Downscaling livestock census data using multivariate predictive models: Sensitivity to modifiable areal unit problem.

Area of Science:

• Spatial statistics
• Geographic information science
• Agricultural economics

Background:

• The modifiable areal unit problem (MAUP) introduces statistical bias in aggregated spatial data.
• Prior research primarily focused on MAUP's impact on upscaling models.
• Understanding MAUP's influence on downscaling is crucial for accurate spatial analysis.

Purpose of the Study:

• To investigate the effects of MAUP on spatial data downscaling methodologies.
• To assess how scale, shape, and sampling methods influence downscaling results.
• To provide insights for improving spatial data analysis and product dissemination.

Main Methods:

• Aggregated fine-resolution chicken and duck census data from Thailand across three administrative levels.
• Disaggregated data using the Gridded Livestock of the World framework with two predictor sampling methods.
• Conducted sensitivity analysis on Pearson's r and RMSE to evaluate downscaling performance.

Main Results:

• Data scale, not administrative unit shape or sampling method, significantly affected downscaling precision.
• Training models with the finest available administrative level yielded better results.
• Spatially clustered datasets were less affected by MAUP than homogenous datasets, showing improved accuracy metrics.

Conclusions:

• Prioritizing the finest administrative level for model training is recommended to mitigate MAUP.
• Aggregation sensitivity analysis is vital for interpreting spatial study results and ensuring robust data products.
• MAUP's impact on downscaling necessitates careful consideration of data aggregation strategies.