بررسی کمبود فشار بخار بر عملکرد و پیش بینی محصول زعفران با استفاده از الگوریتم‌های هوش مصنوعی

بخش کشاورزی یکی از اولین بخش‌هایی است که تحت تاثیر تغییرات اقلیم قرار می‌گیرد. خشکسالی و تغییرات وسیع در وضعیت آب و هوا منجر به کاهش محصولات کشاورزی و در نهایت کاهش بهره‌وری اقتصادی می‌شود. یکی از مهمترین پارامترهای مدیریت دقیق آب کشاورزی، کمبود فشار بخار نام دارد. در این تحقیق ابتدا به برآورد کمبود فشار بخار از داده‌های باز تحلیل بانک اطلاعاتی jra-55 بین سال‌های 1958 تا 2023 پرداخته شد. در گام بعدی بررسی دراز مدت تغییرات کمبود فشار بخار و تاثیر آن بر عملکرد زعفران در چهار محدوده استان خراسان جنوبی شامل بیرجند، قاین، سرایان و طبس انجام شد. نتایج در مناطق مذکور طی 60 سال گذشته افزایش تدریجی میانگین سالانه کمبود فشار بخار با نرخ متوسط هر دهه حدود 60 پاسکال و کاهش سالانه عملکرد زعفران در حدود 0.16 کیلوگرم در هکتار نشان داد و همچنین با افزایش کمبود فشار بخار کاهش عملکرد محصول مشاهده شد. در نهایت به پیش‌بینی عملکرد زعفران برپایه کمبود فشار بخار با استفاده از الگوریتم‌های هوش مصنوعی پرداخته شد. جهت ارزیابی عملکرد مدل‌ها در هر دو مرحله آموزش و تست از چهار شاخص آماری cc ، mae ، rmse و rrmse  استفاده شد. عملکرد مدل جنگل تصادفی در مرحله آموزش و تست بهتر از سایر مدل‌ها بود [0.99=cc؛ 0.06=mae؛ 0.09=rmseو 0.01 =rrmse]، [0.94=cc؛ 0.23=mae؛ 0.28= rmseو 0.01=rrmse]، این تحقیق استفاده از مدل جنگل تصادفی را برای مطالعات آتی پیش‌بینی عملکرد محصول زعفران برپایه کمبود فشار بخار در مدیریت نیاز آبی توصیه می‌کند.

impact of vapor pressure deficit on saffron yield and its prediction using artificial intelligence algorithms

introduction: climate change is one of the most significant environmental challenges of the 21st century, with profound impacts on agriculture, especially in arid and semi-arid regions such as south khorasan province, iran. saffron is a strategic and valuable product, with 90% of the world’s cultivation area and 93.7% of global production of this product belonging to iran. saffron is a low-maintenance plant, well-suited for cultivation in the arid and semi-arid regions of the country, and it also offers high economic returns. considering the specific climatic conditions of iran, where water is one of the limiting factors for agricultural development, saffron is an ideal agricultural product. one of the parameters influencing agricultural crop performance is the vapor pressure deficit (vpd). vpd is not only a critical factor affecting plant physiology but also has a significant impact on the water requirements of plants however, the increasing vapor pressure deficit (vpd) caused by climate change has directly affected its yield. vapor pressure deficit (vpd) is an index representing the difference between the actual moisture content of the air and the maximum moisture the air can hold. this index is influenced by temperature and relative humidity and can be calculated using meteorological data. this study aims to examine the long-term impacts of vpd on saffron yield and predict its performance using artificial intelligence (ai) models. materials and methods: in this study, climatic data including monthly data of temperature, humidity, precipitation, and vapor pressure deficit (vpd) were obtained from the jra-55 database for the years 1958 to 2023 for four regions of south khorasan province (birjand, tabas, esfeden, and sarayan). these data were processed using tools in the arcgis environment. saffron yield data for the years 2005 to 2023 were collected from the south khorasan agricultural organization and vapor pressure deficit was calculated using existing equations. to predict the long-term saffron yield, four artificial intelligence algorithms, including random forest (rf), generalized additive model (gam), random subspace (rss), and m5p, were used. the models were evaluated using the cross-validation technique to avoid overfitting or underfitting. to assess the performance of the models, statistical indices such as correlation coefficient (cc), mean absolute error (mae), root mean square error (rmse), and relative root mean square error (rrmse) were used. results and discussion: the long-term analysis of vapor pressure deficit (vpd) changes and its impact on saffron yield in the studied regions revealed a gradual increase in the annual average vpd over the past 60 years. the decadal increase in average vpd for birjand was 60 pa, sarayan 100 pa, qaen 40 pa, and tabas 50 pa. the reduction in saffron yield averaged annually as 0.16 kg/ha for birjand, 0.2 kg/ha for sarayan, 0.13 kg/ha for qaen, and 0.14 kg/ha for tabas. statistical indices were used to evaluate the performance of the models in both the training and testing stages. the performance of the random forest model was superior to the other models in both the training and testing phases [rrmse= 0.01, rmse= 0.09, mae= 0.06, cc= 0.99] ,[rrmse= 0.01, rmse= 0.28, mae= 0.23,cc= 0.94], the generalized additive model (gam) performed similarly to the random forest model during the training phase and ranked slightly lower in performance during the testing phase. the random subspace (rss) model showed moderate performance in both the training and testing phases, with better performance during training compared to testing. the m5p model demonstrated poorer performance compared to the other models. these findings highlight the significant impact of increasing vpd on saffron yield due to climate change. the rf model, owing to its high accuracy and ability to handle complex relationships among variables, proved to be the best model for saffron yield prediction. conclusion: this study demonstrated that vpd, as a climate-sensitive parameter, plays a critical role in reducing saffron yield. the results indicated that all combined models are, to some extent, suitable for predicting crop yield. among the algorithms used, the random forest model provided more accurate predictions due to combining the results of multiple decision trees and its ability to prevent data overfitting. this study recommends using the random forest model for future studies on saffron yield prediction based on vapor pressure deficit to manage water requirements.