Evaluation of accuracy of neural network method in late spring frost estimating in pistachio growing areas of Kerman

Document Type : Research Article

Authors

1 Department of Water Engineering Faculty of Agriculture Shahid Bahonar University of Kerman. Kerman. Iran

2 Department of Water Engineering, Faculty of Agriculture, Shahid Bahonar University of Kerman, Kerman, Iran

Abstract

Introduction

Spring frost is considered an important threat to agricultural products in high and middle latitudes. The damage caused by the LSFs (Late Spring Frosts) significantly affects vulnerable plant organs markedly affects. This event has caused more economic losses to agriculture than any other climatic hazard in Asia, North America, and Europe. Also, these phenomena have contributed to low crop yields in Iran. The latest statistics released by the Food and Agriculture Organization of the United Nations (FAO) show that Iran is one of the largest producers of agricultural products and the world’s second-biggest producer of pistachios. Kerman province is one of the significant areas of pistachio production. This province has a large share of the pistachio word area plantation. Pistachio spring frost damage resulted in low yields crop yields in the last few years. An important principle in the study of frost is the estimation of this phenomenon. In this study, artificial neural network method methods have been used to estimate late spring frost in the pistachio crop of Kerman city.

Materials and Methods

In this study, the efficiency of this method was investigated in the estimation of Minimum temperature. For this purpose, the Daily data of the synoptic station of Kerman City were obtained from Iran Meteorological Organization from 2000 to 2020. Meteorological data including mean, maximum, and minimum temperatures, relative humidity, wind speed, saturated vapor pressure, and sunshine hours were used. Five different combinations of these variables was considered as input variables in artificial neural network method for minimum temperatures modeling. After entering data into network and modeling with each combination, RMSE and R2 values were calculated. Finally, the combination of 8 variables including average and maximum temperature, the minimum temperature the previous day and two days prior, relative humidity, wind speed, saturated vapor pressure, and sunny hours were selected as the most suitable combination of variables. Then the simulation of the minimum temperature values with %10 of the data was done. The performance of the methods was evaluated using statistical indices of coefficient of determination (R2), mean square of error (RMSE), Mean Bias Error (MBE), and Coefficient of Nash–Sutcliffe (NSE).

Results and Discussion

The accuracy of an analytical method is the degree of agreement of test results generated by the method to the true value. By examining the models, the M1 model was found to be the best model due to the lowest RMSE and higher R2. ANN model results were evaluated using various performance measure indicators. The simulated outcome of the model indicated a strong association with actual data where the correlation coefficient above 0.95 and the MBE index is zero was observed. Also, the RMSE value is positive and close to zero, and the NSE value is above 0.75. Therefore artificial neural network method has high accuracy. In this study, mean annual minimum temperature was estimated using artificial neural network models (from March 10 to May 20 comparison between the observed and calculated data showed that these data are in good agreement. Also, the results showed that temperature fluctuations were high between March 10 and March 31. From 2011 to 2017, an almost uniform temperature trend has been observed between March 10 and March 31. But the years 2000, 2006, and 2020 showed a noticeable decrease in temperature. From 2018 to 2020, this trend of temperature reduction has continued. In April, the temperature values were between 7 and 10 degrees Celsius. The years 2001, 2005, 2006, 2009, 2016, and 2019 have had a noticeable decrease in temperature. In May, the mean minimum temperature was between 10 and 14 degrees Celsius. Therefore, the probability of frost occurrence in early-flowering cultivars was higher in late March than in April and May. The years 2000, 2004, 2005, 2012, 2015, 2019 and 2020 had the highest number of frost days in the last two decades.

Conclusions

The results showed that the artificial neural network method had a high performance in estimating the minimum temperature. The values of the statistical indicators were R2=0.963, RMSE=0.027oC, MBE= 0 and NSE=0.966 respectively. In addition, the ANN method performed good in estimating the number of critical days and frost in pistachio crop. The results showed, although reducing the number of input data in models will result in the reduction of their output precision but Data-driven methods can be used as a useful tool for Minimum temperature estimation.

Keywords

Main Subjects


CAPTCHA Image

Articles in Press, Accepted Manuscript
Available Online from 08 May 2024
  • Receive Date: 27 February 2024
  • Revise Date: 24 April 2024
  • Accept Date: 08 May 2024
  • First Publish Date: 08 May 2024