Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China
Study region: China. Study focus: This research aims to address the limitations of existing VPD reanalysis data, which are characterized by coarse spatial resolution and insufficient accuracy validation. Using ERA5-Land, TerraClimate, GLDAS time series data, and additional predictors such as vegetat...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-02-01
|
| Series: | Journal of Hydrology: Regional Studies |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214581825000163 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832591788778455040 |
|---|---|
| author | Mi Wang Zhuowei Hu Xiangping Liu Wenxing Hou |
| author_facet | Mi Wang Zhuowei Hu Xiangping Liu Wenxing Hou |
| author_sort | Mi Wang |
| collection | DOAJ |
| description | Study region: China. Study focus: This research aims to address the limitations of existing VPD reanalysis data, which are characterized by coarse spatial resolution and insufficient accuracy validation. Using ERA5-Land, TerraClimate, GLDAS time series data, and additional predictors such as vegetation indices, land cover types, terrain, and surface temperature, the study developed a random forest (RF) model to estimate VPD (named RF-VPD). Two independent experiments, station-to-station and year-to-year, were conducted to evaluate the models accuracy against in-situ data. New hydrological insights for the region: Machine learning-based downscaling methods offer a potential solution to enhance the accuracy of VPD spatiotemporal distribution. In this study, we utilized ERA5-Land, TerraClimate, and GLDAS time series, vegetation indices, land cover types, terrain, and surface temperature data as predictors to estimate VPD using three machine learning models, named RF-VPD. VPD station data were computed based on meteorological data from 2481 meteorological stations in China, and two independent experiments (station-to-station and year-to-year) were employed to evaluate the estimation accuracy of RF-VPD using this data as a reference. Additionally, the station-to-station statistics of the VPD reanalysis data and the consistency between the RF-VPD and in-situ VPD data were examined. Overall, the random forest model performed the best in predicting VPD (R²=0.98, NSE=0.98, MAE=0.04, RMSE=0.06, RSR=0.13, and KGE=0.98). The high accuracy demonstrated in year-to-year and station-to-station experiments, coupled with the strong consistency between RF-VPD data and in-situ observations, underscores the robustness and applicability of RF-VPD models in capturing spatiotemporal patterns. Furthermore, our study demonstrates the dynamic impact of hydrothermal transfer processes and meteorological factors on RF-VPD products. The high-resolution RF-VPD product derived from this method has significant potential for practical applications, including drought monitoring, agricultural productivity assessments, fire risk evaluations, and water resource management. Furthermore, the developed RF-VPD model can be a valuable complement to existing VPD datasets, offering higher accuracy and resolution. By improving the prediction of VPD dynamics, this research provides important tools for climate change adaptation, ecosystem monitoring, and sustainable resource management. |
| format | Article |
| id | doaj-art-b034d24887124206a778638294847040 |
| institution | Kabale University |
| issn | 2214-5818 |
| language | English |
| publishDate | 2025-02-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Journal of Hydrology: Regional Studies |
| spelling | doaj-art-b034d24887124206a7786382948470402025-01-22T05:42:23ZengElsevierJournal of Hydrology: Regional Studies2214-58182025-02-0157102192Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in ChinaMi Wang0Zhuowei Hu1Xiangping Liu2Wenxing Hou3College of Resources Environment and Tourism, Capital Normal University, Beijing 100048, ChinaCorresponding author.; College of Resources Environment and Tourism, Capital Normal University, Beijing 100048, ChinaCollege of Resources Environment and Tourism, Capital Normal University, Beijing 100048, ChinaCollege of Resources Environment and Tourism, Capital Normal University, Beijing 100048, ChinaStudy region: China. Study focus: This research aims to address the limitations of existing VPD reanalysis data, which are characterized by coarse spatial resolution and insufficient accuracy validation. Using ERA5-Land, TerraClimate, GLDAS time series data, and additional predictors such as vegetation indices, land cover types, terrain, and surface temperature, the study developed a random forest (RF) model to estimate VPD (named RF-VPD). Two independent experiments, station-to-station and year-to-year, were conducted to evaluate the models accuracy against in-situ data. New hydrological insights for the region: Machine learning-based downscaling methods offer a potential solution to enhance the accuracy of VPD spatiotemporal distribution. In this study, we utilized ERA5-Land, TerraClimate, and GLDAS time series, vegetation indices, land cover types, terrain, and surface temperature data as predictors to estimate VPD using three machine learning models, named RF-VPD. VPD station data were computed based on meteorological data from 2481 meteorological stations in China, and two independent experiments (station-to-station and year-to-year) were employed to evaluate the estimation accuracy of RF-VPD using this data as a reference. Additionally, the station-to-station statistics of the VPD reanalysis data and the consistency between the RF-VPD and in-situ VPD data were examined. Overall, the random forest model performed the best in predicting VPD (R²=0.98, NSE=0.98, MAE=0.04, RMSE=0.06, RSR=0.13, and KGE=0.98). The high accuracy demonstrated in year-to-year and station-to-station experiments, coupled with the strong consistency between RF-VPD data and in-situ observations, underscores the robustness and applicability of RF-VPD models in capturing spatiotemporal patterns. Furthermore, our study demonstrates the dynamic impact of hydrothermal transfer processes and meteorological factors on RF-VPD products. The high-resolution RF-VPD product derived from this method has significant potential for practical applications, including drought monitoring, agricultural productivity assessments, fire risk evaluations, and water resource management. Furthermore, the developed RF-VPD model can be a valuable complement to existing VPD datasets, offering higher accuracy and resolution. By improving the prediction of VPD dynamics, this research provides important tools for climate change adaptation, ecosystem monitoring, and sustainable resource management.http://www.sciencedirect.com/science/article/pii/S2214581825000163Random forest modelDownscalingMulti-source remote sensing dataVPD |
| spellingShingle | Mi Wang Zhuowei Hu Xiangping Liu Wenxing Hou Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China Journal of Hydrology: Regional Studies Random forest model Downscaling Multi-source remote sensing data VPD |
| title | Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China |
| title_full | Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China |
| title_fullStr | Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China |
| title_full_unstemmed | Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China |
| title_short | Vapor pressure deficit (VPD) downscaling based on multi-source remote sensing, in-situ observation, and machine learning in China |
| title_sort | vapor pressure deficit vpd downscaling based on multi source remote sensing in situ observation and machine learning in china |
| topic | Random forest model Downscaling Multi-source remote sensing data VPD |
| url | http://www.sciencedirect.com/science/article/pii/S2214581825000163 |
| work_keys_str_mv | AT miwang vaporpressuredeficitvpddownscalingbasedonmultisourceremotesensinginsituobservationandmachinelearninginchina AT zhuoweihu vaporpressuredeficitvpddownscalingbasedonmultisourceremotesensinginsituobservationandmachinelearninginchina AT xiangpingliu vaporpressuredeficitvpddownscalingbasedonmultisourceremotesensinginsituobservationandmachinelearninginchina AT wenxinghou vaporpressuredeficitvpddownscalingbasedonmultisourceremotesensinginsituobservationandmachinelearninginchina |