Evaluation of PDIR-Now satellite-based precipitation data in Chaharmahal and Bakhtiari Province

Introduction
The first step in understanding basins is measuring different climatic and hydrological variables and examining their relationships. Primary variables include temperature, precipitation, evapotranspiration, water infiltration rate in the soil, flow discharge, etc. Meanwhile, precipitation is one of the most important and effective variables. The inappropriate distribution of rain gauge stations in different regions of developing countries, on the one hand, and the development of remote sensing sciences on the other hand, have led to the increasing use of precipitation products estimated from satellite images. Therefore, it is important to know the characteristics of these products and their possible errors. One of these products is Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN) satellite-based precipitation data. Over time, this satellite product has been developed and introduced as PERSIANN-CCS and PERSIANN-CDR. One of the shortcomings of these products is that in humid and arid areas, they report  underestimation and overestimation of precipitation, respectively. To solve this problem, PERSIANN product development experts designed a new product, the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks - Dynamic Infrared Rain Rate near real-time (PDIR-Now) product, in 2019. This product creates a dynamic relationship between the precipitation rate and cloud brightness temperature by consideringground conditions affecting the precipitation phenomenon. This has led tosignificant advantages in this algorithm compared to other quantitative precipitation estimation algorithms.
 
Materials and Methods
In this research, the performance of the PDIR-Now product was analyzed. Chaharmahal and Bakhtiari Provinces, despite their small area, have about 10% of the country's water resources and play an essential role in supplying water resources for the neighboring provinces; therefore, it was chosen for this research. Its area is 16421 km², located in the southwest and part of the western mountainous belt of Iran, and with an average height of 2282.7 m above sea level, it is a high-altitude region in terms of topography. To conduct this research, 27 rain gauge, climatology, and synoptic stations that contained data for a common period (2005 to 2020) were selected and their precipitation information was collected monthly and annually, removing statistical deficiencies in some years. In the next step, PDIR-Now information was extracted at scales corresponding to the data of gauge stations on the reference site, and 208 received images were georeferenced and processed in the ArcMap environment. Then, these values were compared with the corresponding precipitation values of ground stations using three coefficients: Nash-Sutcliffe (NS), correlation (R), and relative bias (RB). In the next step, using the IDW geostatistics method, zoning maps of the province were created based on the correlation coefficient value, separately for each of the 12 months of the Gregorian year and also on an annual scale.
 
Results and Discussion
The results indicate that the best relationship between PDIR-Now and the precipitation data from ground stations is established in November, where 100% of the stations have an R higher than 0.5 at the 5% significance. Also, in this month, 88.8% of the stations recorded NS values greater than 0.5. The lowest level of correlation was related to May, where 33.3% of the stations had an R greater than 0.5, and 11.1% of the stations showed NS values greater than 0.5. In general, the best relationship is established between PDIR-Now data and ground station precipitation values in the rainy months, especially when rain dominates over other forms of atmospheric precipitation. The weakest relationship is related to January in the rainy season (November to April). In this province, atmospheric precipitation in January is mostly in the form of snow. Since the characteristics of clouds, including surface temperature and brightness temperature, differ in rain and snow conditions, this could resulted in the lack of occurrence in January. Additionally, on an annual scale, 74.07% of the stations have an R greater than 0.5, and 55.5% recorded NS values above 0.5. The best correlation is for Armand station with coefficients of 0.63, 0.83, and 0.01 for NS, R, and RB, respectively. In contrast, the same coefficients for Bardeh station were less than 0.00, 0.35, and 0.356, respectively, indicating the lowest fit between the two data groups at this station.
 
Conclusion
The general results show that PDIR-Now performs better in the rainy season in the southern and central regions, which are located in the low-altitude areas of the province, where precipitation amounts approach the average. In the low-rainy season, stations with precipitation levels close to the average, rather than the minimum, yield better results. These results were consistent with the findings of other research conducted in this field in Iran and the world in the following ways: 1) PDIR-Now, like many other products, underestimates heavy rainfall but performs better in estimating low and medium rainfall. 2) The performance of most satellite-based precipitation products decreases for altitudes above 1000 m. 3) PERSIAANN satellite-based products are more accurate at the monthly scale than at the annual scale. Additionally, the relationship between ground station data and satellite products in the central regions of the province, which receive average rainfall, consistent with the results of this research. In the present study, the density of rain gauge stations was low in the steep and inaccessible parts of the province, and errors in rain gauge data recording remain a Limitation. However, the number of long-term stations with sufficient statistical intervals in other regions is a significant advantage.