Early-season crop mapping using improved artificial immune network (IAIN) and Sentinel data

Area of Science:

• Agricultural Remote Sensing
• Machine Learning for Crop Classification
• Time Series Analysis in Agriculture

Background:

• Early-season crop identification is challenging, especially in diverse agricultural regions with complex cropping patterns.
• Existing classifiers struggle with irregular time series data common in high-resolution satellite imagery.
• Accurate crop mapping is crucial for effective agricultural management and food security.

Purpose of the Study:

• To propose and evaluate an improved artificial immune network (IAIN) for early-season crop type classification.
• To assess the effectiveness of IAIN in handling irregular time series data from Sentinel-1 and Sentinel-2.
• To identify major crops in Hengshui, China, using short image time series.

Main Methods:

• Generated 15-day composite time series images from Sentinel-1 and Sentinel-2 data (10 m resolution).
• Selected Near-Infrared (NIR) band and Normalized Difference Vegetation Index (NDVI) as optimal features using Jeffries-Matusita distances and random forest Gini importance.
• Employed the proposed IAIN classifier to analyze the time series data for crop identification.

Main Results:

• IAIN achieved high overall accuracy: 99% for winter wheat and 98.55% for summer crops.
• Winter wheat was identified with >95% producer's accuracy (PA) and user's accuracy (UA) 20 days before harvest (April 1-May 15 series).
• Cotton and spring maize were mapped with >95% PA and UA using time series longer than April 1-August 15, enabling mapping 4-6 weeks pre-harvest.

Conclusions:

• The IAIN classifier demonstrates significant potential for early-season crop type mapping, even with irregular time series.
• Utilizing Sentinel-1 and Sentinel-2 data with IAIN offers a robust approach for timely crop identification.
• This methodology provides valuable insights for proactive crop management and agricultural planning.