6.5.1 Soil physics
6.5.1.1 Simulation of grain yield and water productivity of wheat under different tillage, residue and nitrogen management using Aqua Crop model
The evaluation of the FAO AquaCrop model with response to grain yield (GY) (under Conventional tillage (CT) and No Tillage (NT), two levels of residue mulch as subplot factor (Maize residue @ 5t/
ha (R+) and without residue (R0), and three levels of nitrogen as sub-sub plot factor (50, 100 and 150%
of the recommended dose of nitrogen) showed that the observed grain yield could account for 83.4%
variation in the simulated GY. The RMSE between the observed and simulated GY of wheat was 0.255 t/
ha, which account for 7.2% of the mean observed GY, which indicates excellent predictions of GY by the model. RMSEs and RMSEu were 0.193 and 0.197 t/ha, respectively. Higher RMSEu indicates that the error in the model predictions was less than the experimental error. The Wilmott d-index and CRM were 0.99 and 0.026, respectively. Higher d-index support better simulation of GY by the model. A positive value of CRM indicates the model under-predicted the GY of wheat. The evaluation of the Aqua Crop model with respect to final biomass yield (BY) of wheat indicate that the observed above ground biomass yield of wheat accounted for 82.9% variation in the simulated biomass yield. The RMSE between the observed and simulated biomass yield of wheat was 1.40 t/ha, which accounted for 14.14% of mean observed biomass yield.
This indicates good prediction of biomass yield by the model. The RMSEs and RMSEu were 1.335 and 0.423 t/
ha, respectively. Higher RMSEs indicate that the error in model prediction of biomass was higher than that of experimental error. The Wilmott d-index and CRM were 0.97 and 0.120, respectively. The positive value of CRM indicates under-prediction of the biomass yield of wheat by the model. Evaluation of the model with respective to water productivity (WP) of wheat
showed that the model could account 71.5% variation in the observed WP of wheat. The RMSE between observed and simulated WP of wheat was 0.118 kg/
m3, which accounted for 8.45% of the mean observed WP of wheat. The nRMSE value indicates excellent agreement between the observed and simulated WP of wheat. RMSEs and RMSEn were 0.0114 and 0.031 kg/
m3, respectively. The higher RMSEs indicate the error in the model prediction was more than the experimental error. The d-index and CRM were 0.97 and -0.006, respectively. The negative value of CRM indicates that the model over predicts the WP of wheat. Therefore, AquaCrop model ver 6.1, which requires relatively less input parameters, can be used satisfactorily for optimizing different management practices like tillage, residue and nitrogen management practices under different soil and weather condition for improving yield of wheat.
6.5.1.2 Prediction of soil hydraulic conductivity through ML and AI
Saturated hydraulic conductivity (SHC) is an important soil hydraulic parameter that determines the rate of water flow systems through soil. As direct measurements of SHC in field conditions is very difficult, laboor-intensive, time-consuming, and expensive, artificial intelligence (AI) based on artificial neural network (ANN) and support vector machine (SVM) models and multi-linear regression (MLR) model were used to obtain soil hydraulic conductivity from easily measurable soil parameters viz.,particle size distribution, bulk density (BD) and organic carbon (OC). MLR model showed that SHC is negatively correlated with clay and silt while it is positively correlated with sand (%). Results indicated that ANN with 2 number of hidden layer performed best both for training and testing dataset. In prediction of SHC, for training datasets, the root mean square error (RMSE) value of SVM model was found to be lowered by 17 and 6.4% as compared to MLR and ANN, respectively.
SVM resulted similar trend in testing dataset, where RMSE value was lowered by 9.11 and 5.7% as compared to MLR and ANN. Correlation value (r) was found to be greater for SVM (0.77) in training data sets while
for testing data sets, r was highest in case of MLR (r
= 0.80). Results indicated that SVM could account the more complex nature of inputs and output variables.
