شناسایی مناطق مطلوب کشت زعفران زراعی(.L sativus Crocus )با استفاده از مدلهای مبتنی بر هوش مصنوعی در استان خراسان رضوی

زعفران ).L sativus Crocus )از خانواده زنبق و از ادویههای ارزشمند صنایع مختلف است. در این پژوهش با تکیه بر مدلهای یادگیری ماشینی به بررسی زیستگاههای مناسب زعفران در خراسان رضوی پرداخته شد. نمونهبرداری نقاط حضور، در بازدیدهای میدانی طی دوره زمانی 1401-1400 صورت پذیرفت؛ در مجموع 59 نقطه حضور برای زعفران ثبت شد. با استفاده از اطالعات27 متغیر محیطی)3 متغیر فیزیوگرافی و 19متغیر اقلیمی، 4 متغیر خاکشناسی و 1متغیر زمینشناسی(مدلسازی انجام شد.

بررسی مقادیر شاخصهای ارزیابی صحت )KAPPA، TSS و ROC )نشان داد که؛ مدل جنگل تصادفی برای مناطق مستعد کشت زعفران با مقادیر به ترتیب 88/8 و 98 و 99/5 درصد و مدل ترکیبی با مقادیر به ترتیب 94/7 و 98/9 و 99/9 درصد برای این پارامترها بیشترین میزان صحت را داشتهاند. در مدلهای برگزیده جنگل تصادفی و مدل اجماعی مساحت بین 3195 تا 6144 کیلومترمربع معادل 2/74 تا 5/28 درصد از مناطق مورد بررسی پتانسیل متوسط تا خوب برای کشت زعفران است که بیشترین توزیع جغرافیایی زعفران را نشان دادند.

متغیرهای محیطی، میانگین دمای ساالنه، مدل رقومی ارتفاع و بارش ساالنه بیشترین سهم را بر تغییرات توزیع زعفران دارند. براساس نتایج تحلیل تناسب زیستگاه و استفاده از مدلهای یادگیریماشین میتواند به بهبود کشت و توسعه زعفران با محدودیتهای محیطی، به عنوان یکفرصت موثر برای تحقق کشاورزی پایدار و افزایش بهرهوری در تولید کمک کند. این پژوهش به مدیران و کشاورزان اطالعات مهمی ارائه داده و مسیری را برای انتخاب مناطق مناسب برای کشت زعفران با توجه به شرایط محیطی فراهم آورده است.

Introduction: Saffron (Crocus sativus L) holds a special place in the culture and economy of various countries as one of the most valuable and expensive spices. This plant, which isresistant to drought and capable of growing in specific climatic conditions, carries significant economic importance. Its cultivation in regions with limited conditions and low water requirements is considered an excellent opportunity for sustainable agriculture in upland and water-scarce areas. In Iran, saffron is cultivated as a strategic and exportable product, especially in regions like Khorasan, Kerman, Golestan, and Markazi. The cultivation of saffron comes with challenges such as water scarcity, soil pollution, decreased genetic diversity, and climate change, especially in arid and waterscarce regions, which is a cause for concern.

This research, has focused on the habitat suitability for saffron cultivation, investigated the environmental factors and their impact on the growth, yield, and quality of this product by using species distribution models. Additionally, the role of human interventions and climate changes in saffron habitat suitability and methods for increasing productivity and sustainability of saffron cultivation are discussed and examined. Materials and Methods: This research was conducted in northeastern Iran, covering an area of 117,612 square kilometers. Presence data were collected using 1:25,000 topographic maps and the features available on the maps. Relying on machine learning models and utilizing 11 algorithms within the BioMod2 package, an investigation into suitable saffron habitats in Khorasan Razavi was carried out. The presence points were documented during field inspections over the 2021-2022time period in the designated areas, resulting in a total of 59 presence points for the C.

sativus L species. A set of 27 environmental variables, including three physiographic variables, 19 climatic variables (1950-2000), four soil-related variables, and one geological variable, were considered for model development in the current time frame. The modeling process involved using 70% of the presence points for model creation and 30% for model performance evaluation.Additionally, to enhance modeling accuracy, five repetitions were considered. Model accuracy was assessed using the KAPPA, TSS, and ROC indices. Results and Discussion: The accuracy evaluation results in the modeling indicate that the Random Forest (RF) model, with values of 88.9%, 98%, and 99.5% for KAPPA, TSS, and ROC indices, and the Ensemble Models (ESMs) model, with values of 94.7%, 98.9%, and 99.9% for the same parameters, achieved the highest accuracy for predicting suitable saffron cultivation areas. Consequently, the Random Forest and Ensemble Models were chosen as the preferred models and served as the basis for further calculations. The relative importance of environmental variables in modeling the spatial distribution of suitable saffron cultivation areas indicates that the most critical environmental factors are climatic variables (BIO8, BIO15, BIO12, and BIO1), followed by physiographic parameters (elevation above sea level), and ultimately, soil-related factors. The results show that the selected models RF and ESMs cover an area ranging from 3,195 to 6,144 square kilometers, equivalent to 2.8% to 5.5% of the regions under investigation, with a medium to high potential for saffron cultivation, displaying the most extensive geographical distribution of saffron.

In summary, this study has demonstrated the high predictive capability of machine learning models for forecasting areas with similar potential for saffron cultivation, with strong performance evaluated through KAPPA, TSS, and ROC accuracy indices. The study findings suggest that the Random Forest model (RF) with a high accuracy rate of 88.9%, 98%, and 99.5% is the preferred model for this research. The results indicate that bioclimatic and topographic factors are the most influential elements in species distribution. In summary, we evaluated the performance of ensemble models in comparison to individual models using presenceabsence datasets and found that ensemble models generally outperform individual models in most situations. Conclusion: The analysis of habitat suitability and the use of machine learning algorithms to improve saffron cultivation in environmentally constrained areas represent an effective opportunity for achieving sustainable agriculture and increasing productivity in saffron production. This research aids managers and farmers in selecting suitable regions for saffron cultivation based on environmental conditions.

Enhancing management capabilities and productivity in saffron production can contribute to increased production and the economic development of regions associated with saffron cultivation. Considering that the crisis of unemployment and drought in the east of Iran in the past 30 years has led to an increase in immigration, the consequences of this crisis can be the disruption of the population centers in the east of Iran, which has caused people to seek refuge from villages and small population centers to the outskirts of big cities. At the same time, it will lead to crippling marginalization for the big cities of the country. Marginalization around cities causes cultural and security problems, increasing drug sales and crime in cities. For this reason, according to the ecological conditions of saffron plant, this plant can be cultivated in susceptible areas, and this is very useful for the economy of small population centers.