# Do pastoral and agro-pastoral perceptions align with observed climate extremes? Evidence from the Koh-e-Suleiman Range, Pakistan

## TL;DR

This study explores how well local perceptions of climate extremes in Pakistan's Koh-e-Suleiman Range match actual climate trends, finding alignment in some areas but not others.

## Contribution

The study introduces a novel integration of local climate perceptions with observed climate data using statistical and machine learning models in a data-scarce region.

## Key findings

- Perceptions of floods, rain intensity, temperature, and warm spells aligned closely with observed trends (≥80% accuracy).
- Perceptions of drought spells were predominantly inaccurate, with 75.3% of respondents overestimating their occurrence.
- Regression and machine learning analyses identified factors like education, age, and livestock ownership influencing perception accuracy.

## Abstract

This study examined the relationship between climate perceptions and observed trends among pastoralist and agro-pastoralist communities in the Koh-e-Suleiman Range, Pakistan. Household perception data were collected from 198 respondents and analyzed alongside climatic records across two time scales (1980–2022; 2013–2022). Data from the Pakistan Meteorological Department were used to compute 29 extreme climate indices, and trends were assessed using the Mann–Kendall and Sen’s slope tests. Perceptions of seven climate variables were compared with observed trends through accuracy tests, bias classification, regression, and machine-learning models. Perceptions aligned closely with observed trends for floods, rain intensity, temperature, and warm spells (\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\ge$$\end{document} 80% accuracy), and moderately for cold spells (71.7%) and rainfall (60.6%). Perceptions of drought spells were predominantly inaccurate, with 75.3% of respondents overestimating their occurrence. Regression analyses identified education, age, and livestock ownership as associated with perception accuracy. Classification and Regression Tree (CART) machine learning analysis, in contrast, revealed non-linear effects: income, age, and livestock herd size shaped drought spell perception, while livestock numbers and age influenced rainfall perception. These findings highlight the value of integrating observed climate extremes with local perceptions to better understand perception observation alignment and inform context- sensitive climate risk communication in data-scarce pastoral regions.