|
Calibration of ERA5 daily precipitation using MLP, D-Tree, and KNN algorithms in Razavi Khorasan province
|
Majid Rajabi Jaghargh    , Seyed Mohammad Mousavi Baygi * , Seyed Alireza Araghi , Hadi Jabari Noghabi |
| Professor, Department of Water Science and Engineering, Faculty of Agriculture, Ferdowsi University of Mashhad, Mashhad, Iran |
|
|
Abstract: (786 Views) |
| Satellite products are the only available data source with adequate spatial coverage, however, their data do not match the observed values and have biases, although this discrepancy cannot be fixed precisely, however, a solution to reduce the bias is data recalibration. Currently, machine learning techniques are used to improve the accuracy of forecasting various types of weather phenomena, so regression solving such problems through methods based on machine learning and deep learning is very efficient. The daily precipitation of 19 rain gauge stations of the Ministry of Energy between 2010 and 2021 was extracted and compared to the average values of their corresponding daily precipitation pixels in the ERA5 database. To measure the data, three algorithms D-Tree, KNN, and MLP were used. The range of changes of correlation coefficient in MLP, D-Tree, and KNN is equal to [0.87, 0.98], [0.75, 0.97], and [0.4, 0.87], respectively. In addition, the range of changes for RMSE in MLP varies from 0.7 to 2.4 mm per day, and these changes for D-Tree and KNN are calculated between 0.8 to 2.2 and 1.2 to 2.5, respectively. In 75% of stations, RMSE in MLP, D-Tree, and KNN algorithms is less than 1.5, 1.9, and 2.2 mm per day, respectively. The range of bias changes in MLP is [0.18, -0.6 mm per day] and this range of changes for D-Tree and KNN is respectively [0.16, 0.5 mm per day] and [0.6, -0.8 mm per day] have been calculated. The bias of corrected data and observed values in MLP, D-Tree, and KNN algorithms for the middle of the stations is -0.09, -0.11, and -0.16 mm per day, respectively. The evaluation of the performance of three machine learning algorithms (MLP, D-Tree, and KNN) in correcting the daily precipitation of the ERA5 database and the comparison of CC, RMSE, and bias statistical indices for the reproduced data compared to ground values showed that in all three statistical indices, the MLP algorithm works better than the others and has good accuracy for correcting the daily precipitation. |
Article number: 8 |
| Keywords: Calibration, database, statistical indicators, machine learning |
|
|
Full-Text [PDF 1757 kb]
(132 Downloads)
|
|
Type of Study: Research |
Subject:
Special
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
|
Rajabi Jaghargh M, Mousavi Baygi S M, Araghi S A, Jabari Noghabi H. Calibration of ERA5 daily precipitation using MLP, D-Tree, and KNN algorithms in Razavi Khorasan province. Journal of Rainwater Catchment Systems 2024; 12 (1) : 8 URL: http://jircsa.ir/article-1-530-en.html
|
