Nutrient composition varies greatly across landscapes, influencing the spatiotemporal variation and dynamics of populations, yet few studies have investigated this pattern across multiple scales. We tested how nutrient limitation affects herbivore populations, from individual behavior to landscape-level population dynamics, using the Australian plague locust, Chortoicetes terminifera (Walker), with field populations and long-term survey data from across their range. At the individual level, juvenile locusts selected for a carbohydrate-biased intake target of 1 protein (p) to 2 carbohydrate (c), and exhibited the highest growth and shortest development time when fed artificial diets with that same 1p:2c ratio during the final juvenile instar. At the field level, locusts kept in cages with protein-biased plants redressed their nutritional imbalance by selecting carbohydrate-biased diets (up to a 1p:20c ratio) for over a week after being removed from the protein-biased environment, returning to the 1p:2c intake target once the deficiency was met. At the landscape level, locust outbreaks were negatively correlated with soil nitrogen and showed a non-linear relationship with soil phosphorus, peaking at approximately 4% soil phosphorus. We disentangled the interaction between mean annual precipitation and soil nitrogen using comprehensive locust surveys and remotely sensed soil and weather data spanning decades. This study is the first to integrate lab, field, and remote sensing approaches, demonstrating the importance of nutrient balancing and acquisition across scales for herbivores. Specifically for locusts, we show that low-nitrogen environments promote outbreaks, likely by reducing plant protein-to-carbohydrate ratios. Incorporating soil quality data into locust plague forecasting models could help improve prediction accuracy.