Fusion of SAR and MSI Dataset for Rice Mapping using Google Earth Engine (GEE)

The fusion of Synthetic Aperture Radar (SAR) and Multi-Spectral Imagery (MSI) data is a powerful technique for agricultural monitoring, including rice mapping. SAR data is sensitive to surface roughness and moisture content, while MSI provides valuable information on vegetation health, phenology, and chlorophyll content. Combining these data sources enhances the ability to accurately classify and monitor rice fields, especially in areas with cloud cover or during key phenological stages of rice growth.

1. Why Fuse SAR and MSI for Rice Mapping?

• SAR (Synthetic Aperture Radar):

1. Cloud Penetration: SAR can capture data regardless of weather conditions, making it invaluable in areas with persistent cloud cover.
2. Sensitivity to Moisture: SAR is highly sensitive to water content and roughness, making it suitable for monitoring flooded fields and rice paddies.
3. Temporal Consistency: SAR provides continuous data, enabling effective monitoring of the dynamic growth stages of rice.

• MSI (Multi-Spectral Imagery):

1. Vegetation Health Monitoring: MSI from optical sensors like Sentinel-2 or Landsat includes bands sensitive to vegetation health, such as NDVI (Normalized Difference Vegetation Index) and EVI (Enhanced Vegetation Index).
2. Spectral Reflectance: It provides information about leaf area index, chlorophyll content, and vegetation vigor during the growth phases of rice.

2. Data Sources on Google Earth Engine (GEE):

• SAR Data:

1. Sentinel-1: C-band SAR data available on GEE, offering dual polarization (VV and VH) and regular acquisitions (every 6-12 days) over many regions.

• MSI Data:

1. Sentinel-2: MSI data that includes bands in visible, near-infrared (NIR), and short-wave infrared (SWIR) regions, with a revisit time of 5 days and a spatial resolution of 10 to 20 meters.

3. Methodology for Rice Mapping Using SAR and MSI Fusion:

Step 1: Preprocessing SAR Data

• Filtering & Calibration: Apply speckle filtering (e.g., Lee filter) and radiometric calibration to convert SAR data into backscatter intensity in decibels (dB).
• Terrain Correction: Apply terrain correction (orthorectification) using the GEE sentinel1 pre-processing functions to account for terrain-induced distortions.
• Polarization Selection: Use both VV and VH polarization for enhanced feature extraction related to water and vegetation.

// Load and preprocess Sentinel-1 SAR data
var sentinel1 = ee.ImageCollection('COPERNICUS/S1_GRD')
                 .filterBounds(region)
                 .filterDate('2023-01-01', '2023-12-31')
                 .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))
                 .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VH'))
                 .filter(ee.Filter.eq('orbitProperties_pass', 'DESCENDING'))
                 .map(function(image) {
                   return image.select(['VV', 'VH']).divide(10).log10();
                 });

Step 2: Preprocessing MSI Data

• Cloud Masking: Use the built-in cloud masking function to remove cloud-contaminated pixels from Sentinel-2 imagery.
• Vegetation Indices: Calculate vegetation indices such as NDVI or EVI to track rice growth phases.

// Load and preprocess Sentinel-2 MSI data
var sentinel2 = ee.ImageCollection('COPERNICUS/S2_SR')
                 .filterBounds(region)
                 .filterDate('2023-01-01', '2023-12-31')
                 .map(function(image) {
                   var cloudMask = ee.Algorithms.Landsat.simpleCloudScore(image).select('cloud');
                   return image.updateMask(cloudMask.lt(20)).normalizedDifference(['B8', 'B4']).rename('NDVI');
                 });

Step 3: Data Fusion

• Combine SAR and MSI Features: Stack the preprocessed SAR (VV, VH) and MSI (NDVI, etc.) bands into a single multi-band image. This fusion enables the combination of temporal, spectral, and structural information, improving classification performance.

Step 4: Classification (Rice Mapping)

• Training Data: Use ground truth data or manually labeled regions to train a supervised classifier (e.g., Random Forest or SVM) on the fused dataset.
• Classification: Apply the classifier to identify rice fields based on the fused SAR-MSI dataset.

Step 5: Post-Processing and Validation

• Smoothing: Apply spatial filters to reduce noise in the classified output.
• Accuracy Assessment: Use independent validation data to assess the classification accuracy, typically using confusion matrices and overall accuracy.

4. Advantages of SAR-MSI Fusion for Rice Mapping:

1. Enhanced Accuracy: Combining SAR and MSI improves classification accuracy compared to using either data source alone.
2. All-Weather Monitoring: SAR data allows for rice mapping even under cloudy conditions, common during the monsoon season in rice-growing areas.
3. Monitoring Growth Stages: The fusion enables monitoring throughout the rice growth cycle, from flooding and transplanting to maturation and harvesting.

5. Applications in Rice Mapping:

1. Crop Monitoring: Track the health and growth stages of rice crops to optimize irrigation and other agricultural practices.
2. Drought & Flood Impact: Assess the impact of drought or excessive flooding on rice fields using moisture-sensitive SAR data.
3. Yield Prediction: Use temporal patterns in vegetation indices to predict rice yields.
4. Sustainable Agriculture: Contribute to precision agriculture by efficiently managing resources (e.g., water) and reducing environmental impacts.