Refining Agricultural Official Statistics through Mixed Methods: The Integration of Area Sampling Frame and Remote Sensing
Conference
10th International Conference on Agricultural Statistics
Format: CPS Paper - ICAS 2026
Keywords: asf, machine learning, remote sensing
Abstract
As the National Statistics Office (NSO) of Indonesia, Statistics Indonesia (BPS) has taken a strategic step toward integrating big data into the production of agricultural official statistics. While big data has traditionally been used as a complementary analytical resource within national statistical systems, this study advances its role toward becoming an operationally integrated component of official statistical production. The research develops and evaluates a Mixed Method framework that integrates Area Sampling Frame (ASF) survey data, Sentinel-1 Synthetic Aperture Radar (SAR) imagery, and machine learning techniques to improve the timeliness, spatial granularity, and reliability of rice phenology statistics. Within this framework, ASF survey data function as ground-truth references, while satellite-derived backscatter variables extend spatial coverage beyond sampled locations. The modelling stage applies the XGBoost algorithm, followed by a Bayesian post-processing procedure that incorporates temporal transition probabilities to enhance logical consistency in growth-stage predictions. Under controlled modelling conditions, the approach achieved high classification accuracy (91-97%), with Bayesian correction significantly improving temporal stability. When implemented nationwide across all KSA sample points without filtering, performance decreased due to increased heterogeneity but remained substantially improved after post-processing (accuracy increasing from 0.595 to 0.752), demonstrating robustness under real-world conditions. Spatial evaluation revealed regional variability, reflecting differences in agricultural complexity and cropping systems. Importantly, the framework is designed in alignment with the Fundamental Principles of Official Statistics, ensuring transparency, methodological rigor, and institutional accountability. The results demonstrate that integrating survey data, remote sensing, and probabilistic machine learning provides a feasible and governance-consistent pathway for modernizing agricultural statistics, strengthening evidence-based policymaking, enhancing food security monitoring, and improving statistical coverage across Indonesia’s geographically diverse archipelago.