Spatial portability of random forest models to estimate site-specific air temperature for prediction of emergence dates of the Asian corn borer in North Korea

Authors: YOO, BYUNG HYUN - Seoul National University; KIM, KWANG SOO - Seoul National University; PARK, JIN YU - Seoul National University; MOON, KYUNG HWAN - National Institute Of Horticultural & Herbal Science (NIHHS); AHN, JUNG JUN - National Institute Of Horticultural & Herbal Science (NIHHS); Fleisher, David

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/9/2021
Publication Date: 6/19/2022
Citation: Yoo, B., Kim, K., Park, J., Moon, K., Ahn, J., Fleisher, D.H. 2022. Spatial portability of random forest models to estimate site-specific air temperature for prediction of emergence dates of the Asian corn borer in North Korea. Computers and Electronics in Agriculture. 199(e107113). https://doi.org/10.1016/j.compag.2022.107113.
DOI: https://doi.org/10.1016/j.compag.2022.107113

Interpretive Summary: The outbreak of insect pests, such as Asian corn borer, can seriously damage crop yield and quality and impact food security. Mathematical models that can anticipate the appearance of these insects can be used by farmers to prepare effective management strategies. However, these tools need accurate daily temperature information at the farm level which is often difficult to obtain in more food-insecure countries. Satellite data can potentially be used to fill this gap. We evaluated the accuracy of developing air temperature estimates using some of this satellite information by linking the data with a statistical model. This model was then used to estimate the emergence date of the insect pest at thousands of locations in the Korean peninsula. The results showed high agreement with measured data at multiple test sites, which indicates that satellite-based temperature estimates were accurate across a broad spatial area. In this application, we showed how the new technology can help growers prepare ahead of time for the appearance of destructive pests and save their crops. These methods of estimating air temperatures can also be used for other decision support tools in food production areas for which such data is not available.

Technical Abstract: Air temperature estimates obtained from satellite data products can facilitate site-specific prediction of pest outbreaks including Asian Corn Borer (Ostrinia furnacalis), a significant crop pest. We demonstrate an approach to estimate daily air temperature with spatial portability at 1-km spatial resolution using two products, land surface temperature (LST) and atmospheric profile (AP) products based on MODIS satellites. Quantitative and qualitative variables were used as inputs to assess and improve spatial portability of random forest (RF) models that are used to convert this data into ground-based estimates. These variables were grouped into sets for the purpose of training the RF model and included LST data products (LT), AP data products (AP), geographical properties (GE), data quality and cloud conditions (QC), land cover type (LC), and auxiliary properties (AX). Different set combinations were evaluated for use in building the RF model and cross-validated using satellite data and observed daily weather data from the Korean peninsula from 2013 to 2019 at 1-km resolution. The RF model that used AT, LT and QC variable sets as inputs had the highest values for the concordance correlation coefficient (CCC) and spatial portability index (SPI) metrics across test sites. Emergence date of the corn borer pest was predicted with high values of CCC (0.94-0.97) using the RF model derived temperature data. These results suggest that both LST and AP satellite products would contribute to higher spatial portability of daily air temperature estimation. Using this technology for temperature estimation purpose coupled with crop pest phenology prediction methods, as demonstrated, will facilitate grower management strategies for coping with biotic pressures on crop production and is of particular use in countries that are food insecure with sparce on-ground weather station networks.