In our project, we developed a method of staying on step ahead of the pandemic effects in the agricultural sector. Due to the risks of COVID-19, the government was forced to restrict people’s movements including the transportation of goods, crops and other necessities. This led to a series of events which led to the disruption of the supply and demand chain. Despite having more productions of cereals than 2019[1], the GDP from this sector was low. [2] This was predominantly due to the miscalculated lockdown and movement restrictions, which stopped the cereals from reaching the customers. This, on one hand caused the financial loss for the famers and the government and on the other hand caused shortage of food for the general people. Our project deals with this problem before it even occurs allowing us to take sufficient risk management steps.

From this very project, we have developed a crop production indicator to predict the GDP loss in order to take early actions. During this pandemic situation where the mobility is a huge factor for any country, our indicator will predict the production considering the natural factors that affect its production. First the necessary weather variables wind, temperature, precipitation and humidity are taken to predict the weather for a certain period of time in the following year. Based on those parameters, we will be able to predict the crop production for that same period of time in the future. As a result, if we can predict the production beforehand, we can take necessary steps that will to prevent GDP loss that could occur due to overproduction or underproduction or if the lockdown is implemented due to Covid-19 restrictions.

The unemployment rate in Spain increased to 16.26% in the 3rd quarter of the 2020 compared to the 13.78% at the end of 2019. [3] We can plan ahead of engaging more manpower into agriculture when the production is expected to be high utilizing the people who have lost their jobs. If the production is more and depending on the mobility condition of the country, if there is lockdown throughout the country, we can take steps and precautions beforehand so that the produced crops aren't wasted due to long storage or inability to transport them. If the production is less compared to the demand, the investors and the government can invest in this sector or increase the budget accordingly. We hope to achieve a model that will predict the agricultural production beforehand and will help our farmers to work accordingly and help the government to protect its economy. 

Farmers, despite being the backbone of any nation, playing the most important role in keeping up with the food demands of the country are very much neglected when it comes to them receiving their deserved returns by selling the harvested crops. This is mostly seen in the under developed and developing nations of the world. Farmers are the group of people that have suffered the most during this pandemic. Despite producing more than the necessary harvest, they got very little in return due to strict mobility. This inspired us in coming up with an idea that gives the farmers a chance of getting what they deserve and the government an early warning to preserve its GDP.

Our approach has always been risk management. Data allows us to manage those risks even before they arrive. Spain and all the other countries in the world suffered because of the lockdown as the economic factor which is the contribution from different sectors like agriculture was not considered. We used a machine learning model to predict the crop production for the following year and thus allowing us to predict the potential loss due to an unplanned lockdown.

We used the temperature, wind speed, relative humidity and precipitation data to train a machine learning model predicting the harvested parcels for a specific day. We extracted the labels which are the measurements of the harvested parcels with the given dates. Using those dates we extracted the temperature, wind speed, relative humidity and precipitation of that specific date for building the model. Hence we got a dataset with labels which consisted of the feature vectors latitude, longitude, year, day of year, precipitation, wind speed, relative humidity, temperature. There were total 24 labelled data from each region. We then split the dataset into training and validation sets. As we have a time series data here, training set contained data from 19 May,2019 upto 5 July,2020 consisting of 20 data points for each region which were merged into one training set. The rest of the 4 data points for each region was merged to make the validation set. Training set had 220 data points and validation set had 44 data points. Training set was used to train a few regression models then the validation set was used to check these metrics (mean absolute error, mean squared error, root mean squared error and r2 score). XGBoost regression model performed the best having the lowest error metrics which was used to give the final prediction. We used python as the programming language. We used the scikit learn and pandas library as a tool to build the model. Using our model we predicted the harvested parcels from 1 January, 2021 up to 22 June, 2021 which showed the similar trend as 2020.The model was trained on a small amount of data which is its biggest pitfall. We are overfitting the data due to that. We couldn't predict the harvested parcel beyond 22 June, 2021 as we had no recorded data for the variables temperature, precipitation, wind speed and relative humidity. We can only make predictions if we have these variables which is another shortcoming of our approach. As a result, we used the prediction to cross check the trend the harvested parcels is following using the trend from previous year which it did match.

If provided with the necessary weather data from previous years, we can use it to train the LSTM model which would in return give the necessary values for the following years allowing us to predict the crop production using them. An LSTM based temperature and precipitation model [4] has been shown in the demo which can be modified accordingly for the other weather parameters.

As a team, it was an amazing experience to work with people from different backgrounds. Despite living in the same country, it brought four unknown people together and work with one goal in mind. There was a communication gap in the beginning which is understandable since we getting to know each other for the first time and everyone was thinking in their own way. But when we put our heads together, we came up with a pretty good idea that everyone was enthusiastic to be a part of. 

The data from the space agency has helped us in creating a complete picture of the Covid-19 impact on agriculture. From the EO dashboard, we pulled the data of harvest parcels of different regions of Spain under the harvesting evolution over time indicator. This helped us to compare the cereal harvest from the previous year which lead us to find the actual reason of impact in this sector.

Despite having a higher harvest, Trading Economics data[2] have showed us that GDP contribution from crops greatly decreased from mid-July. Upon monitoring the, mobility data from the EO dashboard and Google LLC "Google COVID-19 Community Mobility Reports [5] we realized the reason of fall in GDP. From here we came upon the idea of developing an indicator for predicting crop production and for this we needed the necessary weather variables which we obtained from the NASA Prediction of Worldwide Energy Resources. Correlating this data, with the ones we got from the EO dashboard we were able to develop our algorithm for our project. 

PowerPoint Demo Link: Team-Tech-Nerds.pptx - Google Drive

Our solution is not ready for integration into the EO Dashboard. But provided with enough data it will be possible. First and foremost, we need the temperature, humidity, wind speed and precipitation data from previous years in order to implement our LSTM algorithm and predict these parameters for a certain year. Using those parameters, we will be able to predict the crop production for the same year.

The link to our code which gives us the prediction of 2021 is given here.

The link to all the previous years' harvest parcels vs number of days graphs and the used and modified datasets are given here.

We’ve used the existing weather parameters and observed the predicted results to follow the trend of the previous years which shows that with enough training data we’d be able to bring even more accurate results for dates far into the future. We’d be using a model similar to the LSTM based temperature and precipitation prediction model for data refinement. 

1.  Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

Earth Observing Dashboard (eodashboard.org)

2.Spain GDP From Agriculture | 1995-2021 Data | 2022-2023 Forecast | Historical | Chart (tradingeconomics.com)

3.• Spain: unemployment rate 2005-2020 | Statista

4.(PDF) Prediction of Temperature and Rainfall in Bangladesh using Long Short Term Memory Recurrent Neural Networks (researchgate.net)

5.COVID-19 Community Mobility Reports (google.com)

Earth Observing Dashboard (eodashboard.org)

#agriculture #farmers #mobility #GDP #economic_impact #harvest #data_analysis #crop_production #covid19 #pandemic #weather_forecast

This project has been submitted for consideration during the Judging process.