Natural and anthropogenic disturbances act as important drivers of tree mortality, shaping the structure, composition and biomass distribution of forests. Disturbance regimes may emerge from different characteristics of disturbance events over time and space. We design a model- based experiment to investigate the links between disturbance regimes at the landscape scale and spatial features of biomass patterns. The effects on biomass of a wide range of disturbance regimes are simulated by varying three different parameters, i.e. μ (probability scale), α (clustering degree), and β (intensity slope) that shape the extent, frequency, and intensity of disturbance events, respectively. A simple dynamic carbon cycle model is used to simulate 200 years of plant biomass dynamics in response to circa +2000 different disturbance regimes, depending on the different combinations of μ, α, and β. Each parameter combination yields a spatially explicit estimate of plant biomass for which sixteen synthesis statistics are estimated on the spatial distributions of biomass, including information-based and texture features. Based on a multi-output regression approach we link these synthesis statistics with additional gross primary production (GPP) constraints to retrieve the three disturbance parameters. In doing so we evaluate the confidence in inferring disturbance regimes from spatial distributions of biomass. Our results show that all three parameters can be confidently retrieved. The Nash-Sutcliffe efficiency for the prediction of the μ, α, and β is 97.3%, 96.6%, and 97.9%, respectively. A feature importance analysis reveals that the distribution statistics dominate the prediction of μ and β, while features quantifying texture have a stronger connection with α. Overall, this study clarifies the association between biomass patterns emerging from different underlying disturbance regimes, while overcoming the previously found equifinality between mortality rates and total biomass. Given the links between decadal vegetation dynamics and the uncertainties in the role of terrestrial ecosystems in the global biogeochemical cycles, a better understanding and the quantification of disturbance regimes would improve our current understanding of controls and feedback at the biosphere-atmosphere interface in the current Earth system models.

In a model simulating dynamics of a system, parameters can represent system sensitivities and unresolved processes, therefore affecting model accuracy and uncertainty. Taking a light use efficiency (LUE) model as an example, which is a typical approach to estimate gross primary productivity (GPP), we propose a Simultaneous Parameter Inversion and Extrapolation approach (SPIE) to overcome issues stemming from plant-functional-type(PFT)-dependent parameterizations. SPIE refers to predicting model parameters using an artificial neural network based on collected variables, including PFT, climate types, bioclimatic variables, vegetation features, atmospheric nitrogen and phosphorus deposition and soil properties. The neural network was optimized to minimize GPP errors and constrain LUE model sensitivity functions. We compared SPIE with 11 typical parameter extrapolating methods, including PFT- and climate-specific parameterizations, global and PFT-based parameter optimization, site-similarity, and regression approaches. All methods were assessed using Nash-Sutcliffe model efficiency(NSE), determination coefficient and normalized root mean squared error, and contrasted with site-specific calibrations. Ten-fold cross-validated results showed that SPIE had the best performance across sites, various temporal scales and assessing metrics. None of the approaches performed similar to site-level calibrations(NSE=0.95), but SPIE was the only approach showing positive NSE(0.68). The Shapley value, layer-wise relevance and partial dependence showed that vegetation features, bioclimatic variables, soil properties and some PFTs are determining parameters. The proposed parameter extrapolation approach overcomes strong limitations observed in many standard parameterization methods. We argue that expanding SPIE to other models overcomes current limits and serves as an entry point to investigate the robustness and generalization of different models.