Comparison of CART Algorithm and Cropping Calendar in Estimating Paddy Growth Stage in Karawang Regency, West Java

Classification And Regression Trees (CART) is one of the classic and simple algorithm in predictive modeling machine learning. This study aims to compare the result of paddy growth stage estimates based on CART model of Sentinel-1A Synthetic Aperture Radar (SAR) data and Cropping Calendar (KATAM). The construction of the CART model utilises real data field from Area Frame Sampling (Kerangka Sampling Area or KSA) in Karawang Regency observed on 2020. The CART algorithm makes predictions using a tree structure or hierarchical structure. The CART algorithm focuses on finding a decision tree model that has a Gini impurities value = 0. The rules for classifying class based on the physical polarization spectrum which is represented by pixel digital number from Vertical-Vertical (VV), Vertical-Horizontal (VH), and VV/VH of SAR image properties. This study found that the initial planting time is different. The CART model estimates the initial planting time is on September, while the KATAM estimates on November-December.

I. Introduction

Indonesia is an agricultural country with one of the main agricultural commodities is rice. Data from Statistic Indonesia (BPS) stated that rice production in 2020 reached 54.65 million tons of dry milled grain (in Indonesian called Gabah Kering Giling or GKG). Since rice production is strategic activities, therefore some methods have developed to monitor rice and paddy field including paddy growth stage. There are two methods that have been developed for paddy growth stage observation which are satellite-based remote sensing and statistic-based terrestrial observation. Satellite-based remote sensing method has the advantage on area coverage and repeat observations and well known as temporal resolution, whereas statistic-based terrestrial observation method has the advantage on informing actual paddy condition directly from the field [1].