The kidney bean simulation module in InfoCrop v2.1 generic model, used in this study, was developed at the Environmental Modelling lab of the Centre for Environment Science and Climate Resilient Agriculture, ICAR-Indian Agricultural Research Institute (Naresh Kumar et al., unpublished). The model was designed to simulate the growth, development and yield of kidney bean in response to environmental factors such as soil, weather and management practices. The InfoCrop V2.1 generic model has the following crop growth and soil processes;
- Crop growth: Phenology, radiation use, leaf area growth and senescence, dry matter production and partitioning, source and sink balance, nitrogen uptake and partitioning and effects of nitrogen, water, temperature and CO2 on growth processes.
- Soil water flow: Root water uptake; water soil sub-surface movement, soil inter-layer movement, drainage, evaporation and runoff.
- Soil nitrogen dynamics: Mineralization, fixation, soil inter-layer movement, nitrification, denitrification, volatilization and leaching.
- Soil carbon dynamics: Mineralization and immobilization of soil organic matter.
- Environmental impact: Temperature, water, flooding, nitrogen stresses
The input data for calibration of the model included data on daily weather, soil characteristics, varietal coefficients and crop management (Table 1).
2.2. Model calibration and validation
For calibration of the model, a field experiment was conducted in January-April 2020, in the research farm of ICAR-Indian Agricultural Research Institute, New Delhi-110012, India. The experiment aimed to investigate the interactive effect of different irrigation and temperature regimes on the growth and yield attributes of kidney bean. The best performing treatment i.e. kidney bean grown in seasonal mean temperature of 24.2℃ with irrigation provided at sowing, vegetative and pod-filling phase was selected for calibration. Additionally, the model was also calibrated for other treatments i.e. kidney bean grown in seasonal mean temperature of 19.6℃ with irrigation provided at sowing and vegetative phase. To set up the model, preliminary varietal coefficients were obtained from published literature. To calibrate the model, data on soil characteristics; crop phenology, time-series LAI, total dry matter and its partitioning and seed yield; crop management practices and daily weather for selected treatments were used as model input. To simulate proper phenology, leaf area index (LAI), dry matter and seed yield, several iterations were done to closely match the observed with the simulated values.
The model was evaluated using statistical indicators i.e. root mean squared error (RMSE) (Fox, 1981); agreement index (AI) and model efficiency (ME) (Wallach et al., 2006). The formulae for these indicators are given below.
Where n is the number of samples, Si and Oi are the simulated and observed values respectively, O̅ is the mean of the observed data, S̅ is the mean of the simulated data. RMSE estimates the root mean square deviation between simulated and observed values. A higher RMSE value indicates low accuracy of the simulation. AI measures the agreement of the simulated values with observed values in terms of trend. Model efficiency measures the model’s ability to match observed values.
The model had high accuracy in simulating the phenology of the crop in terms of days to 50% first flowering, days to physiological maturity with an error in the range of 2-5 days. The LAI was simulated with an R2 of 0.81 and with a high AI of 0.93. Model efficiency in simulating LAI was (R2 =0.7) with an acceptable level of RMSE (0.55). Additionally, the model could closely simulate the dry matter production at different stages of crop growth as well as seed yield.
After satisfactory calibration of the kidney bean model, validation was done for remaining treatments along with data from published literature for other locations in India (Parbani, Pune and Anand) and Kenya (Eldoret, Machakos and Kabete). These treatments included different management conditions such as varieties, sowing dates, irrigation and nitrogen levels, apart from variations in soil and climate. A total of 87 treatments (Table 2) were used and the model was validated for phenology, total dry matter and seed yield.
