MIT Solve

Solution overview

Our Solution

YieldCast: Crop Planner with Satellite Image

Tagline

Intelligent forecaster of yield

Pitch us on your solution

Although agriculture employs around 41% of labour force of Bangladesh, a huge mismatch between demand and supply is ever prevailing in this sector. This causes a deleterious price fluctuation and thus negatively affects the farmers as well as the consumers of foods.

We will develop an evolving and open yield forecasting platform that will timely forecast yield taking care of as many data points as available. To address this problem, we are creating a Machine Learning platform based on satellite images, meteorological and historical weather data. Primarily, we will focus on the most common staple food of Bangladesh - rice.

The results will be most beneficial for the farmers. However,the government, wholesalers as well as any other stakeholders in the food supply chain can be the consumer of this. The output of our complex system will be web and mobile based application with different plans for different stakeholders.

What is the problem you are solving?

Agriculture plays a pivotal role in the growth and stability of Bangladesh economy. More than three-quarters of total rural population derive their livelihood from agricultural sector. However, around 60% of farmers are functionally landless and depend on sharecropping on others’ lands. The average farm size is too small to support a family adequately. To add to these problems, the farmers are not getting fair prices. Besides, neither the farmers nor the government has proper forecasting data beforehand to plan accordingly. The basic problems are

Government buys very little portion of paddy from farmers. Although the primary goal of government procurement programme is to give “price support” to farmers, which is not benefiting them much.
Price offered to the farmers by third parties is non-commensurate to the farmers’ production costs.

Farmers destroying their crops, instead of selling with the offered unfair price, are getting more common these days in order to show protest. For example, farmers recently blocked a highway in rice-rich Rangpur by spreading paddy grains on the road – demanding the government to purchase paddy directly from farms, and ensure fair price. Even some farmers set their paddy fields on fire in protest against the low prices.

Who are you serving?

Our platform will serve the mass people of Bangladesh, especially farmers, researchers and the Government. Bangladesh's agriculture is highly weather-dependent and the whole harvest can be wiped out in a matter of hours when cyclones strike.

With the help of our platform, the growth status and the productivity can be predicted. And the prediction will be more accurate as the harvest time comes closer. The results will be driven integrating the meteorological data, therefore the farmers can be alerted well ahead of time to take necessary steps to mitigate any natural upcoming disasters.

Therefore, the government can directly plan if import is required or if Bangladesh can export fulfilling own needs and can also plan the direct procurement from the farmers. The reports (fully/partial) will be provided to the farmers through different local offices of government or NGOs and will be open to mass public. Therefore, the whole crop supply chain will remain transparent. This will ensure fair price to the farmers and proper market price and availability to the mass people. It will not only help the farmers and government but also ensure a positive impact on the local economy and food security of mass people as well.

What is your solution?

We want to build a platform using high resolution satellite image and machine learning that will allow us to ingest, process, and analyze massive data. We started researching what could be done with satellite and weather data for agricultural development of Bangladesh.

During the growing season, we can analyze the spectral signatures for paddy to give us information about crop health and productivity. We will combine this information with detailed meteorological data and knowledge of how weather variables affect the production of rice during the growing season to build our forecast.

We have an expert team, containing specialists of Artificial Intelligence, Natural Language Processing and Image Processing, researching on this. They have initially proposed three steps for the task:

Step 1: Identify area of Interest and extract Normalized Difference Vegetation Index (NDVI)

Figuring out where specific crops are being grown in a growing season using satellite image presents many unique challenges. Unlike roads and buildings, a crop’s appearance changes throughout the year posing more difficulties to identify specific crop types.

We’ll generate a time-series of NDVI layers with a variety of land cover classes including water, cropped area, and various vegetation classes of a particular zone. NDVI is an indicator of vegetative growth — low NDVI values corresponding to a lack of green vegetation (e.g. a field immediately after planting) and high ones to greener.

Step 2: NDVI time-series clustering

Conventional dissimilarity measures do not perform well for clustering time-series data. For example, in Euclidian distance, two time-series of same class (i.e 2 Sine curves with distinct transformations) can have a large Euclidian distance between them. This would present problems when measuring distance between two paddy fields planted at different dates. Dynamic Time Warping finds the optimal non-linear alignment between two curves, resulting in less pessimistic distance measures between transformed curves of the same class. The resulting distance matrix can be used as an input to various unsupervised learning algorithms like k-means. After clustering, we will examine various characteristics of the resulting groups, including timing of peak NDVI, and the apparent planting and harvest dates, and attempt to label them as specific crop classes based on the cropping calendar.

Step 3: Supervised classification

Once we are able to confidently label one or more clusters, we can use the labeled pixels as locations for generating a training data set for a supervised classification model for our platform.

Select only the most relevant.

Increase opportunities for people - especially those traditionally left behind – to access digital and 21st century skills, meet employer demands, and access the jobs of today and tomorrow

Where our solution team is headquartered or located:

Dhaka, Bangladesh

In which sector would you categorize your solution?

Agriculture

Our solution's stage of development:

Concept