Mapping of forage Production in Poor Rangelands Haftkel Rangelands Using Sentile-2 Images

Background and objectives: Determining the exact amount of forage production can be of great help to rangeland managers and relevant specialists in determining proper stocking rate. With implementing proper sampling design, remote sensing data could be used to accurately estimate forage production due to the extent of rangelands areas, cost, time spent and other problems in data gathering from field. The objectives of this study was to select the best model and the best remote sensing index in order to map forage production using field data and vegetation indices of NDVI, SAVI, MSAVI2, DVI and GCI extracted from satellite images of Sentinel 2. Methodology: A sampling network with a total of 58 plots (1×1 meters) were established in the studied area and cut and weight method was used to measure forage production. Then, vegetation indices of NDVI, SAVI, MSAVI2, DVI and GCI were created with SNAP software. The values of the mentioned indices were extracted from the location of the plots, using the ArcGIS 10.4 software. The normality of the data was checked by the Kolmogorov Smirnov test. Then their relationships were analyzed with regression in SPSS 16 software. Also, multiple linear regression was used to investigate the relationship between plant indicators and forage production. The train model was created by 70% of the total plots and 30% of the data were used to test the model. Coefficient of determination (R2) and root mean square error (RMSE) were used to select the best model and index or indices. Finally, the selected model was used to create the map of forage production (Kg/hec), using ArcGIS 10.4. The values of final map as estimated data and a total of 58 plots as observed data were evaluated by independent t-test. Results: The results related to the relationship between forage production and plant indices with univariate linear regression showed that all the used indices had a significant relationship with forage production. The univariate linear regression model with MSAVI2 index had the highest coefficient of determination and the lowest RMSE (Y= 649.3-8523.7×MSAVI2; R2= 0.68 and RMSE= 16). The results also showed that the accuracy of the DVI index (R2= 0.66; RMSE= 19) was higher than the NDVI index (R2= 0.58; RMSE= 22) for estimating forage production in studied area. By applying the assumptions of multivariate linear regression model, only two indices of GCI and MSAVI2 were included in the model, and the amount of R2 and RMSE were the same as univariate linear model with MSAVI2 index. The results of independent t-test indicated that there were not significant differences between observed data and the ones estimated by selected model (p<0.05). The minimum, mean and maximum of forage production in the final map were 10, 220 and 475 kg/hec, respectively. Conclusion: According to the equality of the root mean square error and the coefficient of determination of the multiple and linear regression models and also the results of independent t-test that indicted no differences between observed and estimated forage production, we suggest using the MSAVI2 index to estimate forage production in warm semi-arid rangelands.