Description of the study sites
The study was conducted at two sites (Sirinka and Harbu) located in the semi-arid areas of Amhara region, northeastern Ethiopia. The Sirinka site is situated at an altitude of 1850 m above sea level (masl) with latitude 11° 45′ 00″ N and longitude of 39° 36′ 36″ E while the Harbu site is located at an altitude of 1450 masl with latitude of 10° 55′ 00″ N and longitude of 39° 47′ 00″ E. The region receives annual total rainfall of 945 mm with mean annual maximum and minimum temperatures of 27.3 °C and 13.6 °C, respectively. Rainfall in the region is low, erratic and uneven in distribution (Adem et al. 2016). The soil type is characterized as Eutric Vertisol (Adem et al. 2016). The study region is dominated by rugged mountains with undulating hills and valley bottoms. Both sites received bimodal rainfall with a small rainfall season that extends from February to April/May (locally known as Belg) and the main rainfall season (locally known as Kiremt) extends from June to September. Terminal water deficit caused by dry spells is a major constraint for crop production. Major crops grown in the study sites are sorghum, maize, chickpea, haricot bean, field pea, lentil, teff (Eragrostis teff). Mixed farming (crops and livestock) is a major production system. Monoculture is dominant whereas crop rotation (cereals with pulse crops) and intercropping are practiced to some extent. The majority of field crops are grown under rainfed conditions during the main rainy season and some crops such as field pea, teff, and mung bean are grown during the short rainy season. Sorghum can be grown in the short as well as during the long rainy seasons based on the nature of the sorghum cultivars (short or long maturing types). There are four distinct seasons in Ethiopia namely Summer (June, July and August), Autumn (September, October and November), Winter (December, January and February) and Spring (March, April and May) as indicated in Fig. 1.
Description of the DSSAT and CERES-sorghum model
The DSSAT technology is the most widely used software across many countries. Currently, it incorporates more than 42 different crops including cereals, grains, grain legumes, and root crops (Hoogenboom 2003). The DSSAT is the first package with weather simulation generators. Its process-oriented and is designed to work independently of location, season, crop cultivar, and management system. It is capable of simulating the effects of weather, soil water, genotype, and soil and crop nitrogen dynamics on crop growth and yields (Jones et al. 2003). DSSAT and its crop simulation models have been used in a wide range of applications in many countries. DSSAT integrates the effects of soil, crop phenotypes, weather, and management options and analyzes the results in minutes. The CERES-sorghum model is one of the models in the DSSAT with major components of vegetative and reproductive development, carbon balance, water balance, and nitrogen balance (Singh and Virmani 1996). The model can simulate the growth, development, and yield using a daily time step from sowing to maturity of the crop. Differences in growth, development, and yield of crop cultivars are affected by genetic coefficients (cultivar-specific parameters) which are inputs to the crop model. The model can simulate physiological processes that describe the crop response to weather factors such as temperature, precipitation, and solar radiation including the effect of soil characteristics on water availability for crop growth.
Model inputs
Field experiments and data collection procedures
For calibrating the crop model, field experiment was conducted at Sirinka in 2019 main crop season in a plot size of 10 m * 10 m replicated three times. In the analysis we considered each individual replicate as a pair data (observed-simulated) to calculate R2, RMSE, nRMSE and d statistis valuze for each parameters. Sorghum cultivar named Girana-1 was used as a test crop and was planted in a spacing of 0.75 m * 0.15 m. Recently recommended blended fertilizer (NPSB) with nutrient contents of 18.9% N, 37.7% P2O5, 6.95% S and 0.18% B was applied during the sowing time of the crop at a rate of 100 kg ha−1. Nitrogen fertilizer in the form of urea (46%N) was applied during the sowing time a rate of 25 kg ha−1 and additional 25 kg ha−1 was applied 35 days after the crop emergence.
For observation of anthesis date, physiological maturity date, grain-filling period five plants were randomly selected from each plot and tagged. Days to anthesis was recorded as the number of days from the date of sowing to the date at which 50% of the plants in a plot start heading. Days to physiological maturity was recorded as the number of days from the date of sowing to the date at which 75% of the plants in a plot physiologically matured. Grain-filling period is the numbers of days from 50% flowering to 75% physiological maturity, For those measurements on weight bases a sub-sample (from all plant part) were taken to dry in an oven for 72 h at 60 °C to a constant weight and their weights were determined by using a sensitive balance. Leaf area at 50% anthesis was measured by multiplying leaf length with maximum leaf width and was adjusted by correction factor of 0.75 (i.e. 0.75 * leaf length * maximum leaf width) as suggested by Francis et al. (1969). Thus, the Leaf area index (LAI) was calculated by dividing the leaf area by the sampled ground area. The crop model was evaluated using anthesis date, phenological maturity date and grain yield collected from field trial conducted in 2013, 2014, 2015 and 2017 at Sirinka.
Crop management data
Recommended management practices for sorghum crop are required as input by the model. Thus, information on planting date, planting method, planting distribution, plant population, row spacing, planting depth, cultivar selection, irrigation amount and schedule, fertilizer type and amount, and tillage type were obtained from the nearest Agricultural Research Centre at Sirinka located in the study region.
Soil data
About two weeks before sowing of the crop, soil samples were collected from 1.6 m soil depth near the experimental site for chemical and physical analysis. A total of four distinct soil horizons were identified. Soil samples were collected based on soil horizon and were analyzed for soil texture, pH, organic carbon, total nitrogen, available phosphorous, exchangeable cations, electrical conductivity, bulk density, drained upper limit of soil water content, lower limit soil water content and saturated water content. The soil texture was determined by the modified Bouyoucos hydrometer method (Bouyoucos 1962) using sodium hexametaphosphate as dispersing agent. The soil pH was determined potentiometrically using a digital pH meter in a 1:2.5 soil water suspension (Van Reeuwijk 2002). Organic carbon was determined by wet digestion method whereas total nitrogen was determined through Kjeldahl digestion, distillation and titration procedures of the wet digestion method (Black 1965). Available phosphorus was determined colorimetrically using Olsen’s method (Olsen 1954). The Cation exchange capacity was estimated titrimetrically by distillation of ammonium that was displaced by sodium from NaCl solution (Chapman 1965). The soil water dynamics were estimated by inputting soil texture, soil organic matter content and soil bulk density into a soil file creation utility program of the DSSAT software package.
Weather data and RCP scenarios
Daily data of maximum and minimum air temperatures (°C), daily rainfall (mm) and daily total solar radiation (M J M−2 day−1) for the period 1981–2020 was obtained from the nearest weather stations at Sirinka and Kombolcha. The Weather Man utility program of DSSAT 4.6 was used to convert the sunshine hours to solar radiation (M J M−2 Day−1). Future climate data for the 2030s (2020–2049) and 2050s (2040–2069) were obtained from the 17 CMIP5 GCM outputs run under RCP4.5 and RCP8.5 scenarios downloaded from International Center for Tropical Agriculture (CIAT) climate change portal (http://ccafs-climate.org/) and downscaled to the target site using MarkSim software (Jones and Thornton 2013). WorldClim V1.3 was used to interpolate the climate at the required point. This climate database may be considered representative of the current climatic conditions. It uses historical weather data from several databases. Thus, MarkSim uses the climate records for any given location. In this study, two climate change scenarios (RCP4.5, RCP8.5) were used to predict impact of projected climate change on sorghum production and to explore crop adaptation strategies. The study assumes CO2 fertilization effect on sorghum Thus, we used 380 ppm of CO2 for the baseline period whereas 423 ppm and 432 ppm were used for 2030s and 499 and 571 ppm for 2050s for RCP 4.5 and RCP8.5 scenarios, respectively (IPCC 2013). RCP’s are greenhouse gas concentration trajectories adopted by the IPCC for its fifth assessment (IPCC 2013). In the RCP4.5 scenario, Greenhouse gas (GHG) concentrations rise with increasing speed until the forcing is 4.5 W m−2 in the year 2100. This is a moderate emission scenario of concentration rise whereas, in RCP8.5, GHG concentrations rise with increasing speed until the forcing is 8 W m−2 in the year 2100. This is a high scenario of concentration rise.
Model calibration and evaluation procedures
The CERES-Sorghum model in the DSSAT model was calibrated using field experimental data of 2019 main cropping season conducted at Sirinka site. Calibration is defined as adjustment of model parameters so that the predicted results are very close to the results obtained from the field experiments. The model used genetic coefficients that determine phenology, growth, and yield characteristics of a given crop cultivar. The calibration of the model was performed through a trial and error method by applying small change (+ 5%) on each parameter and by adjusting the genetic coefficients that determine the phenology of the crop followed by yield and yield components. The adjusted genetic coefficients were used in the subsequent evaluation of the crop model. In the calibration and evaluation phases, the observed dates of anthesis, physiological maturity, and yield were statistically compared to the simulated values using a set of statistical approaches such as the root mean square error (RMSE) (Loague and Green 1991), normalized root mean square error (nRMSE), index of agreement (d) (Willmott et al. 1985), and coefficient of determination (R2). The RMSE is the standard deviation of the residuals (prediction errors). The residuals measure how far the data points are from the regression line. It tells us how concentrated the data is around the line of best fit. R2 is a statistical measure of how well the regression predictions approximate the real data points. An R2 of 1 indicates that the regression predictions perfectly fit the data. The Index of Agreement (d) developed by Willmott (1981) is used as a standardized measure of the degree of model prediction error and varies between 0 and 1. A value of 1 indicates a perfect match, and 0 indicates no agreement at all (Willmott 1981). The nRMSE gives the measure (%) of the relative difference between simulated and observed data. Less value indicates good fit of the model
$$RMSE = \sqrt {\frac{{\sum\nolimits_{i = 1}^{n} {\left( {P_{i} - O_{i} } \right)^{2} } }}{n}}$$
where n = number of observations, Pi = predicted value for the ith measurement and Oi = observed value for the ith measurement. Thus, lower value indicates good fit of the model.
$$\mathrm{n}RMSE=\frac{RMSE}{N}\times 100$$
where N is the mean of the observed variables. nRMSE gives the measure (%) of the relative difference between simulated and observed data. Less value indicates good fit of the model
$$d=1-\left[\frac{{\sum }_{i=1}^{n}{\left(Pi-Oi\right)}^{2}}{{\sum }_{i=0}^{n}{\left(\left|Pi-O\right|\right)+\left(\left|Oi-O\right|\right)}^{2}}\right]$$
The d-statistic was calculated as (0 ≤ d ≤ 1). The more values close to unity are regarded as best agreement between the predicted and observed data (Musongaleli et al. 2014). When d = 1 indicates excellent. Where n: number of observations, Oi and Pi are the observed and predicted values, respectively for the ith data pair; and O is the mean of the observed values.
Analysis of impact of projected climate change on sorghum
The CERES-sorghum model in combination with the seasonal analysis program in DSSAT was used to simulate phenology, growth, and yield of sorghum under the present and future climate conditions of the study area. The sorghum cultivar (Girana-1) was used as the test crop. Simulations were carried out for the baseline period (1981–2010) and for the projected climate changes in 2030s (2020–2049) and 2050s (2040–2069) under RCP4.5 and RCP8.5 scenarios. In this study, all the simulations were started on July-2 as it is the average planting date for sorghum in the area most farmers practice. The long rain season usually starts at the end of June to the first week of July. Thus, this study also assumed that the soil profile was at the upper limit of soil water availability in that date and the crop was grown under rainfall conditions in the model. It also assumed that soil condition, crop management practices and crop cultivar characteristics are similar to the present situation. Thus, the response of the sorghum cultivar to future climate was evaluated using typical soil and crop management practices (fertilizers application rates, row spacing, planting date, planting method etc.). In the simulation, the crop was planted in 0.75 m * 0.15 m spacing using blended (NPSB) and Urea fertilizers at rates of 100 kg ha−1 and 50 kg ha−1, respectively. This study also assumed no problems of insect, disease and weeds during the simulation periods. The outputs from the crop model like days to anthesis, days to physiological maturity, grain yield and seasonal crop transpiration were computed. The change in phenology and yield were compared as followes.
$$\mathrm{change in antheis or physiological maturity }(\mathrm{\%})=\frac{\mathrm{X predicted}-\mathrm{X base}}{\mathrm{X base}}*100$$
where, X is anthesis or physiological maturity
$$\mathrm{change in grain yield }\left(\mathrm{\%}\right)=\frac{\mathrm{Y Predicted}-\mathrm{Y base}}{\mathrm{Y base}}*100$$
where, Y is grain yield.
Analysis of management scenarios for sorghum
The effect of changes in sowing date, nitrogen rates, and supplemental irrigation were evaluated as sorghum adaptation strategies for their effectiveness to sustain production in the study region. The sowing window for sorghum in the study region is between mid-June and mid-July. Accordingly, the sowing window was categorized as early sowing date, standard (normal) sowing date and late sowing date. Based on this category three sowing dates (15th June, 30th June, and 15th July) were selected. In this regard, sowing on 15th June was considered as early sowing while sowing on 30th June was the normal sowing date as it was practiced by most farmers whereas sowing on July 15 was considered as late sowing for sorghum. Effect of nitrogen was evaluated at three levels (0, 46, and 92 kg N ha−1) and was applied in the form of Urea fertilizer (46%). Regarding irrigated treatments, 100 mm water was applied as supplemental irrigation in ten days interval starting the anthesis period of the crop to reduce the effect of terminal water deficit. Thus, two levels of irrigation treatments (rainfed and supplementaly irrigated) were evaluated. The simulation analysis was performed individually and in combination for all the factors indicated above (change in sowing date, nitrogen rates, and supplemental irrigation) to find the most promising adaptation strategies. Finally, simulated output data were analyzed using analysis of variance (ANOVA) techniques using a statistical analysis system (SAS 2009). Means were compared using the least significant difference (LSD) at 5% probability level. The simulation years were considered as replications as yield in one year under a given treatment was not affected by another year (prior year carryover of soil water was not simulated). Simulation years were unpredictable weather characteristics and therefore formal randomization of the simulation years was not needed.