<![CDATA[Abstract. The national rice supply chain is vulnerable to information asymmetry and the bullwhip effect, aggravated by conventional production forecasting that ignores spatial heterogeneity and relies on lagging data. This study aims to develop and validate an Early Warning System prototype for rice production forecasting that integrates multi-source satellite data, spatial heterogeneity, and model interpretability. Following a Research and Development approach as a proof-of-concept, the study employs a multi-output Random Forest across six rice-producing regencies in East Java during 2019–2024, with ablation validation, SHAP interpretation, and spatiotemporal robustness testing (LORO-CV and LOYO-CV). The results show that integrating spatial context substantially improves productivity accuracy (R² from 0.3419 to 0.8703). SHAP analysis reveals a biophysical dichotomy: quantity components are governed by the temporal dimension, whereas quality is governed by the spatial dimension. The model is temporally robust (coefficient of variation below 15%) with a transparently mapped spatial generalization boundary. Its outputs are transformed into an illustrative economic value estimate and a package of decision maps. Abstrak. Manajemen rantai pasok beras nasional rentan terhadap asimetri informasi dan bullwhip effect, yang diperburuk oleh peramalan produksi konvensional yang mengabaikan heterogenitas spasial dan bertumpu pada data yang bersifat lagging. Penelitian ini bertujuan membangun dan memvalidasi prototipe Sistem Peringatan Dini peramalan produksi padi yang mengintegrasikan data satelit multi-sumber, heterogenitas spasial, dan interpretabilitas model. Mengikuti pendekatan Research and Development sebagai uji konsep (proof-of-concept), penelitian menggunakan Random Forest multi-keluaran pada enam kabupaten sentra padi di Jawa Timur sepanjang 2019–2024, dengan validasi ablation, interpretasi SHAP, serta uji ketangguhan spatiotemporal (LORO-CV dan LOYO-CV). Hasil menunjukkan integrasi konteks spasial meningkatkan akurasi Produktivitas secara substansial (R² 0,3419 menjadi 0,8703). Analisis SHAP mengungkap dikotomi biofisik: komponen kuantitas dikendalikan dimensi temporal, sedangkan kualitas dikendalikan dimensi spasial. Model tangguh secara temporal (Koefisien Variasi di bawah 15%) dengan batas generalisasi spasial yang terpetakan transparan. Keluaran ditransformasikan menjadi estimasi nilai ekonomi ilustratif dan paket peta keputusan.]]>
Copyrights © 2026
