Census parcels cropping system classification from multitemporal remote imagery: a proposed universal methodology.

Area of Science:

• Agricultural remote sensing
• Geospatial analysis
• Image classification

Background:

• Accurate crop assessment is vital for agricultural management and food security.
• Traditional methods of crop assessment are often labor-intensive and time-consuming.
• Remote sensing offers a scalable solution for monitoring agricultural land use.

Purpose of the Study:

• To develop and validate a semi-automated procedure (CROPCLASS) for census parcel crop assessment using multitemporal satellite imagery.
• To create a software tool (CROPCLASS-2.0) for economically feasible and efficient crop classification.
• To assess the accuracy and data requirements of the CROPCLASS methodology.

Main Methods:

• Developed CROPCLASS procedure involving parcel definition, spectral band (SB) and vegetation index (VI) extraction, data matrix (MD) creation, and decision tree (DT) classification.
• Utilized GeoEye-1 satellite images from April to October 2010 for the LaVentilla area, Spain.
• Incorporated ground-truth data for training DT models and validated classification accuracy.

Main Results:

• CROPCLASS-2.0 semi-automates crop assessment with high overall accuracy (OA) ranging from 80.7% for individual crops to 95%-100% within cropping systems.
• DT model accuracy remained consistent across different temporal resolutions of satellite imagery (e.g., monthly, seasonal, bi-monthly intervals).
• Classification of unidentified parcels achieved an OA of 79.5%.

Conclusions:

• The CROPCLASS procedure and software provide an effective and economically feasible solution for semi-automated crop assessment using multitemporal remote sensing data.
• The methodology demonstrates robustness across various temporal data sampling strategies.
• Accurate crop identification and land use classification are achievable with this approach, supporting agricultural monitoring efforts.