. Based on the theory, new models were developed in the transition level to analyze land use change pattern. The analysis quantifies the contribution of a certain land categories to land use change during a specific time interval. The change of land use during 1995-2015 indicated that Tianjin experienced rapid urban development with the area of urban land increased by about 7.5%. This study provided a reference for the sustainable development of land use in Tianjin.
BibTeX
@inproceedings{sang2019intensity,
  author = {Xiao Sang and Qiaozhen Guo and Xiaoxu Wu and Ying Fu and Tongyao Xie and Chengwei He and Jinlong Zang},
  title = {{Intensity and Stationarity Analysis of Land Use Change Based on CART Algorithm}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-48586-3},
}
Figure