Artificial Intelligence in the Introduction of Cover Crops

Aime Christian Tuyishime Author

04/04/2021 Added

38 Plays

Description

This research project was conducted to introduce the practice of cover crops in Rwanda, using Artificial Intelligence to evaluate the performance of cover crops on Rwandan soils by predicting the expected biomass.

Searchable Transcript

Search:

[00:00:02.680]Hello everyone.
[00:00:03.570]My name is Aime Christian Tuyishime
[00:00:05.500]and I'm going to share with you about my project
[00:00:08.410]of introducing cover crops in Rwanda
[00:00:10.520]using artificial intelligence.
[00:00:12.850]And I conducted this project under
[00:00:14.850]the supervision of Dr. Andrea Basche
[00:00:17.350]in the department of agronomy and horticulture.
[00:00:22.780]During this presentation,
[00:00:24.000]I'm going to talk about the current state
[00:00:25.580]of the agriculture sector in Rwanda,
[00:00:27.520]the benefits of practicing cover crops,
[00:00:30.160]and introductional crop yield predicting models and methods
[00:00:33.450]used in this research, the results, lessons,
[00:00:37.130]and next steps.
[00:00:39.640]The agriculture sector plays an important part
[00:00:41.960]in the country's economy.
[00:00:44.130]It represent 39% of the country's GDP,
[00:00:48.240]80% of employments,
[00:00:50.440]90% of local food consumption,
[00:00:53.030]and 63% of foreign exchange earnings.
[00:00:58.430]However, the agriculture sector faces many challenges
[00:01:02.460]including more than 50% of acidic soils,
[00:01:07.500]35 to 246 tons hectare per year of soil loss from erosion,
[00:01:13.510]low water holding capacities,
[00:01:15.510]and low organic matter in the soil.
[00:01:20.490]Practicing cover crops brings many benefits to the soil.
[00:01:24.620]Those includes increased soil nutrients, organic matter,
[00:01:28.820]water holding capacity, and water infiltration.
[00:01:32.200]All this arises from the crop residuals of cover crops,
[00:01:35.770]which helps improve the soil structure.
[00:01:39.670]Cover crops also reduces erosion by curtailing the soil,
[00:01:43.450]which reduces the rain drop impact on the soil surface.
[00:01:46.870]It reduces weeds by bringing competition to them.
[00:01:50.150]It reduces pests by hosting natural controls
[00:01:52.710]which preys on them, and reduces the soil compaction
[00:01:56.160]by using roots which breaks up the soil.
[00:02:01.320]I had an amazing and educative journey
[00:02:03.800]while conducting this research.
[00:02:06.290]First, I conducted the research on the current progress
[00:02:09.450]of cover crops in Rwanda,
[00:02:11.130]which turns out to be not a common practice,
[00:02:14.240]in case it would benefit highly to the Rwandan soils.
[00:02:21.570]As I'm far from Rwanda right now,
[00:02:23.640]doing the field research was impossible for me to do.
[00:02:27.560]I then did the research on crop yield prediction models
[00:02:31.010]where I first researched on the process based models.
[00:02:36.090]After reviewing many research papers,
[00:02:38.650]I actually found that process-based models
[00:02:41.020]were not the right path for me to take.
[00:02:43.970]They take more time, money,
[00:02:46.430]and also require skills to operate.
[00:02:51.040]On the other hand, data based models
[00:02:53.010]like artificial intelligence are fast and cheap
[00:02:56.310]and easy to scale.
[00:02:57.980]They only require programming skills,
[00:02:59.750]which I've been developing.
[00:03:01.700]Hence, this was the best path for me to take.
[00:03:05.920]The next step was finding data
[00:03:07.330]to train my artificial intelligence model,
[00:03:09.860]where I started manually developing the data set.
[00:03:13.100]However, I actually found that would require me
[00:03:15.770]reviewing thousands of research papers.
[00:03:20.090]I then started searching for an already built dataset
[00:03:23.820]which I found and will describe in detail
[00:03:26.840]on the next slide.
[00:03:29.900]I then trained my AI model using Random Forest,
[00:03:33.620]a machine learning algorithm that I will describe in detail
[00:03:36.790]on the next slide.
[00:03:39.980]I used the model to predict cover crop biomass in Rwanda
[00:03:43.410]and developed a map using ArcGIS to visualize the findings.
[00:03:52.450]Here is the information about the dataset that was used
[00:03:55.570]to have started a global experimental dataset
[00:03:57.810]for assessing grain and legume production.
[00:04:00.110]And the main contents of the dataset was site data,
[00:04:03.400]crop data, tillage data, irrigation data,
[00:04:07.470]fertilization data, weeds and insects data,
[00:04:10.630]which were all gathered from 173 articles.
[00:04:18.050]As shown, here is the formula of Random Forest,
[00:04:20.820]the algorithm that have been used
[00:04:22.470]to train and predict biomass in Rwanda.
[00:04:26.010]The general understanding of it
[00:04:27.810]is that it classifies different data points
[00:04:30.830]and based on the ones which are similar
[00:04:34.060]to make predictions on the data that have been given.
[00:04:39.200]The main parameters that have been used in the model
[00:04:41.810]were crop, climate, soil data.
[00:04:44.580]After training the model and testing it,
[00:04:47.560]the RSQ was 0.82, which was a good number
[00:04:51.340]with a root mean squared error of 1.29
[00:04:54.291]and the standard deviation of 3.01.
[00:04:57.850]The prediction of biomass in Rwanda
[00:04:59.820]had the mean of 2.89 tons per hectare,
[00:05:02.680]which is a really good amount of biomass.
[00:05:05.110]It had the maximum of 5.72 tons per hectare,
[00:05:08.700]with a minimum of 1.66 tons per hectare.
[00:05:14.360]As shown on the left, the artificial intelligence model
[00:05:17.400]had a really strong coefficient of correlation
[00:05:19.830]between the predicted crop biomass aerial
[00:05:22.270]and the actual crop biomass aerial.
[00:05:26.300]On the right is the map
[00:05:27.610]showing the predicted cover crop biomass in Rwanda.
[00:05:30.520]As you can see, the northern part seems to have
[00:05:33.740]high crop biomass prediction compared to the southern part.
[00:05:37.330]This can be explained by the difference in elevation,
[00:05:40.650]as the northern part is high in elevation
[00:05:43.440]compared to the southern part.
[00:05:47.470]Artificial intelligence has huge potential in agriculture
[00:05:50.550]due to how fast, cheap, and easy to scale it is.
[00:05:56.640]I'm going scale my research to the whole Sub-Saharan Africa
[00:05:59.630]in this summer of 2021 where we will predict
[00:06:02.630]cover crop biomass in countries like Kenya,
[00:06:05.930]Tanzania, Uganda, Burundi, and Congo.
[00:06:11.660]The last step will be developing a user-friendly interface
[00:06:15.040]that will be used by farmers
[00:06:16.650]to predict the specific biomass
[00:06:19.080]they expect from their fields.
[00:06:23.420]Field application is important
[00:06:26.050]to make sure there is no bias in the model
[00:06:28.600]and measure the accuracy of the model in the field.
[00:06:33.610]Thank you so much for your time.
[00:06:35.380]And I want to thank my supervisor, Dr. Andrea Basche
[00:06:38.640]for the professional advice and guidance
[00:06:41.240]throughout this process.
[00:06:42.830]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

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments