I develop a discrete choice model of deforestation with spatially heterogeneous endogenous climate. Using pixel-level data for the Amazon Rainforest and the Brazilian state of Mato Grosso, one of the most important agricultural hubs in the world, I estimate the model and consider a counterfactual where farmers are allowed to deforest protected areas. Deforestation reduces rainfall and increases temperatures for downwind farmers, creating cascading effects on a network of externalities. Ignoring climate feedback loops significantly underestimates the costs of deforestation,  highlighting the importance of considering climate endogeneity beyond the production-emissions-temperature framework.