Comparative performance evaluation of MLR and ANN in prediction of SWCFC
Models Datasets Performance criteria
MAE r RMSE MAPE
MLR Training 2.46 0.64 2.97 40.29
Testing 3.11 0.80 3.84 43.47
ANN Training 2.09 0.73 2.62 38.7
Testing 2.90 0.78 3.89 57.2
SVM Training 1.85 0.77 2.45 31.2
Testing 2.92 0.73 3.49 41.6
6.5.2 Bio-Physics
6.5.2.1 Modeling the temporal distribution of water, ammonium-N, and nitrate-N in the root zone of wheat using HYDRUS-2D under conservation agriculture
The temporal distribution of both soil water and soil NO3-N under different conservation agriculture (CA) practices during the wheat crop growth were characterized by HYDRUS-2D model. Treatments comprised of conventional tillage (CT), permanent
broad beds (PBB), zero tillage (ZT), PBB with residue (PBB+R) and ZT with residue (ZT+R). Hydraulic inputs of the model, comprising the measured value of Kfs and α and n, obtained as the output of Rosetta Lite model were optimized through inverse modeling.
Model predicted the daily change in soil water content (SWC) of the profile during the simulated period (62- 91 DAS) with good accuracy (R2 = 0.75; RMSE= 0.038).
In general, soil water balance simulated from the model showed 50% lower cumulative drainage, 50%
higher cumulative transpiration along with higher soil water retention, in PBB+R than CT. Reported values of the first-order rate constants, signifying nitrification of urea to NH4(µa) (d−1), nitrification of NH4-N to NO3-N (µn) (d−1) and the distribution coefficient of urea (Kd - in cm3 mg−1) were optimized through inverse modeling and were used as solute transport and reaction input parameters of the model, which predicted the daily change in NO3-N of the profile with better accuracy (R2 = 0.83; RMSE = 4.62). Since NH4-N disappears fast, it could not be measured frequently. Therefore, not enough data could be generated for their use in the calibration and validation of the model. Results of simulation of daily NO3-N concentration indicated a higher concentration of NO3-N in the surface layer and its leaching losses beyond the root zone were relatively lesser in PBB+R, than CT, which resulted in
Pictorial presentation of NO3–N under CT and PBB+R on different days during the simulation period
less contamination of the below ground water. Thus the study clearly, recommended PBB+R to be adopted for wheat cultivation in maize-wheat cropping system, as it enhances the water and nitrogen availability in the root zone and reduce their losses beyond the root zone.
6.5.3 Remote sensing and GIS
6.5.3.1 A prototype for spatial wheat yield forecasting system
A reliable crop yield forecast system is an imperative for stabilized food security. The study attempted to develop a novel regional wheat yield forecasting system by assimilating remote sensing derived LAI and weather forecast into crop simulation model, i.e. InfoCrop-wheat, using minimum observations as model inputs. The CSM model was calibrated and validated using experiments at research farm of IARI, as well as the model was validated for 42 farmers’ fields selected in Pataudi block of Haryana during rabi season 2015-16 and 2016-17 and showed good performance of the model at both the scales. The developed forecasting framework consisted of four components, viz., (i) retrieve LAI from multi-spectral remote sensing images, (ii) assimilate LAI into modified InfoCrop model, (iii) incorporate bias corrected WRF modelled weather forecast and (iv) computer coded prototype system for spatial implementation. LAI was retrieved through inversion of PROSAIL RTM from Sentinel 2A MSI and Landsat-8 OLI imageries and validated using in-situ LAI measurements of farmers’
fields and also assessed the effect of atmospheric correction algorithms, inversion approaches and image resolutions on LAI retrieval. Among the two atmospheric correction algorithms, MODTRAN outperformed libRadtran, while among the inversion approaches, Look-Up-Table outperformed ANN. The inclusion of additional two red-edge bands as available in MSI significantly reduced the uncertainly in LAI retrievals over that obtained by using six bands, while inclusion of only additional VNIR band did not show any significant effect on LAI retrievals. MODTRAN and LUT based inversion reduces average error in LAI retrieval to 0.44 from 1.19. Then, we developed the
novel modified InfoCrop-LAI assimilation framework through successful implementation of Ensemble Kalman filter and Forcing algorithm of multiple LAI assimilations with crop phenology adjustment.
