Site description

The station is located at 23.45˚N and 90.23˚E and 8 m above mean sea level. The study area experiences a sub-tropical monsoon climate. Seventy percent of the total rainfall is received during July to September, with average annual rainfall of 2148 mm. April is the hottest month, with average minimum and maximum temperatures of 23.6˚C and 33.7˚C, respectively. The coldest month is December where the average minimum and maximum temperatures are 14.1˚C and 26.4˚C, respectively. Average sunshine hours are 6.52 hr∙day⁻¹. The soils of the study site were characterized as silt clay loam with moderate drainage.

DSSAT was developed by an international network of scientists, cooperating in the International Benchmark Sites Network for Agrotechnology Transfer project (IBSNAT Project). DSSAT version 4.6 was utilized for the study. The DSSAT simulate crop growth, development and yield using a defined data set on crop management, minimum weather data and soil profile parameters. Some of the crop management data required to simulate DSSAT include crop, cultivar, planting date, row and plant spacing, fertilizer-N levels, tillage practices and organic amendments [7] . Also data inputs include physiological stages of crop growth such as anthesis dates and days to maturity. Minimum weather data sets consist of maximum and minimum temperatures, solar radiation and rainfall. If sunshine hours data is available, then Weatherman module in DSSAT coverts it into solar radiation. Soil profile parameters include depth of soil, soil physical and chemical characters.

Weather data were collected from the weather station located at BRRI Gazipur, which is very near to experimental site. Initial soil profile for the experimental site was collected for physical and chemical characteristics. The data on soil characteristics such as soil moisture, pH, bulk density, soil organic carbon and total-nitrogen were measured and stored in the soil input file (Table 5). The soil of the experiment was specified as silty clay loam. The crop management data were recorded throughout the growing seasons. The input files, such as weather file, soil file, experimental file and A & T files (average measured data file, annual as well as temporal), were prepared to calibrate and validate the Crop Environment Resource Synthesis (CERES)-Maize model. The first year experiment was used for calibration of the model for test cultivars and the second year experiment was used for validation purpose.

CERES-Maize model is a predictive and deterministic model designed to simulate maize growth, soil, water and temperature and soil nitrogen dynamics at a field scale for one growing season. The CERES-Maize model in DSSAT requires a set of six cultivar specific parameter for its calibration (Table 1). Four of them (P1, P2, P5, and PHINT) control the timing of phenological stages, and the remaining two (G2 and G3) characterize the potential yield under optimal conditions. By running of GLUE model, and simultaneous carrying out sensitivity analyses with these coefficients, the coefficients for the tested cultivars were computed, through process of calibration.

Two field experiments were carried out at the research field of Plant Physiology Division, BARI, Gazipur during rabi seasons of 2014-15 and 2015-16. The crops were grown under no stress condition. First year’s data were s used to calibrate genetic coefficients and second year’s data were used for the validation of model.

The site was previously fallow. Four adapted maize cultivars, BARI Hybrid Maize-7, BARI Hybrid Maize-9, NK-40 and Pioneer which are well adapted at farmer’s level were used in this experiment, with yield potential ranging from 9.5to 10.0 t∙ha⁻¹. Seeds of BARI Hybrid Maize-7, and BARI Hybrid Maize-9 were collected from Bangladesh Agricultural Research Institute and others seeds were procured from market. Plant population for each cultivar was 83,333 plants ha⁻¹ (60 cm × 20 cm). Seed rate used in the study was 35 kg∙ha⁻¹. The unit plot size was 4.2 m × 3 m. Seeds were sown on November 18, 2014 with 60 cm × 20 cm spacing, while sowing date in the second year was November 11, 2015. Fertilizers were applied at the rate of 250-55-100-30 kg∙ha⁻¹ N, P, K and S as urea, triple super phosphate (TSP), muriate of potash (MOP) and gypsum. One third of N and whole amount of TSP, MOP and gypsum were applied as basal at sowing. Remaining 2/3 N was top-dressed at 40 and 70 days after sowing (DAS). Irrigation was given as and when required to maintain adequate soil moisture available in the root zone. Standard agronomic practices were followed including weed and insects/pest control.

Soil characterization: Before conducting experiments, soil samples from 0 - 15, 15 - 30, 30 - 60, 60 - 90 and 90 - 120 cm depths were collected. The soil samples were analyzed for texture, bulk density, soil moisture, pH, organic matter, total N, field capacity, wilting point, bulk density, nitrate (NO₃⁻) and ammonium (NH₄⁺) concentrations. Results of various physico-chemical properties of soil have been summarized in Table 2(a) and Table 2(b). Hydraulic conductivity, bulk density and saturated moisture content were derived from in-built pedo- transfer functions module of SOIL of DSSAT.

