ARTIFICIAL INTELLIGENCE IN THE INTRODUCTION OF COVER CROPS IN AFRICA TO REGENERATE THE SOIL

Aime Christian Tuyishime Author

08/06/2021 Added

19 Plays

Description

This research project was conducted to introduce the practice of cover crops in Africa, using Artificial Intelligence to evaluate the performance of cover crops on African soils by predicting the expected biomass. A tool was created to help farmers predict the specific amount of cover crop biomass they might expect in their field.

Searchable Transcript

Search:

[00:00:01.140]Hello everyone. My name is Aime Christian Tuyishime,
[00:00:03.750]and I'm going to share with you about my project of introducing cover crops in
[00:00:07.110]Africa using Artificial Intelligence.
[00:00:09.600]I conducted this project under the supervision of Dr. Andrea Basche,
[00:00:13.380]in the Department of Agronomy and Horticulture.
[00:00:17.130]During this presentation,
[00:00:18.450]I'm going to talk about the case study of the current state of the agriculture
[00:00:21.150]sector in Rwanda,
[00:00:22.290]the benefits of practicing cover crops, an introduction on crop yield
[00:00:25.650]predicting models, methods used in this research,
[00:00:28.830]the results and the potential of the model I created.
[00:00:33.390]The agriculture sector in Rwanda plays an important part in the country's economy.
[00:00:37.410]It represents 39% of the country's GDP, 80% of employments,
[00:00:42.210]90% of local food consumption, and 63% of foreign exchange
[00:00:47.070]earnings. However, the agricultural sector faces many challenges,
[00:00:50.760]including more than 50% of acidic soils,
[00:00:53.820]35 to 246 tons/ha per year,
[00:00:57.540]Soil loss from erosion, low water holding capacity, and low
[00:01:02.580]organic matter in the soil.
[00:01:05.430]Practicing cover crops brings many benefits to the soil.
[00:01:09.330]Those includes: increased soil nutrients,
[00:01:11.700]organic matter, water holding capacity, and water infiltration.
[00:01:16.350]All these arise from the crop residues of cover crops,
[00:01:19.860]which helps improve the soil structure.
[00:01:22.380]Cover crops also reduces erosion by curtailing the soil,
[00:01:26.160]which reduces the rain drop impact on the soil surface.
[00:01:29.700]It reduces weeds by bringing competition to them and reduces pests by hosting
[00:01:34.590]natural controls, which preys on them.
[00:01:37.320]It also reduces the soil compaction by using roots, which breaks up the soil.
[00:01:43.470]I had an amazing and educative journey while conducting this research.
[00:01:48.240]It was divided into two phases where the first one focused on Rwanda and the
[00:01:52.710]second scaled phase one findings to Africa. In phase one,
[00:01:57.060]I conducted a research on the current progress of cover crops in Rwanda,
[00:02:00.870]which turns out to be not a common practice in case it would benefit highly to
[00:02:05.190]the Rwandan soils. As I'm far from Rwanda right now,
[00:02:09.240]doing a field research was impossible for me to do.
[00:02:13.140]I then did a research on crop yield prediction models.
[00:02:16.800]Where I first researched on the process-based models. After reviewing many
[00:02:20.940]research papers,
[00:02:22.080]I found that process-based models were not the right path for me to take.
[00:02:26.760]They take more time, money, and also require skills to operate.
[00:02:31.620]On the other hand, data-based models like Artificial Intelligence are fast,
[00:02:36.150]cheap, and easy to scale.
[00:02:38.190]They only require programming skills which I had been developing.
[00:02:41.700]Hence this was the best path for me to take.
[00:02:44.850]The next step was finding data to train my Artificial Intelligence Model.
[00:02:48.690]where I started manually developing the dataset. However,
[00:02:52.230]I quickly realized that it would require me reviewing many thousands of research
[00:02:56.550]papers. I then started researching for an already built a dataset,
[00:03:01.210]which I found and will describe later. I then trained my AI model
[00:03:06.070]using Random Forest, a Machine Learning Algorithm that I would describe in
[00:03:10.480]detail on the next slide.
[00:03:12.550]I used the model to predict the cover crop biomass in Rwanda and developed a map
[00:03:16.690]using ArcGIS to visualize the findings. For the
[00:03:21.640]second phase,
