کاربرد گوگل ارث انجین در استخراج تغییرات پوشش گیاهی و تفکیک کشت محصولات فاریاب (مطالعه موردی: دشت رودبار)

ارائه نقشه اراضی زیر کشت آبی، از ارکان اصلی در سیاست‏گذاری‏‏ های حوزه کشاورزی و برنامه‏ریزی‏های بهبود مدیریت آب است. این موضوع در دشت رودبار که اینک با بیلان منفی روبه‏رو است، اهمیت زیادی دارد. در این مطالعه، از الگوریتم ناپارامتری یادگیری ماشین بردار پشتیبان برای پردازش و طبقه‏بندی اراضی زیر کشت آبی با استفاده از تصاویر ماهواره‏ای لندست استفاده شد. فرایند طبقه‌بندی نیز در بستر گوگل ارث انجین (gee) صورت گرفت و از زبان برنامه‌نویسی javascript برای پس‏پردازش‏ها و شاخص ndvi برای ارائه نقشه اراضی زیر کشت آبی استفاده شد. در تدوین کد مورد استفاده، سال به سه دوره چهار ماهه تقسیم شد، سپس تصاویر ماهواره لندست 8 در هر دوره (2013 تا 2021) استخراج و برای آن، تصویری بر اساس حداکثر مقدار پیکسل‏ها تولید شد. بدین منظور با استفاده از الگوریتم mvc (ترکیب مقادیر حداکثر)، تصاویر موجود در هر دوره بررسی شد و برای هر پیکسل حداکثر مقدار متناظر آن، بین تمام تصاویر به عنوان ارزش نهایی آن پیکسل در نظر گرفته و در نهایت، تصویری جدید ایجاد شد. در مرحله بعد با ترکیب سه تصویر تولید شده (با استفاده از ترکیب رنگ کاذب) و اختصاص هر کدام از تصاویر به یکی از باندهای قرمز، سبز و آبی، تصویر جدیدی به وجود آمد و از آن برای استخراج نقشه نوع کشت استفاده شد. برای بررسی دقت طبقه‌بندی مناطق زیر کشت نیز از نمونه‌های آموزشی زمینی و تصاویر با وضوح بالا (گوگل ارث)، همچنین ادغام با مجموعه داده‌های موجود و استفاده از دانش تخصصی و محلی در منطقه مورد مطالعه استفاده شد که دقت طبقه‌بندی کلی 81٪ بود. همچنین بررسی تغییرات شاخص ndvi طی سال های 2013 تا 2021 نشان داد که در سال 2013، کمترین مقدار پوشش گیاهی و در سال‏های 2019 تا 2020، بیشترین مقدار پوشش گیاهی وجود داشت.

application of google earth engine to map the vegetation cover and separation of irrigated cultivated areas at rudbar plain

1- introductionimprovements in classification accuracy over irrigated areas are essential to enhance agricultural water management and inform policy and decision-making on water management and land use planning. advances in remote sensing technologies in conjunction with the emergence of big data and cloud-based processing platforms such as google earth engine (gee) are facilitating the classification of irrigated areas within improved accuracies in a timely and cost-effective manner, thus, enhancing the monitoring of these factors at both local and global scales. this process is aided by freely available high-resolution remotely sensed products and novel non-parametric machine learning algorithms for land use classification. this issue is very important in less developed areas such as the south of kerman province where the economy is based on agriculture. groundwater use in agriculture is soaring in arid and semi-arid regions such as rudbar plain having a negative balance of underground water. in the same regard, this study applied google earth engine to classify irrigated areas in rudbar plain.2- methodologythis study used a non-parametric machine learning algorithm, i.e., support vector machine algorithm, to classify near-accurate irrigated areas using high-resolution satellite images. all the steps of this study, including preprocessing, classification, and accuracy assessment, were performed within the gee platform. preprocessing included cloud/snow masking and maximum imagery generation, and classification was based on the support vector machine algorithm. to this end, a gee javascript code was used to access and analyze the data.  time series of vegetation indices, such as the normalized difference vegetation index (ndvi), are widely used for crop mapping. therefore, in this study, we proposed a method for compositing the multi-temporal ndvi, in order to map irrigated areas with the landsat 8 images in google earth engine. the algorithm composites the multi-temporal ndvi into three key values including ndvi1, ndvi2, and ndvi3. so at first, the year was divided into three periods of 4 months. then the maximum ndvi values for each pixel during each period were calculated. for this purpose, the maximum value composite was used to convert 16 days resolution ndvi data into maximum ndvi data for each period. therefore, the three data sets, namely ndvi1, ndvi2 and ndvi3, which respectively correspond to the first four months of the year, second four months of the year and third four months of the year were calculated. to this end, a gee javascript code was applied to images. the classification process was automated on a big data management platform, i.e., the google earth engine (gee). irrigated area is specified using false color combination with the selection of ndvi1, ndvi2 and ndvi3 indexes intended for the development of rgb. the existing datasets were used to train and validate the land cover. a random sampling method was used to balance the number of training point’s classifications. landcover categories were grouped into five types to separate cultivated areas from the rest of the land uses.3- results in this study, a new method for identifying irrigated lands was introduced using google earth engine. the approach enhanced the classification accuracy of irrigated areas using ground-based training samples and google earth and fusion with existing datasets and the use of expert and local knowledge of the study area. the overall classification accuracy was 81%. as a result of this study, maps of cropping patterns include five category 1: date palm trees and citrus fruits, 2: potatoes and onions; 3: tomato; 4: cucumber and 5: sweet corn, sesame and sour tea. the areas of cultivation of category 1, category 2, category 3, category 4, and category 5 are respectively 170 km2, 283 km2, 133 km2, 56 km2 and 277 km2.  also, the vegetation changes were investigated during the years 2013 to 2020. the results demonstrated the cover of low vegetation in 2013 and 2018 and the cover of high vegetation in 2020.4- discussion & conclusions the combination of methods and approaches in gee facilitated the rapid classification of more accurate irrigated areas with petabyte volume big data. the developed dataset of the cultivated areas has an overall accuracy of over 81%. given that there is no specific pattern and plan for cultivating agricultural products in rudbar plain, the enhanced outputs of the irrigated area mapping are essential for policy and decision-makers to assess vast and complex irrigation systems’ performance in detail. they are critical for the accurate monitoring of irrigation activities from the field to transboundary or national scales. by examining the area under cultivation, it was found that seasonal cultivation is more popular with farmers than multi-year cultivation, and more attention should be paid to the marketing of agricultural products.