Predicción estacional de sequía meteorológica y anomalía de la actividad de incendios

Resumo

Droughts generate very significant impacts both on society and the environment. Understanding the factors that explain their variability and, therefore, allow for their prediction, has been and continues to be a scientific challenge. A reliable prediction of drought will enable an advance estimation of its consequences and, consequently, be a fundamental tool in decision-making. The main objective of this work is based on the adjustment, evaluation, and operational implementation of simple empirical seasonal prediction models for droughts and wildfires. The first part of the work presents an operational prototype with high spatial resolution to predict seasonal meteorological drought in Spain (4SPAIN). The prediction system is based on an empirical method known as Ensemble Streamflow Prediction (ESP). The model predicts drought using the Standardized Precipitation Index (SPI) from monthly precipitation data. Model validation demonstrates its predictive capability, providing probabilistic seasonal predictions with a four-month lead time. However, results should be interpreted with caution after the dry season. Secondly, a similar model is proposed for the entire planet using ESP. The role of incorporating uncertainty associated with the use of different precipitation databases included in DROught Probabilistic (DROP) is analyzed. Thus, ESP is applied to each of the DROP databases, resulting in SPI (ESP). The ensemble of models is referred to as 4DROP. Additionally, the added value of including dynamic seasonal precipitation predictions is examined, specifically the predictions from the ECMWF's Seasonal Forecasting System 5 (SEAS5). SPI predictions will be obtained by applying the 4DROP methodology, using observations as initial conditions and SEAS5 predictions. The properties of the new ensemble are referred to as S5. While both systems exhibit similar skill, the S5 model demonstrates better performance in longer prediction lead times, especially in tropical regions. Furthermore, it is important to highlight that prediction skill increases significantly when considering the complete set of eleven predictions, as opposed to predictions based on individual database sources. Finally, a simple empirical model based on logistic regression is proposed to predict the anomaly of burned area by fires as a function of climatic conditions (droughts) both preceding and concurrent to the fire season (CLIBA). The model is replicated in the eleven members of DROP, resulting in eleven probabilistic predictions of the burned area anomaly. The ensemble of all these predictions is identified as CLIBA-DROP. Similar to 4DROP, the inclusion of SEAS5 (S5) in the system as concurrent conditions is tested, while maintaining observations for preceding conditions. This process is optimized for seasonal predictions with a 4-month lead time. The ensemble of CLIBA(S5) applied to DROP databases is referred to as CLIBA-4FIRE. The analysis of results using the probabilistic ROC area metric shows a statistically significant relationship in approximately 68% of the total area with available burned area data (in at least 10 out of 20 years) when considering the CLIBA-DROP model, i.e., with observed conditions. CLIBA-4FIRE, which includes concurrent conditions from S5, exhibits predictive capability in around 60% of the area. In other words, the performance of CLIBA-4FIRE shows very promising results. This approach underscores the importance of taking into account the uncertainty of observations in seasonal prediction and enables the development of operational products for issuing early warnings (https://matv.shinyapps.io/app_APPS/). The codes and data are made available at (https://github.com/MTAV26/thesis/).