Federico Bert’s research while affiliated with Council for Agricultural Research and Agricultural Economy Analysis and other places

Each year, farmers must decide crops and their agronomical management for the plots of their farms. This decision—subject to climatic interaction and crop prices context—will determine farm earnings. We introduce a framework, called MORDMAgro, based on Many Objective Robust Decision‐Making methodology to support farmers' decisions. Through the use of a scenario approach, the framework aims to assist in situations where there is no agreement on how to represent the uncertain critical parameters that affect the outcome, that is, crop prices or weather conditions. It considers seven decision objectives that focus on costs, margins, utilities, returns, losses, gains, and regrets to integrate a comprehensive range of farmers’ goals. The framework outputs robust strategies to farmers, that is, land allocation to crops that return acceptable outcomes for as many scenarios as possible, rather than finding an “optimal” strategy that optimizes one or several objectives. It also identifies critical scenario factors that decrease a decision's payback by a classification tree algorithm. We applied the framework to a case study of a farm in the Argentine Pampas to identify robust strategies from typical cropping alternatives based on wheat, maize, and soybean. We share all scripting and data to ensure reproducibility and foster the framework's usage.

Over the last decades, the rapid replacement of native forests by crops and pastures in the Argentinean semiarid Chaco plains has triggered unprecedented groundwater level raises resulting from deep drainage increases, leading to the first massive waterlogging event on records (~25.000 Ha flooded in 2015 near Bandera, one of the most cultivated clusters of the Chaco). In this paper, we link this episode to the ongoing deforestation and cropping scheme shifts through the combined analysis of remote sensing data, agricultural surveys, local farmer information and hydrologic modeling. From 2000 to 2015, the agricultural area of Bandera increased from 21% to 50%, mostly at the expense of dry forests. In this period, agriculture migrated from more intensive (i.e., double‐cropping) to more water‐conservative (i.e., late‐summer single crops) schemes, as a general strategy to reduce drought risks. These changes reduced regional evapotranspiration and increased the intensity of deep drainage in wet years. Contrasting cropping schemes displayed significant evapotranspiration differences, but all of them experienced substantial drainage losses (~100‐200 mm) during the wettest year (2014/15), suggesting that cropping adjustments have a limited capacity to halt the generation of water excesses. Nearly 50% of the cropped area in Bandera could not be sown or harvested following the groundwater recharge event of 2014/2015. In the ongoing context of shallow and rising water tables, the introduction of novel cropping schemes that include deep‐rooted perennials, to promote transpirative groundwater discharge, seems crucial to avoid the recurrence of water excesses and their associated dryland salinity risk in the region.

We present Hydroman, a flexible spatially explicit model coupling human and hydrological processes to explore shallow water tables and land cover interactions in flat agricultural landscapes, modeled after the Argentine Pampas. With fewer parameters, Hydroman aligned well with established hydrological models, and was validated with observed water table patterns and crop yield data. Simulations with different climate, phreatic and land cover conditions confirmed that climate remains the main driver, but crops also influence water levels and yields, depending on the growing cycle. We also examined the impacts of two alternative sowing strategies. Risk aversion proves robust in minimizing crop losses, but often results in less sowing, exacerbating flooding. Strict rotators risk more, but help stabilize the groundwater levels. Reintroducing pasture further stabilizes the system. Future work will engage farmers to derive and assess land cover strategies that maximize yield and minimize losses, and transfer our modeling approach to other applications.

We present a Bayesian hierarchical space-time stochastic weather generator (BayGEN) to generate daily precipitation and minimum and maximum temperatures. BayGEN employs a hierarchical framework with data, process, and parameter layers. In the data layer, precipitation occurrence at each site is modeled using probit regression using a spatially distributed latent Gaussian process; precipitation amounts are modeled as gamma random variables; and minimum and maximum temperatures are modeled as realizations from Gaussian processes. The latent Gaussian process that drives the precipitation occurrence process is modeled in the process layer. In the parameter layer, the model parameters of the data and process layers are modeled as spatially distributed Gaussian processes, consequently enabling the simulation of daily weather at arbitrary (unobserved) locations or on a regular grid. All model parameters are endowed with weakly informative prior distributions. The No-U Turn sampler, an adaptive form of Hamiltonian Monte Carlo, is used to maximize the model likelihood function and obtain posterior samples of each parameter. Posterior samples of the model parameters propagate uncertainty to the weather simulations, an important feature that makes BayGEN unique compared to traditional weather generators. We demonstrate the utility of BayGEN with application to daily weather generation in a basin of the Argentine Pampas. Furthermore, we evaluate the implications of crop yield by driving a crop simulation model with weather simulations from BayGEN and an equivalent non-Bayesian weather generator.

RESUMEN La región pampeana de la República Argentina, una de las mayores llanuras del mundo, ha registrado en los últimos 50 años un fuerte ascenso en los niveles freáticos, con el consecuente aumento en la frecuencia de inundaciones. Esta dinámica tiene origen en dos procesos que se desarrollaron en ese período. En primer lugar, la zona presentó una tendencia hacia del aumento en las precipitaciones anuales. En segundo lugar se produjo un fuerte aumento del área dedicada a la agricultura, desplazando zonas con pasturas y pastizales, es decir, hubo un cambio en el uso del suelo. A través de ensayos numéricos con un modelo hidrológico (distribuido en el espacio y continuo en el tiempo, debidamente calibrado y verificado), se muestra en este trabajo que el aumento de las precipitaciones es el fenómeno que explica en mayor medida el incremento observado en los niveles freáticos, pero que la vegetación también juega un rol altamente significativo. Más aún, se pone de manifiesto la no linealidad de la respuesta del sistema hidrológico a los cambios en la precipitación y el uso del suelo, ya que la combinación de ambos efectos produce un resultado bastante inferior a la suma de cada uno de los efectos por separado. Adicionalmente, el modelo indica que existe una relación exponencial entre la profundidad de la napa y las áreas inundadas, estableciéndose una profundidad freática de 2 metros como el valor umbral a partir de la cual las áreas inundadas crecen significativamente.

