The coupling between ecosystems and the climate implies that quantifying their dynamics is crucial for our attempts to prevent undesired changes in our environment. The main goal of this project is to use non-linear, stochastic modeling in the context of climate and ecosystems dynamics. This proposal combines methods from statistical physics, non-linear dynamics, game theory and random dynamical systems to study the various effects of noise on climate and ecosystems dynamics. The specific goals of this project are:(i) Using the record of past climate observations to build dynamically-weighted forecasting ensembles adjusted to specific climate variables. These ensembles will enable inter-comparison of the models. Using the sequential compound decision method will reduce the uncertainties of climate predictions. (ii) Investigating the dynamics of non-linear vegetation models, in the homoclinic snaking regime, in the presence of global and local noise. Early-warning signals and the possibility of desertification not through critical transition but through a series of transitions between localized states will be studied. (iii) Analyzing data describing the process of vegetation mortality under controlled and natural drought conditions. Spatial and temporal patterns in the mortality process will be studied and related to quantitative models. Theoretical models will be developed and tested against the data. The expected results of this research will not only improve our understanding of climate and ecosystems dynamics but will also advance fundamental physics research such as the interplay between complex non-linear dynamics and stochastic effects.