the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
The Application of Machine Learning Algorithms to the Global Forecast of Temperature-Humidity Index with High Temporal Resolution
Abstract. Climate change poses a significant threat to agriculture, with potential impacts on food security, economic stability, and human livelihoods. Dairy cattle, a crucial component of the livestock sector, are particularly vulnerable to heat stress, which can adversely affect milk production, immune function, feed intake, and in extreme cases, lead to mortality. The Temperature Humidity Index (THI) is a widely used metric to quantify the combined effects of temperature and humidity on cattle. However, most studies estimate THI using daily-level data, which fails to capture the full extent of daily thermal load and cumulative heat stress, especially during nights when cooling is inadequate. To address this limitation, we developed a machine learning approach to temporally downscale daily climate data to hourly THI values. Utilizing historical ERA5 reanalysis data, we trained an XGBoost model and generated hourly THI datasets for 12 NEX-GDDP-CMIP6 climate models under two emission scenarios (SSP2-4.5 and SSP5-8.5) extending to the end of the century. This high-resolution THI data provides a more accurate assessment of heat stress in dairy cattle, enabling better predictions and management strategies to mitigate the impacts of climate change on this vital agricultural sector.
- Preprint
(14721 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2024-344', Anonymous Referee #1, 12 Sep 2024
This article employs the XGBoost model to temporally downscale daily climate data, generating THI data that quantifies the impact of temperature and humidity on cattle. The model training requires 24-48 days, but generating a year's data only takes 3-8 hours. Overall, the topic is interesting. However, I have some concerns as follows:
- " We opted for the XGBoost model for its computational efficiency compared to Random Forest and other analogous algorithms, specifically for our use case." The authors claim that XGBoost offers higher computational efficiency than Random Forest and similar algorithms in their specific case. Please provide specific comparative values.
- The spatial resolution of the THI is 0.25°, which seems to be a very coarse pixel size. How can you ensure that the values accurately represent such a large spatial area? Additionally, the input data for the method includes ERA5 and NEX-GDDP-CMIP6 with a spatial resolution of 0.25° as well. Would this method still be applicable if high-resolution data were available?
- The ERA5 reanalysis dataset provides historical hourly data. Knowing its quality would be helpful. Can you utilize ground-truth data to validate the used variables in the ERA5 dataset? Without assurance of the input data quality, the quality of the trained model cannot be guaranteed.
- More details on the structure of the XGBoost regressor model are needed.
- The formula for THI indicates that its value depends on temperature and relative humidity, with the latter being calculable from ambient and dew point temperatures using equations 1-3. This suggests a strong dependency of THI on these two temperature data. An error analysis will be helpful. The author could do an experimental study to explore how errors in these temperatures affect THI accuracy.
- "The performance of the trained models was assessed using ground truth data derived from the ERA5 dataset." Please provide more details about the ground truth data, including its spatial representativeness and temporal resolution.
Citation: https://doi.org/10.5194/essd-2024-344-RC1 -
RC2: 'Comment on essd-2024-344', Anonymous Referee #2, 04 Oct 2024
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2024-344/essd-2024-344-RC2-supplement.pdf
- AC1: 'Comment on essd-2024-344', Pantelis Georgiades, 14 Nov 2024
Data sets
Temperature Humidity Index GDDP-NEX-CMIP6 ML projections Pantelis Georgiades https://doi.org/10.26050/WDCC/THI
Model code and software
The Application of Machine Learning Algorithms to the Global Forecast of Temperature-Humidity Index with High Temporal Resolution Pantelis Georgiades https://github.com/pantelisgeor/Temperature-Humidity-Index-ML
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
370 | 54 | 57 | 481 | 15 | 9 |
- HTML: 370
- PDF: 54
- XML: 57
- Total: 481
- BibTeX: 15
- EndNote: 9
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1