Daily weather data for both the growing seasons, including precipitation (mm), minimum and maximum air temperatures (˚C), and sunshine hours were collated from the weather observatory. Solar radiation (MJ∙m⁻²∙d⁻¹) was calculated from daily sunshine hours by using Weatherman of DSSAT. The weather files were created by using WEATHERMAN module of DSSAT for running of the model.

Phenology: Every day we visited the crop field and recorded phonological information in terms of number of days required for attaining a particular growth stage. Ten plants from each plot were tagged at two-leaf stage which was used for destructive sampling to determine tassel initiation stage. Three randomly located plants of each cultivar from each replication were removed at regular intervals (every 2 - 3 day) starting from 12 days after emergence. The apex was dissected under a stereoscopic microscope and observed carefully. When the tip of tassel was visible under microscope data was recorded.

Fifty percentage tasseling date was recorded when tasseling was noticed in 50% of tagged plants. For observing physiological maturity we collected grains from different parts of the selected cobs like base, middle and tip when the cob became straw color at 2 - 3 days interval. We recorded physiological maturity date when 50% of the grains showed black layer which indicate that no further accumulation of assimilates in grain is possible.

Leaf area was measured by an automatic area meter (LI 3100 C, LI-COR, USA). For dry matter, leaves, stems and cobs were separated and dried in an oven at 80˚C for 72 hours. Photosynthetically active radiation (PAR) was assumed as 50% of solar radiation, which were used for radiation use efficiency calculation.

Grain yield and biomass at harvest were measured. As subsample, five plants were selected and the plant components were separated into stover husks and ears. The variables determined include the number of seeds per unit area (seed number∙m⁻²), seed weight (dry∙g∙m⁻²), cob weight (dry∙g∙m⁻²), and stover weight (dry∙g∙m⁻²). Maize was harvested at physiological maturity (BARI Hybrid Maize-7 and BARI Hybrid Maize-9 at 145 DAS, NK-40 at 146 DAS and Pioneer at 147 DAS). The yield components data were collected from 5 randomly selected plants prior to harvest from each plot. At∙harvest, the yield data was recorded plot wise and analyzed statistically.

The DSSAT crop models require genetic coefficients, which are specific for each cultivar to properly describe the processes related to growth, development and grain production. These coefficients allow the model to simulate performance of diverse genotypes under different soil, weather and management conditions [15] . The model was calibrated using field measured values of weather parameters, crop management and soil properties during 2014-2015 cropping season. Genetic coefficients were estimated by using observed anthesis and physiological maturity dates and grain yield of maize cultivars growing in 2014-15 and validation with data of 2015-2016 [16] [17] .

Since none of the cultivars were previously introduced within DSSAT, we created them in the genetic file (MZCER046.CUL) of DSSAT-CSM. Initial values of the genetic coefficients were obtained from the medium maturity group cultivar DEKALB 591, already available in the DSSAT. The computed crop specific parameters (CSPs) values for DEKALB 591 cultivar were copied into MZCER046 CUL file to operate the simulation. An iterative approach was used to obtain reasonable genetic coefficients through trial and error adjustments until there was a match between the observed and simulated dates of anthesis and physiological maturity and grain yield [18] [19] . The derived genetic coefficients were used for model performance evaluation. In this study, water and nitrogen balance simulation controls were switched off, to ensure that no stress for water or nitrogen was experienced in the course of crop growth.

Model performance was evaluated by comparing the simulated versus observed values from the experiment of 2015-2016. Data for model validation include anthesis date, physiological maturity date, grain yield, potential grain weight and above ground biomass. Comparisons between measured and predicted maize yield and its attributes were carried out through percent error propagation.

Statistical analysis, through coefficient of prediction (R²), percent error and p- values was done in yield and its attributing characters were carried out to judge the performance of simulation compared to the observed values.

