Technique for improving seasonal climate forecasts

• Grant number: 21/09647-3
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Duration: January 01, 2022 - November 30, 2022
• Field of knowledge: Physical Sciences and Mathematics - Geosciences - Meteorology
• Principal Investigator: Camila Cossetin Ferreira
• Grantee: Camila Cossetin Ferreira
• Host Company: Wetterlab Ltda
• CNAE: Pesquisa e desenvolvimento experimental em ciências físicas e naturais
• City: São Paulo
• Pesquisadores principais: Rodrigo Yamamoto
• Associated grant(s): 23/04207-0 - Technique for improving seasonal climate forecasts, AP.PIPE
• Associated scholarship(s): 22/00728-3 - Statistical correction methods, BP.TT; 21/15088-7 - Technique for improving seasonal climate forecasts, BP.PIPE

Abstract

The climate changes we have experienced in recent decades and which are present in the majority of future climate projections have highlighted the need for social planning in different spheres: from more resilient cities to companies that respond to climate fluctuations. Climate variability in seasonal scale, in particular, has a major impact on agriculture, power generation, water resources management, among others. Despite this importance, seasonal climate forecasts are still not widespread and this is due to the mismatch between user needs and available information. The main factors responsible for the difficulty in disseminating long-term forecasts are: unreliable forecasts, complexity in obtaining and interpreting data, and the absence of products that meet the particularities of each sector. In this project, we intend to explore these issues through innovative methodologies and thus offer a suitable product for market demands. Specifically, we propose the development of a correction method for seasonal climate forecasts from a dynamic model and a distinct form of data communication and availability.Generally, the correction techniques adopted for weather and climate forecasts are based on comparing the data to be corrected with reference data. There are numerous ways to adjust two time series, such as linear regression, comparison between the probability distributions of the two sets, and artificial intelligence algorithms. Most of these techniques have a common point, they relate the series as a whole, generalizing the model error and suggesting a universal correction. The difference of the methodology we propose is to fragment the time series to be corrected into small parts and look for similar patterns in the past, to then select the reference data corresponding to these specific periods and apply a statistical correction technique. In this way, each type of phenomenon has a correction adapted according to the particularity of its forecast error. For example, the correction for precipitation due to a cold front during winter would be different from the correction for precipitation triggered by a convective system in summer. To optimize the search process for similar patterns in the historical period, the recognition will be done in the frequency domain using the Fast Fourier Transform. The input data for the pattern recognition algorithm will come from a lagged ensemble, defined through a previous evaluation. The statistical methods for correction will also be tested and according to their performance the most appropriate will be selected.It is worth noting the user's confidence depends not only on the accuracy of forecasts, but also the way of communicating an error. Showing where and when the model fails is crucial to the proper use of long-term forecasts. Thus, as important as developing a method to improve the quality of forecasts is to develop a validation system and thereafter transmit this information clearly. To meet this need and present a new type of user interaction, we will develop a web platform for requesting and delivering data, reducing bureaucracy and automating these processes.In summary, as a result of this project, we intend to offer products that are easy to access and interpret, which simplify and encourage the use of long-term forecasts in the management of strategic areas, filling a gap identified in the market and presenting new paradigms in terms of climate forecasts. (AU)