Trend analysis and performance of machine learning models for agroclimatology parameters in Bosso, Nigeria
Climate change necessitates regional transitions to mitigate its impact and enhance agricultural practices. This has led to increased efforts to adopt climate-smart agriculture in Nigeria, focusing on the concurrent analysis of various climatic parameters over the long term across different locatio...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/117235/ http://psasir.upm.edu.my/id/eprint/117235/1/117235.pdf |
| Summary: | Climate change necessitates regional transitions to mitigate its impact and enhance agricultural practices. This has led to
increased efforts to adopt climate-smart agriculture in Nigeria, focusing on the concurrent analysis of various climatic parameters over the long term across different locations. This study examines the adoption of climate-smart agriculture in Nigeria by analysing 15,949 daily observations for over 43 years that covers the period from 1st January 1981 through 31st August 2024 for three (3) agroclimatology parameters viz temperature, precipitation and relative humidity in Bosso, Niger State, a significant agricultural region affected by climate variability. The Mann–Kendall test was employed to assess linear trends, while machine learning (ML) models analysed non-linear characteristics using both custom and built-in functions in R-package. The findings from the trend analysis indicates a gradual drying trend, warming, and a slight decrease in precipitation, all of which significantly impact water availability and agricultural productivity. The Random Forest model performs best on the training data for temperature, relative humidity and precipitation with least errors and higher R2. For testing data, Support Vector Regression (SVR) excelled in predicting atmospheric conditions which are temperature (R2 = 0.8955,
RMSE = 0.7311, MSE = 0.5345, MAE = 0.5760) and relative humidity (R2 = 0.9214, RMSE = 4.5241, MSE = 20.4676,
MAE = 3.2613). However, SVR was outperformed by Gradient Boosting Model (GBM) in predicting precipitation, which
exhibited extreme values (R2=0.4983, RMSE=4.1192, MSE=16.9681, MAE=2.1010). The research recommends adopting heat and drought-resistant crops, enhancing irrigation systems, and improving soil health to cope with rising temperatures and reduced rainfall. In conclusion, stakeholders are encouraged to leverage the findings from the Mann-Kendal test and machine learning models for effective understanding of dynamics in the climate data, forecasting and timely dissemination of information, facilitating transition to climate-smart agriculture. The findings from this research will provide insights to guide
stakeholders in policymaking and promote awareness of climate-resilient agricultural practices among farmers and the public. |
|---|