كليدواژه :
مقياس سازي , نقشه برداري , داده هاي كمكي , خاك
چكيده فارسي :
مقياس مكاني مورد نياز اطلاعات منابع طبيعي با آن مقياسي كه در دسترس مي باشد، بسيار متفاوت است. يك روش براي تطابق مقياس ها، استفاده از مفهوم مقياس سازي است. در اين مطالعه با استفاده از روشي جديد يزرگ مقياس سازي نقشه رقومي كربن آلي خاك از قدرت تفكيك مكاني 150 متر به 30 متر در منطقه اي به وسعت 300هكتار واقع در استان كردستان انجام شد. در ابتدا، نقشه رقومي كربن آلي خاك با قدرت تفكيك مكاني 30 متر با استفاده از داده هاي
نمونه برداري شده از سطح خاك، داده هاي كمكي (استخراج شده از مدل رقومي ارتفاع و تصاوير ماهواره لندست) و مدل درختي به دست آمد؛ سپس، با استفاده از روش كوچك مقياس سازي ساده اي به نام ميان گيري بلوكي نقشه كربن آلي خاك با قدرت تفكيك مكاني 150متر تهيه گرديد. در مرحله بعد، با استفاده از الگوريتم بزرگ مقياس ازي،
نقشه كربن آلي خاك از قدرت تفكيك مكاني 150 متر به 30 متر تبديل گرديد. با فرض اين كه ارتباط داده هاي كمكي و كربن آلي خاك مي تواند يك رابطه غير خطي باشد، در اين مطالعه از روش رگرسيون تعميم داده شده و شبكه عصبي مصنوعي استفاده گرديد. براي ارزيابي الگوريتم بزرگ مقياس سازي از ضريب تبيين بين نقشه حاصل از مدل درختي و نقشه بزرگ مقياس سازي بهره گرفته شد. نتايج مدلسازي در مرحله اول نشان داد كه بعضي از متغيرهاي كمكي مانند شاخص گياهي نرمال شده، شاخص خيسي، شاخص همواري دره با درجه تفكيك بالا و انحناي طولي شيب بيشترين تاثير را بر پيش بيني كربن آلي خاك دارند. نتايج پيش بيني رگرسيون درختي در مرحله آزمون نيز نشان داد كه مدل به خوبي توانسته كربن آلي خاك با دقت مكاني 30متر را مدل سازي كند (ريشه مربعات خطا برابر با 0/15 و ضريب تبيين 0/78 مي باشد). نتايج بزرگ مقياس سازي نيز نشان داد كه روش مورد استفاده جهت بزرگ مقياس سازي (شبكه عصبي مصنوعي و رگرسيون تعميم داده شده) به خوبي با ضريب تبيين 0/81 و 0/70 توانسته اند تغييرات مكاني كربن آلي خاك را در مقياس بزرگتر مدل سازي كنند.
چكيده لاتين :
Introduction There is no high-resolution map which describes SOC in Iran. Moreover the, traditional soil mapping approach currently used in Iran is greatly time consuming and cumbersome, particularly, when a large number of samples are required. Therefore, to overcome this problem, the application of digital soil mapping (DSM) techniques could be an efficient alternative approach. In digital soil mapping, soil properties are digitally mapped based on their relationships with the collection of cheaper-to-measure ancillary data. Previous studies indicated that the digital elevation model (DEM) and remotely sensed data are the most common ancillary data for soil organic C (SOC) prediction. However, the spatial scale of DSM products is often mismatched to the scale at which it is required. One way of harmonizing digital spatial data is the application of either up-scaling or down-scaling methods. Therefore, the main purpose of the present research is the application of a general method for downscaling.
Material and methods Our study area is located in Kurdistan Province, about 12 km northwest of Baneh, Iran. It lies between 36° 01′ and 36° 05′ North latitudes and 45° 40′ and 45° 50′ East longitudes with a total area of 3000 ha. The climate is semiarid with distinct differences between dry and wet seasons. The average annual rainfall and temperature are 700 mm and 13.8 °C respectively. Soil moisture and temperature regimes are Xeric and Mesic, respectively. Conditional hyper latin cube sampling (cHLS) was applied to collect data in the field. Then, 188 representative soil profiles were digged, described and sampled according to soil survey manual. This resulted in 585 soil samples that were analysed in the laboratory to measure SOC. SOC was then measured in each sample. To make a relationship between SOC and auxiliary data a regression tree model was adopted. In this paper, a new method for down-scaling of SOC data from 150-m to 30-m spatial resolution was applied and compared. Firstly, the digital map of SOC was prepared using sampled data, auxiliary data (i.e. derived from DEM and Landsat 8 ETM+ images) and regression tree model. Then, a map of SOC with 150-m grid resolution was prepared through a simple method of up-scaling termed averaged blocking. After that, the SOC map with 150-m by 150-m grid resolution down to a 30-m by 30-m grid resolution using downscaling method. To assess the performance of downscaling method, coefficient of determination was calculated between the downscaled SOC map and that which was created using the regression tree model.
Results and discussion Raw SOC data showed a normal distribution pattern. The min and max SOC values ranged from 0.21 to 1.83 percent. The coefficient of variation for SOC was high (more than 36.00%), which indicated a broad range of values across the study area. The results showed that the regression tree model could predict SOC with reasonable accuracy up to R2=0.78. Moreover, Landsat spectral data and terrain parameters were found to be the most useful auxiliary data for mapping of SOC across the study area. More specifically, normalized vegetation index (NDVI), wetness index (WI), multi-resolution valley bottom flatness index (MrVBF), profile curvature (PrC.) and the first principal component (PC1) of six Landsat spectral data had the highest correlation with the SOC data. This is unsurprising because it is well-confirmed that NDVI is related to total plant cover. Therefore, it could be correlated with SOC linked to vegetation density. In fact, the SOC storage at the soil surface is mainly controlled by the balance of carbon inputs through plant residues which is controlled by vegetation density. Results also indicated the down-scaling approaches (i.e. artificial neural network and generalized additive regression model) could map better the spatial variability of SOC in finer resolutions (30-m) with R2 of 0.81 and 0.70, respectively. The differences in the obtained R2 values could be ascribed to their various mathematical assumptions.
Conclusion In this paper, a general method defined by Malone et al. (2009) was adapted to down-scale the coarse spatial resolution of SOC to that of SOC map with finer resolution. Our results indicated that the above-mentioned methods (artificial neural network and generalized additive regression model) had a good performance for down-scaling the SOC data from 150-m to 30-m. Overall, fine-resolution soil maps are useful for many soil and environmental scientists and land managers in Iran. Therefore, we recommend the approach applied in this study area be used to map the SOC at the desired resolution in other parts of Iran, particularly those areas having the same agro-ecological zones.
