COSC2669 Case Studies in Computer Science
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Case Studies in Computer Science
WIL Project – Group 4
- Helping Farmers adapt to the impacts of Climate Change
2 Climate Change
- Climate change is starting to be felt all over the world.
- In the last year the heat waves in Europe and the floods in Pakistan.
- Climate change affected Farming and Agriculture.
- Victoria is not immune to its impacts
- We wanted to help farmers understand and adapt to the impacts of Climate Change.
3 BoM Data Source
- We looked at BoM data on rainfall and temperature for regions in Victoria
- We focused in in the Wimmera as this is a major agricultural area in our state.
- We summarised rainfall data from over 20 weather stations in the Wimmera
- The plot shows rainfall per decade and month since 1970
o Shown Darker:Earlier to Light:Later
- It shows 2 interesting features:
o Over the last 5 decades there’s been a reduction in rainfall
o Rainfall has occurred earlier in the year
- Note 2020’s only have 2 years so data is more distributed.
4 Farming Challenge and Adaptation – The Problem
We believe Farming and Agriculture will be particularly impacted by climate change
Many farmers learn how to management by personal experience, learning from their family, and using local traditions and practices. Farms are complex businesses to run.
Instead of relying on tradition or guesswork, we’re proposing a 2 stage Proof of Concept to assist farmers to account for the impact of climate change, and to help better select the varieties of crops to suite the new conditions
5 Proposal – A Potential Solution
We proposed to develop a model to help predict rainfall patterns for the near-term future, and then develop a crop variety recommender to use these predictions to assist farmers and agricultural planners select the best crop varieties for the new conditions.
To make the scope of the project achievable for this subject we focused on:
- Developing a predictive model based on LSTM to estimate the rainfall for a given weather station forward 1 to 2 years.
- Using the rainfall prediction and other farm specific data, a recommender to suggest the top 5 wheat varieties.
6 Focus on Wimmera Wheat
Using weather and rainfall data from BoM we extract rainfall, temperature, and cloud cover for all the weather stations in the Wimmera.
The Wimmera, one of Victoria’s most fertile and productive dry land farming areas, is in the west of Victoria. Its climate is semi-arid to sub-humid with annual rainfall from 380 to 580 mm, and with some areas exceeding 1500mm, with most rain falling in the winter.
It is the principal wheat producing area of Victoria.
7 Weather Forecast Model
A weather prediction model using BoM rainfall, temperature, and evaporation data for Victoria since 1970.
We then extracted weather data for stations in Victoria and allocated to the regions
Mallee, Wimmera, Southwest, North Central, North East, East, and Central
<Add chart showing Reduced rainfall over the last 5 decades)
- Using the Wimmera dataset, we showed:
o <Lu and others to add details on model and prediction>
We developed a model based on LSTM (Lu add details here). This took data from the last 70 years for a selected weather station, and attempts to predict the monthly rainfall for the next 12 months.
For weather station XYZ, the model gave the following results:
The model has accuracy of ???? <Show plot here of past data with model output>
8 Weather Forecast Model Results
For weather station XYZ, the model predicted the following results:
<Show plot of model prediction for net 12 months overlaid over long term trend.>
The model showed:
- X increase/decrease in expected rain compared to long term average
- Z movement earlier or later than long term average
9 Crop Recommender
There are many crops that can be planted in Victoria and each region has different weather patterns and ranges that impact the types and variates that would give the best results for a given region.
For this project we focused on Wheat as this is principle crop grown in the Wimmera.
Using data from Grain Research and Development Corporation (GRDC) 2022 Victorian Crop Sowing Guide, we developed a recommender that allows farmers to enter rainfall data, and other details and get a list of top 5 recommended varieties of wheat to sow.
Wheat variety selection is based on multiple factors including grain yield and quality, disease resistance, maturity, adaptation to the rainfall, elevation, temperature, soil type.
The system uses cosine similarity to order varieties based on the following attributes
o Expected Rainfall
§ High (> 500mm),
§ Medium (350 to 500mm), or
§ Low (< 350mm)
o Maturing Time
o Winter or Spring Sowing
- The recommender takes details of the expected values for a region produces an ordered list of varieties.
Below is a sample of using the recommender:
Crop details include:
Name, Type, Low Med or High Rainfall, Winter Sowing, Maturing Times, plus other details like Height, Coleoptile Length, Lodging, Sprouting, Head Type Colour, Head Type Awn, Soil Tolerance t Boron, and Soil Tolerance to Acid.
These can be included in the algorithm as required.
The project attempts to assist farms and agriculture planners to adapt to the impacts of climate change.
As demonstrated in the recent unprecedented 3rd La Niña weather event impacting the East Coast of Australia this year, predicting future rainfall can be difficulty task.
We’ve attempted to use standard models to predict local rainfall based on longer term trends, and provide a mechanism for farmers to make use of these and other predictions that may greatly benefit the productivity of their farms and of their regions.
To be truly impactful, the rainfall model needs to be expanded to broaden the number of sites it has been trained on to improve accuracy and scope across other regions in Victoria.
The model showed accuracy of ….. <Lu add details on models performance>
The crop recommender showed great promise in selecting and recommending the best varieties of wheat so farmers can optimise the productivity of their land.
The model can be easily expander to include different and more attributes to get a more tailored recommendation for a given location and farm. It could also easily be extended to include other corps although care needs to be taken to match specific crop attributes.