A spatially based quantile regression forest model for mapping rural land values.

Area of Science:

• Agricultural Economics
• Geospatial Analysis
• Machine Learning

Background:

• Rural land valuation is crucial for agricultural land use policy.
• Machine learning enhances mass appraisal, but traditional models often ignore spatial autocorrelation.
• Existing methods like linear regression struggle with non-linear, non-normal spatial data.

Purpose of the Study:

• To develop and evaluate a spatial version of Quantile Regression Forest (sQRF) for rural land valuation.
• To account for spatial autocorrelation in land values using neighboring site information.
• To improve computer-assisted mass appraisal of rural land values.

Main Methods:

• Implemented a spatial Quantile Regression Forest (sQRF) algorithm.
• Compared sQRF predictions against spatial random forest, kriging regression, and linear regression.
• Utilized ancillary environmental variables for prediction.

Main Results:

• sQRF demonstrated strong performance in predicting rural land values.
• sQRF outperformed multiple linear regression in predictive accuracy.
• sQRF provides a direct measure of prediction uncertainty.

Conclusions:

• The developed sQRF model effectively addresses spatial autocorrelation in rural land valuation.
• sQRF offers a robust alternative for computer-assisted mass appraisal, improving accuracy and uncertainty assessment.
• Accurate land valuation models are vital for taxation and land use policy development.