Optimal Algorithm for Zonation of Spring and Autumn Frosts in Kurdistan Province, Using NOAA-AVHRR Images

Introduction
Frost as a harmful climatologic phenomenon affects various human activities and biological processes. Due to mountainous nature of Kurdistan Province as the study area of this research in one hand and type and diversity of effective air masses on the other, the frequency, severity and duration of this phenomenon in the province are outstanding. Confronting frost could be one of the major programs of the province authorities especially in the agricultural sector. Satellite images can be a good way to study the frost because of the continuity and repeatability of the relevant data. With the previous studies and the lack of researches related to the frost and estimation of land surface temperature by satellite images in the country, this study aims at determining the optimal algorithm in order to study and extract the earth's surface frost zones in the spring and autumn in the region using night-time images of the AVHRR sensor.
 
Materials and Methods
The study area in this research is the Kurdistan Province located in the west of Iran. To do this, the daily temperature data from seven weather stations of the region in a 10-year period (2001-2010) were used. After the spring and autumn frost dates were determined in the stations, 24 night-time images were taken from NOAA satellite website, and their thermal bands (AVHRR channels 4 and 5) were used to calculate the temperature. Then, the satellite images were corrected geometrically by ENVI software using GCP files of images, and after that radiometric calibration was performed by histogram equalization method. Likewise, thermal band radiances and brightness temperatures were calculated. To calculate the surface emissivity (ε), the land-use layer must be taken into account. NDVI values were used in this study so that 10 daily images for each year (5 images for spring and 5 images for autumn) in total 100 images for the study period were taken and accordingly, NDVI values were calculated for the images. After eliminating the cloudiness effect and calculating the surface emissivity, land surface temperature equation was applied to the images with the surface emissivity. From the proposed various techniques for calculating the surface temperatures, the Split-Window Technique (SWT) was used in this research and finally, three algorithms were used for calculating the night-time temperatures namely Price (1984: 7236), Coll et al. (1994: 113) and Ulivieri et al. (1994: 62). The equations used in the mentioned algorithms were applied to the images using ERDAS software. The last step in the study was to validate the estimates by comparing the temperatures derived from satellite images with recorded ones at the weather stations using three indexes, namely Mean Absolute Error (MAE), Mean Bias Error (MBE) and Root Mean Square Error (RMSE), and consequently, calculating the correlation coefficients between these two temperature series.
 
Results and Discussion
Land surface temperatures in the Kurdistan Province were estimated using NOAA satellite images by Price, Coll, and Ulivieri algorithms. The resultant error values from MAE, MBE and RMSE indexes indicated a better match between temperatures derived from Coll algorithm and the observed ones in the weather stations. Observed and estimated temperatures based on Coll algorithm at the stations for all of the 24 selected images were presented in a table. After doing corrections and applying various algorithms on the satellite images, temperature zoning maps were prepared to extract and analyze the frost zones. The number of these maps was equivalent to the number of used images, i.e. 24. In some images there were pixels, identified as white color, without any information. Coll algorithm's temperature estimation errors in the stations is ranged from -0.1° to 6.3°C according to MAE, MBE and RMSE indices. Statistically significant correlations were also found at the 0.01 level between observed and estimated temperatures at Sanandaj, Marivan, Bijar and Zarrineh-obato stations, and at the 0.05 level of confidence at Saghez, Ghorveh and Baneh stations.
 
Conclusion
NOAA satellite data are used by scientists and researchers of different fields to separate temperature zones because of their appropriate temporal, spatial and spectral resolution. In this study, we tried to analyze the spring and autumn frosts in the Kurdistan Province using NOAA-AVHRR images, and prepare zoning maps derived from the optimal algorithm. Out of the three algorithms used to estimate the land surface temperature, Coll algorithm led to better results. The use of NDVI index in calculating the surface emissivity was also helpful in estimating the temperature. It is noteworthy that in all three algorithms, temperature estimates at Sanandaj, Marivan and Saghez stations- which have lower altitudes- were better than the elevated stations like Zarrineh-obato, Ghorveh and Bijar. There seems to be a direct or indirect relationship between the altitude and the accuracy of estimation. This needs to be investigated. Generally the elevation role in the occurrence of night-time frost in the province is obvious, both in terms of intensity and development.