This study demonstrated that assimilation of LAI through EnkF improved not only crop yield prediction performance but also phenology and growth of wheat using standard management inputs and minimum actual observations. Finally, we demonstrated the wheat growth and yield forecasting system assimilating LAI through ensemble kalman filter and bias corrected weather forecast from dynamical WRF model into InfoCrop-wheat model for a region. The workable system has shown the acceptable accuracy in forecasting phenology, total dry matter and yield of spring wheat at fine scale and minimized the large management input data requirements. It has potential to be adopted for actual applications in many national projects like FASAL and PMFBY of Govt. of India.
6.5.4 Agricultural meteorology
6.5.4.1 Multi stage wheat yield estimation using weather-based models
Wheat yield data and weather variable during crop growing period (46th to 15th SMW) for last 35 years data were collected from Hisar, Ludhiana, Amritsar, Patiala and IARI, New Delhi. Stepwise multiple linear regression (SMLR), Principal component analysis in combination with SMLR, Artificial Neural Network (ANN) alone and in combination with principal components analysis (PCA), Least absolute shrinkage and selection operator (LASSO) and elastic net (ENET) analysis are carried out by fixing 70% of the data for calibration and remaining dataset for validation.
Results showed that out of six multivariate models lasso and elastic net are excellent for four out of five station and good for one station, because of prevention in over fitting and reducing regression coefficient by penalisation. nRMSEv range between 5.0 to 15.9%
for Elastic Net, 4.2 to 16.9 for LASSO, 3.7 to 20.0% for SMLR, 6.2 to 15.9% for PCA-SMLR, 7.8 to 22.3% ANN and 11 to 16.1% PCA-ANN, respectively.
Multistage wheat yield prediction was done at tillering, flowering and grain filling of crop by considering 46th to 4th, 46th to 8th and 46th to 11th SMW for model development. On examining these multivariate models for stage-wise prediction of wheat yield, percentage deviation was found between -0.1 to 25.6, 0.9 to 22.8, -0.7 to 22.5% during tillering, flowering, and grain filling stage respectively. On the basis of percentage deviation of estimated yield by observed yield, prediction accuracy at different growth stage was found better by Elastic net and LASSO model followed by SMLR model. PCA-SMLR, ANN and PCA-ANN model were giving least prediction accuracy in all phenological stage. From this study it may be concluded that LASSO, Elastic Net and SMLR model based on weather parameters can be used for district level yield forecast at different crop growth stage of the crop.
