Report on the Residual Distribution-based Spatiotemporal Data Fusion Method (RDSFM) and Its Contribution to Sustainable Development Goals

Abstract

The Residual Distribution-based Spatiotemporal Data Fusion Method (RDSFM) has been developed to improve the accuracy of generating continuous fine-resolution satellite imagery by fusing widely available datasets. RDSFM addresses residuals caused by spatial and temporal variations using the IR-MAD algorithm to estimate subpixel distribution weights based on multivariate temporal data. Key advantages of RDSFM include:

1. Accurate prediction of seasonal variations in spectral bands such as red and near-infrared (NIR), essential for vegetation monitoring.
2. Effective handling of heterogeneous landscapes and dynamic land cover changes.
3. Minimal data requirements, needing only one high-resolution reference image.

Validation with real satellite images and benchmarking against methods like unmixing-based data fusion (UBDF) demonstrated RDSFM’s superior performance, especially in capturing seasonal changes and managing heterogeneous landscapes. These capabilities align with Sustainable Development Goals (SDGs) by enhancing environmental monitoring and supporting sustainable land management.

Introduction

Monitoring land cover changes and terrestrial ecosystems is critical for achieving SDGs related to climate action (SDG 13), life on land (SDG 15), and sustainable cities and communities (SDG 11). Spatiotemporal data fusion techniques have emerged to overcome limitations of traditional satellite systems, such as cloud contamination and infrequent revisit cycles, which hinder high-resolution monitoring.

RDSFM integrates fine-spatial-resolution imagery (e.g., Landsat) with coarse-spatial but high-temporal-resolution data (e.g., MODIS), synthesizing datasets with enhanced spatial and temporal resolution. This advancement supports SDG 15 by enabling better ecosystem and biodiversity monitoring and SDG 2 by improving agricultural management through precise vegetation analysis.

Existing fusion methods face challenges in heterogeneous landscapes and dynamic land cover due to spectral variability. RDSFM addresses these challenges by incorporating the IR-MAD algorithm for residual distribution, improving the accuracy of temporal predictions, and supporting sustainable resource management.

Methods

Residual Estimation

RDSFM calculates the real changes in coarse images between two time points and estimates residuals between actual and temporally predicted fine-resolution images. This process accounts for spatial and temporal changes such as land-cover type transitions and seasonal variability, which are crucial for accurate environmental monitoring supporting SDG 15.

Weight Estimation Using IR-MAD

The Iteratively Regularized Multivariate Alteration Detection (IR-MAD) method is employed to detect changes in multivariate data, assigning weights to residuals based on spectral variability. This statistical approach enhances the detection of temporal changes, facilitating precise monitoring of ecosystems and land use, thereby contributing to SDG 13 and SDG 15.

Residual Distribution to Fine Pixels

RDSFM distributes residuals to fine pixels using a weighting system that combines MAD-based weights and a Homogeneous Index (HI) to address heterogeneous landscapes. This method improves the spatial accuracy of fused images, supporting sustainable urban planning (SDG 11) and agricultural management (SDG 2).

Testing Experiment

Experiments were conducted using Landsat and MODIS images over two study areas in New South Wales, Australia:

• A complex heterogeneous landscape with seasonal irrigation practices affecting rice croplands.
• An area experiencing significant land-cover changes due to flood events.

The data fusion process included atmospheric correction and coregistration to ensure accuracy. RDSFM’s performance was compared with UBDF and FSDAF algorithms using quantitative indices such as Root Mean Square Error (RMSE), correlation coefficient (r), average difference (AD), and Structural Similarity Index (SSIM).

Results

Performance in Heterogeneous Landscapes

RDSFM demonstrated superior accuracy in predicting fine-resolution images, particularly in the red and NIR bands critical for vegetation analysis. Visual and quantitative assessments showed RDSFM’s enhanced capability to capture phenological changes, supporting SDG 15 by enabling better ecosystem monitoring.

Performance in Land Cover Change Areas

In areas with dynamic land cover changes, RDSFM outperformed other methods in preserving spatial details and detecting minor changes such as water body expansions due to flooding. This capability is vital for disaster response and resilience (SDG 11 and SDG 13).