[00:03:22.480]A literature review was done on machine learning with the aim of doing hyper
[00:03:26.590]parameter tuning and optimization of the model.
[00:03:29.440]I then conducted a search for African agriculture datasets,
[00:03:32.980]which were used to make cover crop biomass predictions and developed maps to
[00:03:37.180]illustrate the results.
[00:03:39.400]As the goal of this project is develop a tool for farmers to use in predicting
[00:03:43.930]the cover crop biomass they expect from their fields,
[00:03:47.110]I had developed a prototype of the user interface, which is on the last slide.
[00:03:54.040]Here are the datasets that were used in the project.
[00:03:57.190]The first dataset that was used in testing and training is titled: "A global
[00:04:01.540]experimental data set for assessing grain legume productions."
[00:04:05.110]The second and the third datasets were used for prediction.
[00:04:08.950]There were titled: "A global data set for crop production under conventional tillage
[00:04:13.510]and no tillage systems" and "Conservation agriculture in Sub-Saharan Africa,
[00:04:18.520]crop yields from experiments,"
[00:04:20.350]respectively. As shown here is the formula of Random Forest,
[00:04:25.690]the algorithm that have been used to train and predict the cover crop biomass.
[00:04:29.830]The general understanding of it is that it classifies different data points
[00:04:34.390]based on how similar and correlated they are,
[00:04:37.060]Hence making predictions based on that.
[00:04:39.850]Their main parameters that have been used in the model were crop, climate, and
[00:04:44.560]soil data. After training the model and testing it,
[00:04:48.940]the R-Squared was 0.82 with a root mean square error
[00:04:53.740]of 1.29 and a standard deviation of 3.01.
[00:04:59.200]The prediction of biomass in Rwanda had the mean of 2.89 tons/ha
[00:05:03.520]which is a really good amount of biomass. It had,
[00:05:06.820]the maximum of 5.2 tons/ha,
[00:05:10.540]with a minimum of 1.66 tons/ha.
[00:05:15.790]As shown on the left, the Artificial Intelligence model had a really strong
[00:05:19.480]coefficient of correlation between the predicted crop biomass aerial and the actual
[00:05:23.560]crop biomass aerial. On the right, is the map showing the predicted cover crop
[00:05:27.670]biomass in Rwanda. As you can see,
[00:05:30.040]the Northern park seems to have higher crop biomass prediction compared to the
[00:05:34.570]Southern part. This can be explained by the difference in elevation,
[00:05:38.980]as the Northern part is higher in elevation compared to the Southern part.
[00:05:43.900]The difference also can be explained by the difference in soil fertility, as
[00:05:48.790]the Northern part is near volcanos, hence potential for soil nutrients in the soil.
[00:05:55.180]Cover crop biomass prediction was done for 116 sites in 20 different countries.
[00:06:00.680]The results of the prediction show huge potential of cover crops as shown by the
[00:06:05.240]second map, which was the average of fields per country.
[00:06:09.560]There two ratings were good ranging from 2-4 tons/ha,
[00:06:13.640]and very good with the range from 4.1 to 6 tons/ha
[00:06:18.230]of cover crop biomass.
[00:06:22.400]On the left is the user interface prototype.
[00:06:25.310]The app will help every from around the globe to predict the amount of biomass
[00:06:29.720]they might expect before planting cover crops.
[00:06:32.750]This reduces the uncertainty they might have on practicing cover crops and
[00:06:37.430]protects them from any losses they might encounter from blindly practicing
[00:06:42.170]them. The app is fast and simple to use.
[00:06:46.010]It is cheap as it open-source and free, with a user-friendly interface.
[00:06:51.470]It also has global scalability and applicability,
[00:06:54.950]as I used global datasets to train and test it.
[00:06:58.700]The app is projected to be publicly launched by the end of 2021.
[00:07:04.220]Thank you so much for your time, and I want to thank my supervisor, Dr.
[00:07:08.540]Andrea Basche for the professional advice and guidance throughout this process.
[00:07:13.580]Thank you so much.

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

ARTIFICIAL INTELLIGENCE IN THE INTRODUCTION OF COVER CROPS IN AFRICA TO REGENERATE THE SOIL

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments