Cloud-based crop type mapping for Nandi County, Kenya

Crop type mapping is essential to agriculture applications, including yield estimates, crop planting acreage statistics, agricultural market predictions, and land use change analysis that support relevant decision-making. Since 2008, the U.S. Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) has produced the annual Cropland Data Layer (CDL) with 108 crop types and 26 other land-use types for the CONUS Climate Console at 30 meters spatial resolution. According to Boryan et al., 2011, the See5 decision tree classifier model is trained for each Landsat scene with ground truthing data reported by farmers on what is grown at USDA’s Farm Service Agency Common Land Units (CLU). The overall classification accuracy for main crops (e.g., corn and soybean) ranged from 5% to 95% in 2009. Annual CDL product is available freely at the CropScape web service system. It has been widely used in various agricultural applications and decision-making, such as flood monitoring and crop loss estimation, crop mask extraction for early-season winter wheat identification (Boryan et al., 2019), and land cover change estimation. Some previous researchers studied remote sensing in-season crop mapping using machine learning models. The pixel-level machine learning classification algorithm requires high-quality crop-type ground truth data to integrate corresponding satellite images as training samples to train the machine learning classifier. The quality of these crop labels significantly impacts classification performance because the classifier learns how to distinguish crop types from the labels. This research focuses on producing crop-type maps for Nandi county, Kenya. We surveyed crop type’s ground truth data in Nandi County that integrates time series Sentinel-2 imagery to construct the training sample dataset. The supervised machine learning model - Random Forest was trained by training samples and employed in crop-type classification for Nandi County. The assessment result confirmed that this workflow performs well for crop type classification with reasonable accuracy. The remaining sections of the report are structured as follows: Section 2 - Study Area and Method, Section 3- Sequential Experiments and Results and Section 4- Conclusion. Section 2 provides a comprehensive overview of the study area and outlines the methodology employed in the research. It lays the foundation for understanding the geographical context and the procedures followed in the study. Section 3 delves into a detailed account of sequential experiments conducted and their corresponding results. Section 4 summarizes the results of this report.