Interview: Dr. C S Murthy, Director, Mahalanobis National Crop Forecast Centre
“Mainstreaming digital technologies into the agriculture value chain needs the participation of multiple stakeholders,” notes Dr. C S Murthy in this interview for AGI’s Newsletter on Digital Agriculture.
How does MNCFC use Remote Sensing and SAR data for crop assessments? Can they be used for all crops? If yes/no, why?
Satellite data of moderate spatial resolutions of both optical and microwave sensors are being analyzed extensively at MNCFC for crop mapping, crop health monitoring, and crop risk assessment. Analysis of microwave data acquired by Synthetic Aperture Radar (SAR) covers different crops. For many years, crop mapping with SAR data is being carried out for selected crops such as paddy and jute. This data is now being extended to other crops like soybean, maize, and red gram since the results are promising. SAR-based indices are being used for crop condition assessment for all crops by way of extracting these indices using the crop maps generated with optical indices. Thus, synergistic use of SAR and optical remote sensing data is followed at MNCFC.
How are crop models used for yield prediction? Have you developed your own crop models at MNCFC?
There are three widely followed models for generating yield estimates namely (a) Semi-physical, (b) Crop simulation, and (c) Machine Learning / Deep Learning. These crop models produce pre-harvest crop yield estimates, right now for major crops.
Semi-physical models are based on the biochemical process of plants involving light absorption for photosynthesis, radiation use efficiency, stress factors, accumulated biomass, and grain yield. It is recognized as a better approach than simple empirical modeling. Its strength lies in adopting a process-based framework with limited parameterization. Critical elements of this model: (a) precise information on crop variety, planting, and harvest dates, (b) empirical derivation of FAPAR with NDVI, and (c) derivation of water stress and temperature stress factors.
Crop growth models simulate the plant processes to estimate various bio-physical parameters and final crop yield. These models need intensive parameterization starting from genetic coefficients of crop variety under cultivation, crop sowing time, crop management practices – fertilizer applications, irrigation supplies, pest/disease occurrence, etc. When applied over larger geographies, these models tend to perform poorly due to limited parameterization and inadequate representation of varied crop-growing environments
Artificial Intelligence (AI) has gained importance in solving the non-linear relationships between variables. AI includes Machine Learning and Deep Learning models, such as the random forest (RF), Support Vector Machine (SVM), and different variants of neural network models (NN). In recent years, AI models are being increasingly used for crop yield estimation. Satellite-derived vegetation indices, meteorological data, hydrological variables, and edaphic factors are used as inputs in these models. Good quality data for training and validation is a prerequisite for the successful implementation of these models.
At MNCFC, crop yield estimation models are being implemented in collaboration with SAC and IMD under the FASAL project. Under crop insurance, pilot studies on crop yield estimation at the insurance unit level are being carried out through the AgriTech industry.
MNCFC is planning to initiate many collaborative studies with both academics and the industry to improve crop yield estimation in the country.
Apart from crop yield forecasts, MNCFC also has projects centered on drought assessment (NADAMS) and crop insurance (KISAN). Please throw light on the use of Geospatial technologies in these projects.
Monitoring and assessment of drought conditions and timely dissemination of information to stakeholders constitute the most vital part of the drought management system. Under the current drought monitoring project called NADAMS, several drought indicators such as rainfall deviations, rainy-days deviations, Moisture Adequacy Index, soil moisture index, and crop condition anomalies based on NDVI, and LSWI are generated at the sub-district level and provided to states, by MNCFC. These indices along with crop sown area data are used by States for declaring drought as per the guidelines of the National Drought Manual 2016.
Further, MNCFC generates several indices using weather and satellite datasets from the beginning of the season to support Crop Weather Watch review meetings of the Ministry. Our endeavor is to strengthen the crop monitoring system with all possible geospatial information products. These products are provided to all states taking part in Crop Weather Watch meetings. In Kharif 2022, the drought conditions prevailing in eastern India comprising parts of Uttar Pradesh, Jharkhand, Bihar, and West Bengal were closely tracked by MNCFC helping the Ministry and States with technology-based inputs.
What role can Multi-sensor (SAR & Multispectral/Hyperspectral) and Multi-resolution data integration play in agricultural analysis?
The synergistic use of multiple spectral indices is always beneficial for monitoring crops. Every spectral index has some strengths and limitations. Combined use of non-overlapping spectral indices leads to enhanced information content on crop status. In the early part of the season, NDVI is less sensitive because vegetation cover is not significant.
LSWI is more sensitive to surface wetness/dryness and hence it plays a better role than NDVI in the early part of the crop season. Similarly, SAR backscatter data shows higher sensitivity to surface dryness and wetness and hence can be used for drought detection early in the season. During the crop growth stage, backscatter, and optical indices like NDVI and LSWI together capture both surficial and structural parameters of the crop. Biophysical parameters like LAI, and FAPAR available at coarse resolution needs to be integrated with moderate resolution spectral indices to improve crop surveillance systems. Hyperspectral data provides the unique information content on crops and operational models using such datasets are yet to be evolved.
Are you using Data Science and Machine Learning for predictive analytics? If yes, please elaborate.
Remote sensing data of moderate resolutions around 10m is available free once in 5-10 days, thereby reducing the cost of surveillance over large areas. Similarly, mobile technology has tremendously improved the field data collection system by producing real-time crop status and crop management data.
ML/DL models are well-proven for analyzing large streams of data from multiple sources – satellites, weather stations, Mobile Apps, etc. in agriculture. These models are efficient to investigate the associations, establish relationships, and perform predictions on crop health and risk occurrence.
MNCFC is currently working towards realizing crop mapping and crop yield estimation using ML/DL models for major crops in the country. Smartphone-based crop surveillance using AI-based photo analytics is another important area for developing farmer advisory systems to be realized in the country. MNCFC has initiated collaborative efforts in this direction.
Lastly, what are some areas where MNCFC would like to explore partnerships/solutions from the Indian Geospatial industry?
There are enormous opportunities for developing and mainstreaming geospatial solutions in the agriculture value chain in the country. Spatial analytics, data mining, data engineering, evidence-based tools, etc., could churn out many useful information products. The benefits of these new technologies are yet to reach the farmers.
Many farmer-centric services such as weather advisory, pest/management, and market advisory could be improved to a larger extent by bringing new forms of knowledge and tools through digitalization and datafication. There is scope for developing new business models with these technologies to boost the country’s agricultural economy.
Therefore, the potential areas for collaborative efforts with the geospatial industry include (a) farmer-centric advisory services on weather and pest/disease management and market intelligence, (b) innovative crop insurance products for crop risk management, and (c) strengthening crop estimation surveys and crop surveillance systems.
Mainstreaming digital technologies into the agriculture value chain needs the participation of multiple stakeholders – farmers, input suppliers, traders, administrators, researchers representing private sectors, governments, and non-profit organizations. The benefits of technology are effectively reaped only when the institutional mechanism is in place.
Leave a Comment