<![CDATA[Background: Rainfed paddy fields are the most climate-vulnerable segment of food-crop land, yet they are the least adequately represented in spatial databases. On the east coast of Aceh, Indonesia, North Aceh Regency holds 8,356 ha of rainfed paddy distributed across 27 sub-districts, equivalent to 21.02% of its 39,762 ha of cultivated baseline paddy land, while Bireuen Regency holds 2,784.91 ha across 13 sub-districts, equivalent to 18.64% of its 14,944 ha. Taken together, 11,140.91 ha, or 20.37% of the paddy land in both regencies, depends entirely on rainfall. Between 2021 and 2023 the harvested rice area contracted by 29.07% in North Aceh and by 10.14% in Bireuen, while yields in North Aceh declined from 5.77 to 5.38 t ha⁻¹. Objective: Optical monitoring in this region is constrained by cloud cover that peaks in October and November, precisely the land-preparation and transplanting window of rainfed fields. Methods: This study adopts a systematic evidence-synthesis design covering 13 published land cover classification and rice mapping studies, combined with an analysis of official secondary statistics, in order to quantify the scale of the rainfed paddy problem on the east coast of Aceh, to benchmark the performance of the Random Forest algorithm on the Google Earth Engine platform using published empirical evidence, and to formulate a cloud-resilient operational monitoring protocol. Results: The synthesis shows that optical-based Random Forest achieves a mean overall accuracy of 94.34% (range 89.00–98.81%; standard deviation 3.52) with a mean Kappa coefficient of 0.8926, and consistently outperforms CART by 4.87 percentage points of overall accuracy and by 0.107 Kappa points. Optical–radar fusion schemes attain comparable accuracy (mean 94.65%) without depending on the availability of cloud-free imagery, and the evidence indicates that preserving the temporal dimension of the image time series raises accuracy by up to 14.7% for Sentinel-1 data. Conclusion: On this basis, a ten-class monitoring protocol comprising 104 predictor variables, a four-layer cloud-handling chain, and explicit rules for separating rainfed from irrigated paddy is proposed, accompanied by a ready-to-run Google Earth Engine script provided as an appendix.]]>
