Resumen de Improvement of seasonal forecasting techniques applied to water resources and forest fires
Raül Marcos Matamoros
This thesis studies the benefits of different calibration approaches on seasonal forecasting and the improvement of seasonal prognosis of water resources and forest fires. Droughts and wildfires are an inherent problem to the Mediterranean and are likely to worsen if climate change continues. Both hazards are a source of important economic costs, environmental damage and, in the case of wildfires, even life losses. These impacts have encouraged policy- and decision-makers to reduce vulnerability by placing great efforts in the development of mitigation and adaptation strategies. Seasonal forecasting could help with this task by foretelling the behaviour of water resources and wildfire with months in advance. Furthermore, it has the capacity to provide operational frameworks that can work both in present and future climate conditions. However, seasonal predictability in extra-tropical latitudes is usually considered rather limited and, consequently, seasonal forecasts are seldom used in decision-making. There are studies, though, suggesting that calibration methods could help improving current model’s output. Thus, the existing gap between end-user goals and theoretical research needs more work to demonstrate the utility of seasonal forecasts. To achieve this objective this study has been divided in three sub-objectives: skill assessment, seasonal forecast of water resources and seasonal forecast of forest fires. The skill assessment comprises an evaluation of the skill of the raw ECMWF System-4 output in Europe, Spain, Catalonia and the Muga river basin; and the study of the impact on the ECMWF System- 4 performance of the MOS-analog and linear regression calibrations in comparison to mean bias correction. As for the seasonal forecast of water resources the application began with the modelling of the Boadella reservoir in-flow, out-flow and volume anomalies through a Multiple Linear Regression (MLR) procedure. Afterwards, the seasonal predictability of the Boadella predictands has been evaluated through several seasonal forecast approaches. Finally, regarding the seasonal forecast of wildfires the first step has been to model summer (JJAS) burned area in Catalonia through a MLR with antecedent and current year drought conditions. Subsequently, the MLR performance has been tested under different seasonal forecast configurations. Among the results obtained it has been found that most of the skill of the ECMWF System-4 is focused in the first lead. At this horizon deterministic forecasts improve climatology and persistence in the majority of months and for all the variables considered. The probabilistic assessment, on the other hand, showed this skill was mainly centred in the winter months. Also, the added value of calibration post-processing techniques has been checked. These techniques always ameliorate the skill of the original model output by correcting first order biases. Nevertheless, the MOS-analog outcome has also hinted the possibility to go beyond these results if the analog pool was sufficiently increased. In reference to the reservoir’s applications the perfect prognosis approach revealed that in-flow, volume and out-flow anomalies can be modelled through multiple linear regression (MLR). In all three cases summer months showed enhanced predictability way beyond the first lead, a significant result for water managers. Moreover, the results proved that volume anomaly seasonal forecasts could begin the operational switch from customary climatology to another forecast strategy based on MLR models. This is also true for some months in the outflow’s modelling. For the in-flow case, though, there is still further research needed before reaching that sate. Finally, regarding forest fires, exploiting the relationship between summer burned area and preceding drought conditions can lead to MLR models that provide a seasonal estimate of the expected above/below-normal summer fire burned area in Mediterranean-type regions.
