Home Articles Uprooting Uncertainty – UAVs for agricultural insurance sector

Uprooting Uncertainty – UAVs for agricultural insurance sector

Sudhakar Manda, Chief - Remote Sensing and GISSudhakar Manda, Chief – Remote Sensing and GIS, Skymet Weather Services, reveals how using geospatial technologies lead to more precise crop yield estimates

Are unmanned aerial vehicles (UAVs) or drones a game-changer for the agricultural insurance sector?

The insurance sector has been starving for detailed, correct and reliable data to take decisions. Though satellites have been there providing a wide range of spectral and spatial information, which is useful for different crop parameters, satellite data does not provide enough granularity when it comes to answering specific questions to settle losses — whether a person buying a crop insurance product has actually sown that particular crop or not, whether the farmer has sown the crop in the exact hectares of area he has purchased the policy for, whether lodging has affected 30% of his crop field or 75%, etc. Solutions based on drones, when suitably crafted, can definitely prove to be a reliable, precise estimation tool for losses if there have been any.

Another major advantage of UAVs is that they can be flown at will and low altitudes, i.e., below the clouds. Since monsoon is the most important cropping season and a significant part of the season goes off without any optical satellite data due to lack of cloud penetration, UAVs come as the only solution to gather imagery data.

How conducive is the policy environment in India for using UAVs for commercial purposes?

The existing policies need far-reaching changes in order to make the commercial usage of UAVs a practical alternative. There are several government agencies, such as, Directorate General of Civil Aviation, Home Ministry, Ministry of Defense, Survey of India, etc., who need to be approached for getting approval for flying. This takes time, which is a major hindrance for the industry. While most other applications can wait indefinitely, it is the agriculture sector that needs the approvals to be processed within a given framework of few days or weeks — as the window to capture crop information is very limited. We understand the concerns of the government and are certainly not in favor of leaving the skies open for anyone and everyone; but at the same time, a controlled environment needs to be worked out where legitimate drone users can be allowed to operate.

Image taken from a drone

How are you using UAVs for crop insurance?

We have successfully implemented crop yield estimation solutions by taking crucial data from UAV images as replacement to field inputs, and combining it with weather and other datasets for crop simulation models. Finally, the effort is towards arriving at farm field level yield estimates and to tell whether a particular farmer is likely to suffer or has actually suffered losses or not. Apart from that, we derive very accurate acreage/undisputable estimates for fraud detection, connecting the farmer’s data to cadastral information and a host of other new datasets like soil health card, etc., to improve upon precision.

What new weather forecast technologies are you utilizing?

We use latest state-of-the-art global dynamical models to arrive at short-, medium- and long-term forecasts. These dynamical models are tuned for Indian region by incorporating latest terrain data and proper physics scheme. Additionally, we apply statistical corrections based on our own network derived weather parameters. Further the forecast is corrected by group of synopticians. One of the key areas where we feel that significant improvement is needed is the rain gauging network. Because of its very nature, rainfall is often very localized and variable in both intensity and distribution. We would be keen to see setting up of Doppler radar network by private parties to be permitted in our country that will improve the baseline for data and forecast multi-folds.

What challenges do you face in your workflow in India?

The cropping system in India is quite varied, and even within a single cropping season (like rabi), the sowing dates vary. At times, double sowing of short duration varieties or re-sowing in failed crop areas is quite common. This becomes very problematic when seed companies claim that if all sold seeds are sown, the acreage should be of certain hectares. However, when satellite image classification is performed over the peak growing season, the acreage turns out to be quite low. This creates a lot of confusion when total production arrivals to markets are estimated and prices for the produce are determined for that season.

Parcel overlay

Therefore, it became necessary for us to identify the source of difference in these two numbers. We decided to go for a multi-date classification of crops right from November, when all fields were in harvested state, and pursue it till April which brought forth some unique observations and to final satisfaction of client and explained why the numbers are different.

How have Hexagon Geospatial solutions helped you solve this problem?

We used ERDAS software in general, and ERDAS Modeler in the later stages, to develop algorithms that we were able to segregate not only different crops, but also three different phases viz:

• Early sown areas: Sown in late October/early November, and harvested in late January/early February.
• Regular sown areas: Sown in late November/early December, and harvested in March, which is the normal crop duration.
• Late sown areas: The early sown areas were re-sown in February and harvested in April-May.

In total, six to seven images were analyzed together to assess the presence or absence of crops on ground. This was achieved through stacking of multi-date NDVI outputs and grouping them into early/mid and late sown pixels. This could be easily done in ERDAS Modeler and results were to our satisfaction.

Has Hexagon Geospatial’s solution improved your business processes?

Because of Hexagon Geospatial’s solution, we could accomplish this feat as Modeler allows flexibility to define our own algorithms and modify them as required. The regular classifications are useful in crop classification of one date, and multi-date imagery analysis for sowing/harvesting trends, etc., is possible too.