Drought is one of the most devastating natural disasters in the world, affecting millions of people in various ways. Effective monitoring and assessment are therefore essential. This study aimed to develop an Earth observation-derived combined agricultural drought index (CADI) for agricultural drought monitoring in the Amhara Region. The development of CADI for the entire study period, that is, from 2000 to 2021, was constructed using a substantial machine learning (ML) algorithm, i.e., Random Forest (RF), by using SPI data calculated from selected fifty-one stations of the study area for each summer month. SPI values of each month were used to train the RF model and test the model; thus, 70% of the value of SPI was utilized for training the RF algorithms, and 30% of the value of SPI was utilized for testing. For the entire study period (2000 to 2021), remote sensing-based input parameters, which are Land Surface Temperature (LST), Temperature Condition Index (TCI), Evapotranspiration (ET), Normalized Difference Vegetation Index (NDVI), Vegetation Condition Index (VCI), Enhanced Vegetation Index (EVI), Precipitation Condition Index (PCI), and Soil Moisture Condition Index (SMCI), in addition to these, three topographic variables (Elevation, Slope, and Aspect) were utilized to develop CADI on a monthly scale for the entire study period (2000 to 2021). Crop yield and the Emergency Events Database (EM-DAT) drought periods were utilized to assess the effectiveness of the Combined Agricultural Drought Index. The result of CADI indicated that temporal and spatial variations of agricultural drought records have been observed in the study period with different severity levels. Hence, during the 22 years, 14, 16, 12, & 13 in June, July, August, and September were affected by droughts of different severity levels, respectively. Particularly, the CADI-derived maps depict that 2002, 2009, and 2015 were the years most affected by drought in terms of area coverage and severity levels (severe-to-extreme drought cases) in the Amhara region, Ethiopia. During these years, drought has affected about 77.14%, 83.04%, and 86.93% of the study area in 2002, 2009, and 2015, respectively, and 2007 was the wettest year during the entire study period as well as seasonally. The correlation coefficient results between the combined agricultural drought index and crop yield were very good, ranging from r = 0.51 to 0.95, indicating the index's utility for agricultural drought monitoring. From these, the highest correlation was observed in the East Gojam Zone and the lowest in the Wag Himra Zone. Overall, the combined agricultural drought index performed well in capturing the spatiotemporal patterns of the historic agricultural drought occurrences. CADI also performed well in distinguishing between drought and non-drought years, with higher sensitivity and spatial resolution than traditional drought indices. Thus, to mitigate the negative effects of drought in the region, the model can be used to develop an early warning system and agricultural drought monitoring. The CADI-based spatial analysis confirms its utility as a reliable tool for detecting and managing agricultural drought, aiding policymakers and stakeholders in developing region-specific drought preparedness and mitigation strategies. As a conclusion, this study confirms that CADI is a robust and adaptable index that enhances early warning capabilities and supports informed decision-making for agricultural drought risk management.