Development of an agricultural drought assessment system : integration of agrohydrological modelling, remote sensing and geographical information

Iran faces widespread droughts regularly, causing large economical and social damages. The agricultural sector is with 80-90 % by far the largest user of water in Iran and is often the first sector to be affected by drought. Unfortunately, water management in agriculture is also rather poor and hence water productivity of crops WP is far below potential. The growing water scarcity due to drought and the increasing water demands of industries, households and environment, are major threats to sustainable agricultural development in Iran. Therefore, the development of a reliable agricultural drought assessment system would be very beneficial for proper operational decision making on farms, for early warning, for identification of potential vulnerability of areas and for mitigation of drought impacts. Given the current water scarcity, the limited available amount of water should be used as efficient as possible. To explore on-farm strategies which result in higher WP-values and thus economic gains, the physically based agrohydrological model Soil Water Atmosphere Plant (SWAP), was calibrated and validated using measured data at 8 selected farmer’s fields (wheat, fodder maize, sunflower and sugar beet) in the Borkhar irrigation district in Iran during the agricultural year 2004-05. Using the calibrated SWAP model, on-farm strategies i.e. deficit irrigation scheduling, optimal irrigation intervals and extent of cultivated area, were analyzed based on relations between WP- indicators and water consumption. The results showed a large potential of the improvement of water productivity under limited water supply in the Borkhar irrigation district. Although agrohydrological models like SWAP offer the possibilities for predicting crop yield, such models may become inaccurate because of uncertainty of input parameters like irrigation scheduling, soil hydraulic parameters and planting dates. This holds especially true when applying distributed modelling at regional scale. Hence to reduce the uncertainty in application of SWAP at regional scales, remotely sensed data of leaf area index and evapotranspiration were used in combination with a geographical information system. The remotely sensed data were inserted into the distributed SWAP model using data assimilation techniques i.e. sequential updating. Data of LAI were derived from Visible and Near Infrared (VNIR) spectral bands of remote sensing data with moderate to high spatial resolution. However, due to resolution limitations of existing remotely sensed data i.e. thermal bands, these data could not be used directly for routine ET estimation of individual fields. Therefore, a new disaggregation method based on linear disaggregation of ET components within each MODIS pixel, was developed and applied to the simulated MODIS data. The results of the proposed approach were further compared with two other disaggregation approaches being based on weighted ratios, as derived from dividing ET maps of high and low spatial resolution data. The biggest advantage of the proposed linear disaggregation approach was that the number of high spatial resolution images needed in this method is low, i.e. the approach can even be applied using one land cover map only. As in many regions access to high spatial resolution thermal images is currently not possible, the linear disaggregation method can still be used to assess drought impacts far in advance. Water balance components as computed by SWAP are quite sensitive to the upper boundary conditions, and hence to irrigation times and application depths. In order to know how much water has been applied, the cumulative actual ET data were therefore used in an automatic calibration mode, i.e. inverse modelling of irrigation scheduling. The ability of inverse modelling to reproduce the initial irrigation times and depth, was first investigated using forward cumulative SWAP simulated ET data based on 5, 15 and 30 days. Thereafter, the cumulative disaggregated remotely sensed ET data based on 5 days were used in the inverse modelling process. The results showed that the performance of inverse modelling is promising in identifying the irrigation time and depth of irrigation using 5 days based cumulative ET data. However, irrigation amounts, which rewet the soil profile beyond field capacity and thus cause excessive percolation, could not be detected by the applied inverse modelling approach. Also, assimilation of remotely sensed data into a distributed SWAP by automatic calibration needed a large amount of computation time, especially at regional scale. Hence, to insert the valuable information from remotely sensed land surface data into the SWAP model at regional scale, a simple updating assimilation technique was used. The SWAP model was implemented in a distributed way using the spatial distributed information of soil types, land use and water supply on a raster basis with a grid size of 250 m. In order to link spatial information data with SWAP, a coupling program was written by the author in MATLAB. This program took care of the transfer of in- and output data from one system to the other, as well as to run the model for each pixel. To have a prediction of crop yield far in advance, the sequential updating process of remotely sensed based data (LAI and/or relative evapotranspiration ET/ETp) was halted at one respectively two months before the end of the wheat growing season. During the sequential updating process known weather data were used, while for the remaining part of the growing season different scenarios were considered based on weather data of a dry, wet and normal year. A value for the optimum gain factor Kg, that performed best with respect to the observations, was selected Simulation with assimilation of both LAI and ET/ETp -data at both the regional and field scale (bias about %) was very promising in forecasting crop production one month in advance. However, longer term predictions i.e. two months in advance, resulted in a higher bias between the simulated and statistical data. It appeared that in the assimilation process LAI data have a dominant influence. Because of this dominant influence, it is suggested to repeat the assimilation process using the LAI data of the most advanced satellite i.e. IRS-P6 (ResourceSAT1&2) with higher spatial and temporal resolution. The surface water in the Borkhar irrigation canal network is provided by diversion of the water from the Zayande Rud river. Since this river is mainly fed by the snow melt from January to April, a comprehensive drought assessment system on seasonal basis can be developed by integration of the developed agricultural drought assessment system with the estimates of available surface water being derived from snow pack and snow cover.