6.5.4.2 Simulation of biomass and seed yield mustard cultivars using AquaCrop model
AquaCrop model (v 6.1) was calibrated for above ground biomass accumulation and seed yield using 4 types parametric values, namely, conservative and non-conservative crop parameters, soil parameters and management parameters. Conservative crop parameters were obtained from the published literature, non-conservative crop, soil and management parameters were generated using the field experiment data of earlier years of the same field for three mustard cultivars namely Pusa Vijay, Pusa Mustard 21 and Pusa Bold in the model data files. AquaCrop Model (v 6.1) was then run with the daily weather data of rabi season
of 2013-14 for aboveground biomass calibration of three mustard cultivars: Pusa Vijay, Pusa Mustard-21 and Pusa Bold. The calibrated final biomass was 12.70, 11.05 and 9.81 t/ha, respectively. The difference in observed and simulated final biomass was +1.35, +1.80 and -1.6%, respectively. So, the final biomass calibration was within ± 10 % difference. After running the calibrated AquaCrop Model with the weather data of rabi season 2013-14, the simulated seed yield was obtained for mustard cultivars, Pusa Vijay, Pusa Mustard-21 and Pusa Bold. The differences between the simulated and observed seed yield were found to be -9.20, +3.32 and +2.47% (i.e. within ± 10 %)
6.5.4.3 Estimate surface energy fluxes using BREB (Bowen Ratio Energy Balance) method
Bowen ratio energy balance (BREB) method is a micrometeorological method by combining Bowen ratio with energy balance components of earth. In this study, a field experiment was conducted on maize (variety: PMH-1) and wheat (variety: HD 2967) in the main experimental farm of ICAR during kharif, 2018 and rabi 2018-19. Homogeneous crop was grown in both the season and micrometeorological tower was installed inside the crop field. Micrometeorological tower is having temperature, humidity, wind speed in five levels (0.5 , 1, 2, 4, and 8m), net radiometer at 2m height, PAR sensor at 2m height, wind vane at 2m height, soil moisture and soil temperature at three different depths (5, 15, 25 cm) and soil heat flux plate at 2 depths (5 and 15 cm). During kharif 2018, sowing was done on 19th July. Cultivar PMH-1 emerged in 8-10 DAS and matured at 105 DAS. Wheat crop was grown Observed and calibrated final biomass and yield of three mustard cultivars during rabi 2013-14
Cultivar Biomass (t/ha) Seed yield (t/ha)
Observed Calibrated Difference (%) Observed Calibrated Difference (%)
V-1 11.2 12.71 +1.35 2.50 2.27 - 9.20
V-2 10.85 11.05 +1.80 2.11 2.18 3.32
V-3 9.97 9.81 -1.6 1.90 1.95 2.47
in rabi 2018-19. Crop was sown on 25th November, 2018.
Cultivar HD 2967 germinated in 7-8 days. CRI stage was appeared in 22 DAS. Harvesting was done at 142 DAS. Highest LAI of maize was 4.56 at 82 DAS with a CV varied from 5.64 to 16.05 % and CI varied from 0.15 to 0.29. Mean grain yield of maize crop was recorded 5.76 t ha-1 during kharif, 2018. During rabi season 2018- 19, wheat grain yield varied from 4.11- 4.42 t ha-1. The mean grain yield was observed 4.25 t ha-1.
Trend of energy balance and Bowen ratio for (a) a cloudy and (b) cloud free day over wheat crop during rabi, 2018-19
6.5.4.4 Weather based Agromet Advisory
Agro-met advisory bulletins are being prepared in Hindi as well as in English on every Tuesday and Friday based on past weather, real time weather and weather forecast received for next five days from IMD, New Delhi. The bulletin is passed on to the farmers
through SMS / telephone / E-mail .The bulletins are also sent to ATIC, KVK Shikohpur, KVK Ujawa, IKSL, NGO, e-choupal, KrishiDarsan, All India Radio, DD Kisan, and local Hindi newspapers through E-mail for wider dissemination among farmers. These advisories were uploaded on the Institute website (www.iari.res.in) along with daily weather data and medium range weather forecast. These advisories were also uploaded on the IMD website (www.imdagrimet.gov.in) and farmer portal (http://farmer.gov.in ) in both Hindi and English. These advisories along with crop status are sent to IMD, Pune for preparation of national bulletins.
During 2019, total 105 agro-advisory bulletins were prepared in Hindi as well as in English. SMS were sent to the farmers through m- Kisan portal. Weather forecast and agromet advisory bulletin is fruitful for farmers, through this they can select high yielding varieties of different crop and vegetables, other farming practices such as sowing, weeding, irrigation, fertilizer, pesticides spray (time and doses) can be done at right time. Feedback received from the farmers from different villages of NCR Delhi showed that agromet advisory bulletin is useful since as it helps in reducing cost of cultivation, saving of input resources and increases in net profit.