The genetic coefficients, as obtained through run of GLUE as part of Calibration exercise, are shown in Table 3. The genotypic coefficient P1 was 225˚C day for BARI Hybrid Maize-7 and BARI Hybrid Maize-9, while that was 226˚C day for PIONEER and NK-40. The P2 was same for all the varieties and P5 was 956˚C day for BARI Hybrid Maize-7 and BARI Hybrid Maize-9, while that was 965 for PIONEER and 964 for NK-40. There was considerable variation among the cultivars in G2 and G3 but PHINT were the same amongst the cultivars. PHINT is critical in determining the duration of vegetative growth in maize, it is lower in temperate but higher in tropical climate. No cultivar specific parameters were found in literature for comparing the values we generated in our study. As culti-

var specific parameters are not available in DSSAT model a same maturity cultivar was included in the DSSAT v4.5 cultivar database [11] for generation of genetic coefficient [20] . The generated cultivar specific parameters were within the range of DSSAT cultivar database. So, we can use the generated genetic coefficient in model application for Gazipur environment.

Figure 1 shows the performance evaluation of the model in terms of temporal course of leaf area index (LAI) during calibration process. Simulated LAI at different growth stages matched well with observed LAI although simulated LAI was slightly higher than the observed values. Per cent deviation was higher in the earlier stages of the crop growth, and subsequently the difference got narrowed down. The LAI was the highest at around 90 DAS thereafter it declined during 2014-15 season. A typical LAI pattern begins with a slow increase in the season, followed by a rapid increase of LAI until a maximum value is reached. LAI then declines as the leaves senesce and plants reaches physiologically maturity [21] .

Figure 2 Shows simulated and observed LAI during 2015-2016 seasons as process for the model’s validation, and performance seems to be satisfactory. In the beginning till anthesis, simulated values were slightly higher than the observed ones, but after anthesis the reverse trend was observed. However, the difference was marginal. Inter-varietal differences were not so significant throughout the growing periods. Figure 3 Shows relationship between simulated and observed maximum LAI value. The regression line was near to 1:1 line, indicating that the model was performing well under the test environment. The coefficient of determination (R²) was around 87% in case of trend line.

Table 4 shows the observed and simulated values for days to anthesis, days to maturity, grain yield and tops weight at harvest. In 2014-15 seasons, days to anthesis for BARI Hybrid Maize-7 and BARI Hybrid Maize-9 were 81 days, while that was 82 days for PIONEER and NK-40 and simulated values were 80 days for all the four cultivars, with error percentage of 1.23 to 2.44%. However, during 2015-2016 seasons, error percentage was a little bit higher, which ranged from 3.75% to 4.94%.

BHM7 = BARI Hybrid Maize-7, BHM9 = BARI Hybrid Maize-9; Obs = observed; Sim = simulated.

In 2014-15, days to physiological maturity for BARI developed maize varieties were 142 against 139 days of the simulated values. On the other hand, PIONEER and NK-40 took 143 days for physiological maturity against 140 days of the simulated values. The error percentage ranged from 1.45 to 2.11 across the years. There were little variation between observed and simulated grain yields of hybrid maize cultivars over the years, observed grain yield was a little bit higher compared to simulated yield. Similar trend was found in tops weight at∙harvest.

Table 5 shows the simulated and observed days to anthesis, days to maturity, grain yield and tops weight at∙harvest. During 2015-16 for validation, simulated days to anthesis was over estimated compared to simulated values irrespective of cultivars where error percentages ranged from 3.75% to 4.94%, but days to maturity was in good agreement between observe and simulated values. Grain yield was over estimated compared to simulated results, with error percentage ranged from 2.13% to 6.81%. Similar trend was found in tops weight at∙harvest with error percentage 2.86% to 4.43%. The estimation error for anthesis and physiological maturity appear reasonable and in line with results of other studies in which field data were used to estimate genetic coefficients [22] [23] [24] [25] .

Figure 4 shows relationship between simulated and observed biomass at harvest. Observed and simulated biomass matched well where coefficient of determination (R²) was around 97%, and thereby could demonstrate satisfactory performance of the model.

Figure 5 shows the performance of DSSAT model in terms of observed versus simulated grain yield. The regression line was more or less near to 1:1 line, indicating that the model was performing well under the test environment. Similarly, goodness of fit (R²) as well as regression coefficients between observed and si-

BHM7 = BARI Hybrid Maize-7, BHM9 = BARI Hybrid Maize-9; Obs = observed; Sim = simulated.

mulated data was significant. The coefficient of prediction was to the extent of 90% in case of trend run between the observed and simulated values.

CERES-Maize calibration and validation for major cultivars for Bangladesh production environment opens the way for use of the DSSAT model for inputs and resource management, yield forecasting and climate change impact analysis. The performance of the model after its calibration was satisfactory and the results were within significant limits and were similar to the results of Jones, and Thornton, 2003 and Ma, et al., 2006. The model, in the present study, was funder non-limiting conditions of water and nitrogen, which is the major production environment of maize crop in Bangladesh.