Review article: Towards improved drought prediction in the Mediterranean region – modeling approaches and future directions

Abstract

Abstract. There is a scientific consensus that the Mediterranean region (MedR) is warming and as the temperature continues to rise, droughts and heat waves are becoming more frequent, severe, and widespread. Given the detrimental effects of droughts, it is crucial to accelerate the development of forecasting and early warning systems to minimize their negative impact. This paper reviews the current state of drought modeling and prediction applied in the MedR, including statistical, dynamical, and hybrid statistical–dynamical models. By considering the multifaceted nature of droughts, the study encompasses meteorological, agricultural, and hydrological drought forms and spans a variety of forecast scales, from weekly to annual timelines. Our objective is to pinpoint the knowledge gaps in literature and to propose potential research trajectories to improve the prediction of droughts in this region. The review finds that while each method has its unique strengths and limitations, hybrid statistical–dynamical models appear to hold the most promising potential for skillful prediction with seasonal to annual lead times. However, the application of these methods is still challenging due to the lack of high-quality observational data and the limited computational resources. Finally, the paper concludes by discussing the importance of using a combination of sophisticated methods such as data assimilation techniques, machine learning models, and copula models and of integrating data from different sources (e.g., remote sensing data, in situ measurements, and reanalysis) to improve the accuracy and efficiency of drought forecasting.

Using Calibrated Rainfall Forecasts and Observed Rainfall to Produce Probabilistic Meteorological Drought Forecasts

“…The product produced in this study was probabilistic, allowing verification to be completed objectively across both non-drought and drought conditions. By being probabilistic, its utility for making risk management decisions is increased, compared to a deterministic forecast [15]. The data can also be visualized as probabilistic risk maps, allowing drought risk to be more easily communicated [15].…”

“…By being probabilistic, its utility for making risk management decisions is increased, compared to a deterministic forecast [15]. The data can also be visualized as probabilistic risk maps, allowing drought risk to be more easily communicated [15].…”

“…As noted in earlier literature [15], it is likely that this methodology, being based on dynamical modeling, would be limited in its ability to capture drought state changes caused primarily by stochastic and short-term weather variability (e.g., the absence or presence of extreme rainfall events). This is particularly relevant for Australia, where drought development and recovery are strongly related to heavy rainfall days associated with synoptic weather systems [31].…”

“…In a review of drought prediction systems over the Mediterranean, [15] noted that the majority were based on statistical models, which had a tendency to be limited by assumptions of stationarity, as well as overfitting, especially if the input variables were not independent of each other. The two methods that used dynamical model forecasts as inputs were based on forecasting an index only using forecast information [16] or on using the forecasts in a statistical model [17].…”