Forecasting Temperature And Precipitation Using Machine Learning

Abstract

Precise forecasting of climate patterns ips essential
for efficient environmental management and
strategic planning, especially with rising climate
unpredictability and catastrophic weather
occurrences. This work aims to improve the
prediction accuracy of global climate models by
using a random forest algorithm with seasonal bias
correction, targeting the forecasting of alterations
in precipitation and temperature across diverse
climatic scenarios. Our methodology significantly
enhances the accuracy of climate predictions by
using an extensive dataset that encompasses
historical climate data and future estimates based
on Representative Concentration Pathways.
Following the use of seasonal bias correction, the
Correlation Coefficient for precipitation enhanced
from 0.826 to 0.860, while the Nash–Sutcliffe
Efficiency rose from 0.633 to 0.735. Furthermore,
the Mean Absolute Error and Root Mean Square
Error decreased, consequently augmenting the
model's trustworthiness in forecasting severe
precipitation occurrences.
The Random Forest model enhanced temperature
projections, with a Correlation Coefficient of up to
0.987, signifying robust predictive efficacy. These
enhancements underscore the capability of machine
learning techniques in optimizing climate models,
therefore offering more precise instruments for
policymakers and planners to address climate
concerns. The effective use of these sophisticated
statistical methods highlights the need of ongoing
innovation in climate research, ensuring that
climate forecasts are relevant and dependable for
guiding global climate resilience and adaptation
policies.