In flat environments, groundwater is relatively shallow, tightly associated with surface water and climate, and can have either positive and negative impacts on natural and human systems depending on its depth. A linked modeling and analysis framework that seeks to capture linkages across multiple scales at the climate/water/crop nexus in the Argentine Pampas is presented. This region shows a strong coupling between climate, soil water, and land use due to its extremely flat topography and poorly developed drainage networks. The work describes the components of the framework and, subsequently, presents results from simulations performed with the twin goals of (i) validating the framework as a whole and (ii) demonstrating its usefulness to explore interesting contexts such as unexperienced climate scenarios (wet/dry periods), hypothetical policies (e.g., differential grains export taxes), and adoption of non-structural technologies (e.g., cover crops) to manage water table depth. Keywords: Natural-human systems, MIKE SHE, agent-based model, agriculture, water table, risk management

Ecosystem services (ES) have become a key concept in the assessment of natural resources, as a way to connect human well-being and ecosystems degradation. However, ES quantification is considered a basic problem because provision varies considerably as a result of land use change and site-specific characteristics (i.e. climate, soil, topography, and time). Thus, more detailed studies are needed to assess whether these changes affect ecological variables. We explored the use of environmental and crop management variables in predicting the provision of four ES (soil C balance, soil N balance, N2O emission control and groundwater contamination control) in three agroecosystems located in the Pampa region (Argentina). Data-mining, represented by k-means cluster and classification trees, was used to identify the dependence of ES provision on the variation of both environmental and crop management factors. We used plot level crop management and environmental field information stored in a large database during a 10-year period. The k-means method selected five different clusters. The final configuration showed two contrasting clusters: one with the lowest ES provision, and another one with the highest ES provision. The five clusters were represented in the terminal nodes of the final classification tree. Regarding the predictive power of the variables, crop and year were the most important predictors. Then, differences observed in ES provision resulted from changes in land use (variable “crop”) and crop season (variable “year”). These results are meant to enlighten stakeholders in terms of how to manage Pampean agroecosystems in order to positively influence ES provision.

We present the application of a parametric stochastic weather generator within a nonstationary context, enabling simulations of weather sequences conditioned on interannual and multi-decadal trends. The generalized linear model framework of the weather generator allows any number of covariates to be included, such as large-scale climate indices, local climate information, seasonal precipitation and temperature, among others. Here we focus on the Salado A basin of the Argentine Pampas as a case study, but the methodology is portable to any region. We include domain-averaged (e.g., areal) seasonal total precipitation and mean maximum and minimum temperatures as covariates for conditional simulation. Areal covariates are motivated by a principal component analysis that indicates the seasonal spatial average is the dominant mode of variability across the domain. We find this modification to be effective in capturing the nonstationarity prevalent in interseasonal precipitation and temperature data. We further illustrate the ability of this weather generator to act as a spatiotemporal downscaler of seasonal forecasts and multidecadal projections, both of which are generally of coarse resolution.

In flat plains groundwater affects agricultural production outcomes and risks. Agricultural land use decisions however, may strongly impact groundwater levels available for production. This paper explores the scope for managing groundwater levels through land use decisions in a sub-basin of the Salado River in the Argentine Pampas, a very flat area that plays a key role in world agricultural production. A spatially distributed hydrological model implemented with MIKE SHE software was used to establish the impacts of different land uses on groundwater dynamics, and to assess the interdependencies among spatially-close decision-makers sharing a water table. Additionally, groundwater level changes in response to climate variability were quantified. We found land use has strong effects on water table levels both for oneself (e.g., pastures can lead to significant decreases (up to 4.5 m) in water table levels) and others, in the form of strong interdependencies that exist between farmers sharing a water table where land use decisions of one farmer effect groundwater level of neighboring farms and vice-versa. However, the effectiveness to control groundwater levels through land use decisions are subject to the rather unpredictable effects of rainfall variability. The results presented in this paper provide key insights in relation to physical and social aspects that should be considered for managing groundwater levels through land use decisions, in order to avoid negative and/or maximize positive effects on agricultural production.

Sensitivity analyses (SAs) identify how an output variable of a model is modified by changes in the input variables. These analyses are a good way for assessing the performance of probabilistic models, like Bayesian Networks (BN). However, there are several commonly used SAs in BN literature, and formal comparisons about their outcomes are scarce. We used four previously developed BNs which represent ecosystem services provision in Pampean agroecosystems (Argentina) in order to test two local sensitivity approaches widely used. These SAs were: 1) One-at-a-time, used in BNs but more commonly in linear modelling; and 2) Sensitivity to findings, specific to BN modelling. Results showed that both analyses provided an adequate overview of BN behaviour. Furthermore, analyses produced a similar influence ranking of input variables over each output variable. Even though their interchangeably application could be an alternative in our Bayesian models, we believe that OAT is the suitable one to implement here because of its capacity to demonstrate the relation (positive or negative) between input and output variables. In summary, we provided insights about two sensitivity techniques in BNs based on a case study which may be useful for ecological modellers.