The model performed well in simulating phenology, total dry matter and seed yield across different experiments (Fig 1). Statistical indicators for various parameters were as follows: days to 50% flowering (RMSE: 2 days, AI: 0.98, R2: 0.94, ME: 0.93); days to 50% physiological maturity (RMSE: 3 days, AI: 0.99, R2: 0.94 and ME: 0.99); total dry matter (RMSE: 487 kg ha-1, AI: 0.99, R2: 0.76 and ME: 0.96) and seed yield (RMSE:223 kg ha-1, AI: 0.95, R2: 0.8 and ME: 0.81). A high agreement index for all growth and yield parameters indicates the model’s ability to capture the trend of the effects of different factors on the growth and development of kidney bean.
2.3. Climate change impact assessment
Climate change impact analysis was carried out for rainfed kidney bean production. Simulations were done for ‘Kharif’ (July-September) season in India. However, Kenya experiences a bimodal rainfall regime distinguished as long rains season i.e. March-May (MAM) and short rains season i.e. October-December (OND). Thus, analysis was done for both seasons. High yielding, commonly grown, reference varieties were considered for the analysis i.e. Varun and KAT X56 for India and Kenya respectively. Selected locations in India were: Dahod (74.26°N 22.84°E), Nilgiri (76.5°N 11.42°E), Chickmagalur (75.77°N 13.32°E), Baghpat (77.22°N 28.95°E), Darjeeling (88.26°N 27.04°E), Patna (85.14°N 25.59°E), Jabalpur (79.97°N 23.19°E), Pune (73.86°N 18.52°E), Palani (77.52°N 10.45°E) and Amritsar (74.88°N 31.62°E) and in Kenya viz., Eldoret (0.53°N 35.28°E), Embu (-0.5°N 37.45°E), Kakamega (0.27°N 34.75°E), Kisii (-0.68°N 34.78°E), Kitale (1 °N 34.98°E), Machakos (-1.58°N 37.23°E), Makindu (-2.28°N 37.83°E), Meru (0.08°N 37.65°E), Nakuru (-0.27°N 36.1°E) and Nyeri (-0.43°N 36.97°E). These locations represented major kidney bean growing areas in India and Kenya. Common crop management practices followed by farmers along with general soil characteristics of the selected locations were taken into consideration.
2.3.1. Processing of baseline climate data and future climate projections
Location-wise weather data i.e. daily rainfall, maximum and minimum temperatures were required for simulating climate change impacts. Based on the availability of high-quality data, observed weather data were collected for the baseline period 1976-2005 (India) and 1982-2005 (Kenya) for each of the selected locations. These data were sourced from Indian Meteorological Department (IMD) and Kenya Meteorological Department (KMD).
Data on future climate scenarios (2010-2099) were obtained from the Coordinated Regional Downscaling Experiment (CORDEX) data portal. Two regional climate models (RCMs), common for India and Kenya were selected (Table 3). This study considered two representative concentration pathways (RCPs) i.e. RCP 4.5 and RCP 8.5 to represent mid and high-level emission pathways, respectively. RCM data were bias-corrected using the scaling method and ensembles were created to run simulations for different climate change scenarios i.e. 2020 (2010-2039), 2050 (2040-2069) and 2080 (2070-2099). For the baseline period, CO2 concentration was set as 360 ppm, while under RCP 4.5 projected values of 422 ppm (2020 scenario), 495 ppm (2050 scenario) & 532 ppm (2080 scenario) were used. Under RCP 8.5, projected CO2 concentration of 432 ppm (2020), 572 ppm (2050) and 799 ppm (2080) were used.
2.3.2. Estimating yield for baseline period and future climate scenarios
Data on daily weather, sowing time (according to season), varietal coefficients of reference variety, soil characteristics, crop management and CO2 concentration were input into the model to simulate kidney bean seed yield for baseline period i.e. 1976-2005 (India) and 1982-2005 (Kenya). The mean of the yields from these years was taken as ‘baseline yield’. Similarly, simulations were done for climate change scenarios (2020, 2050 and 2080) under different RCPs. The mean of 30 years yield under each scenario for each RCP was taken as ‘mean yield in scenario’.
Climate change impact on seed yield, expressed as per cent change from mean baseline yield, was calculated as: