Past severe droughts over North America have led to massive water shortages and increases in wildfire frequency. Triggering sources for multi-year droughts in this region include randomly occurring atmospheric blocking patterns, ocean impacts on atmospheric circulation, and climate's response to anthropogenic radiative forcings. A combination of these sources translates into a difficulty to predict the onset and length of such droughts on multi-year timescales. Here we present results from a new multi-year dynamical prediction system that exhibits a high degree of skill in forecasting wildfire probabilities and drought for 10-23 and 10-45 months lead time, which extends far beyond the current seasonal prediction activities for southwestern North America. Using a state-of-the-art earth system model along with 3-dimensional ocean data assimilation and by prescribing the external radiative forcings, this system simulates the observed low-frequency variability of precipitation, soil water, and wildfire probabilities in close agreement with observational records and reanalysis data. The underlying source of multi-year predictability can be traced back to variations of the Atlantic/Pacific sea surface temperature gradient, external radiative forcings, and the low-pass filtering characteristics of soils. Over the past six years, several regions in North America have experienced severe drought conditions. Their impacts cover a wide range of sectors such as agriculture, energy, food security, forestry, drinking water, and tourism 1, 2. Unusually dry and hot conditions were reported for Texas and Mexico in 2010-2011 3 , for the Great Plains in 2012 4 , and for California in 2011-2014 5, 6. The economic damage associated just with the recent California drought has been estimated at ~2.2 billion United States (US) Dollars and a loss of ~17,000 jobs 7. Successful water resource management that relies on knowledge of the present and future hydroclimatic conditions is crucial for mitigating the climate-driven drought risks. Even though the atmosphere has a short dynamical memory of less than several weeks, its evolution is partly affected by slowly varying sea surface temperature (SST) conditions. In the low-frequency range, atmospheric variability is modulated by more predictable climate phenomena such as the El Niño-Southern Oscillation (ENSO) 8 , the Pacific Decadal Oscillation (PDO) 9 , the Atlantic/Pacific SST contrast 10, 11 , and the Atlantic Multidecadal Oscillation 12, 13. The ratio of internally generated atmospheric variability and SST-forced variability thus limits the potential prediction horizon of monthly to seasonally averaged rainfall changes to less than 1 year 14-16. However, there are many land systems (e.g., soils, water reservoirs, vegetation, and perennial snowpack) that effectively filter out the high-frequency rainfall variability and therefore exhibit longer persistence as a result of natural time integration of atmospheric signals 17, 18. Consequently, this low-pass filtering effect enhances the cont...