Beyond Accuracy: Evaluating the Data-Driven, Process-Informed DHR-SDP Model for Rainfall Prediction and Climate Signal Representation
Year: 2025
Type: preprint
Abstract: Rainfall prediction is essential for understanding and mitigating the impacts of global environmental change. Dynamic and statistical models are used for this purpose: dynamic models simulate underlying physical processes but are computationally intensive, while statistical models are usually more cost-efficient. Stochastic climate models that are both data-driven and process-informed are interesting, as they often show enhanced predictive capabilities over conventional statistical models, partly due to their ability to capture the inherent climate dynamics. Ensuring the reliability of both predictions and their conceptual interpretations (e.g., descriptions of climate patterns) requires a robust evaluation that extends beyond assessing predictive accuracy. Here, we rigorously assess a data-driven, process-informed rainfall model—DHR-SDP (Dynamic Harmonic Regression with State-Dependent Parameters)—against mean-monthly rainfall estimates, alternative data-driven models with exogenous variables—ARMAX (Autoregressive Moving Average), ANNX (Autoregressive Neural Network), and ARX-SDP (Autoregressive with State-Dependent Parameters)—and nineteen GCM monthly hindcasts at 1° resolution with bias correction. We consider a process-oriented evaluation testing the model’s ability to represent climate patterns (seasonality and trends). This evaluation is conducted in a complex orographic region of the Southern Andes characterized by spatial rainfall variability. Results show that DHR-SDP reproduces seasonal rainfall dynamics, trends, and monthly predictions reliably, due to its capacity to replicate intra- and inter-annual processes. These findings underscore the potential of data-driven, process-informed approaches to predict and conceptualize local climate behavior, even in complex terrains. Further work should examine the model’s broader applicability to climate-change modeling and projections and clarify relationships among global and local climate variables and phenomena.
Source: Research Square
Institutions: University of Cuenca, Universidad del Azuay
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W4415740992