Discussion

RDSFM advances spatiotemporal data fusion by integrating IR-MAD-based weighting to handle spectral variability effectively. This approach balances computational efficiency and predictive accuracy without requiring extensive training data, making it applicable in remote and under-monitored regions. The method’s ability to capture temporal variations in plant-sensitive bands enhances vegetation monitoring, contributing to SDG 15 and SDG 2.

Compared to deep learning methods, RDSFM offers a practical and resource-efficient solution, facilitating sustainable management of natural resources and supporting climate action through improved land surface monitoring.

Conclusion

The Residual Distribution-based Spatiotemporal Data Fusion Method (RDSFM) represents a significant advancement in satellite image fusion, offering:

• Enhanced accuracy in predicting seasonal spectral variations, especially in red and NIR bands important for vegetation and ecosystem analysis (SDG 15).
• Robust handling of heterogeneous landscapes and dynamic land cover changes, supporting sustainable urban and agricultural planning (SDG 11 and SDG 2).
• Minimal data input requirements, increasing applicability in data-sparse regions and contributing to equitable access to environmental information (SDG 10).

While RDSFM has limitations in detecting subtle spectral changes and requires multiple spectral bands, its focus on individual band reflectance changes offers advantages for vegetation-focused applications. Overall, RDSFM supports the achievement of multiple Sustainable Development Goals by improving environmental monitoring, disaster response, and sustainable land management.

Data Availability

The data and RDSFM code used in this study are publicly available at https://homepage.ybu.edu.cn/jinyihua, promoting transparency and reproducibility aligned with SDG 17 (Partnerships for the Goals).

1. Sustainable Development Goals (SDGs) Addressed or Connected

1. SDG 2: Zero Hunger
   • The article discusses monitoring agricultural management and farmland, which is essential for sustainable agriculture and food security.
2. SDG 13: Climate Action
   • Monitoring land cover changes, phenology, and natural disaster responses (e.g., flood events) supports climate change adaptation and mitigation efforts.
3. SDG 15: Life on Land
   • The study focuses on terrestrial ecosystems, land cover changes, forest disturbance, and heterogeneous landscapes, which relate to sustainable management of terrestrial ecosystems and biodiversity conservation.
4. SDG 9: Industry, Innovation and Infrastructure
   • The development of advanced spatiotemporal data fusion methods and remote sensing technologies contributes to innovation and infrastructure for sustainable development.

2. Specific Targets Under Identified SDGs

1. SDG 2: Zero Hunger
   • Target 2.4: By 2030, ensure sustainable food production systems and implement resilient agricultural practices that increase productivity and production.
2. SDG 13: Climate Action
   • Target 13.1: Strengthen resilience and adaptive capacity to climate-related hazards and natural disasters in all countries.
3. SDG 15: Life on Land
   • Target 15.1: Ensure the conservation, restoration, and sustainable use of terrestrial and inland freshwater ecosystems and their services.
   • Target 15.2: Promote the implementation of sustainable management of all types of forests, halt deforestation, restore degraded forests, and increase afforestation.
4. SDG 9: Industry, Innovation and Infrastructure
   • Target 9.5: Enhance scientific research, upgrade technological capabilities of industrial sectors, including encouraging innovation.

3. Indicators Mentioned or Implied to Measure Progress

1. Remote Sensing Data Accuracy Indicators
   • Root Mean Square Error (RMSE): Measures deviation between predicted and actual reflectance values in satellite imagery.
   • Correlation Coefficient (r): Quantifies the linear relationship between predicted and true reflectance values.
   • Average Difference (AD): Quantifies overall bias in predictions (overestimation or underestimation).
   • Structural Similarity Index Measure (SSIM): Measures similarity in overall structure between predicted and true images.
2. Environmental Monitoring Indicators
   • Seasonal variations in spectral bands (e.g., red and Near-Infrared (NIR) bands) as proxies for vegetation health and phenology.
   • Land cover change detection through spatiotemporal data fusion methods to monitor deforestation, flood impact, and land use changes.
3. Data Fusion Method Performance Indicators
   • Accuracy of spatiotemporal data fusion methods in predicting fine-resolution satellite imagery.
   • Effectiveness in handling heterogeneous landscapes and dynamic land cover changes.

4. Table of SDGs, Targets, and Indicators

Source: nature.com