I have LULC and ET data for the past three decades and have predicated LULC for future time period (e.g., 2025-2055). Now, I want to predict ET using land use land cover changes.
Dear Arti Soni,
You can visit the following link to get some idea regarding your query. It may help you.
Chapter Mapping and Assessment of Evapotranspiration over Different ...
https://iwaponline.com/h2open/article/4/1/92/81999/Detection-and-prediction-of-land-use-change-impact
https://iwaponline.com/h2open/article/4/1/92/81999/Detection-and-prediction-of-land-use-change-impact
Article The impact of land-use/land cover changes on water balance o...
Forecasting water components is a very difficult project. The water balance may be written as Q (flow) = P - ET - losses (such as deep seepage) within a hydrologic unit. As land use, land cover changes, both Q and ET can be influenced or altered in various ways. Changes in vegetation type, impervious surfaces, hydrologic modifications, soil compaction are all possible avenues of change. And forecasting the amount and timing of P (precipitation) is no easy task. If you can find information from hydrologic experiment stations within your physiographic area, some of them have evaluated the water balance differences for land cover types and/or land use. The Coweeta Hydrologic Experiment Station in NC, USA showed the Q and estimated ET differences in pine, hardwood and grass cover types, roughly 30, 20 and 10 inches, respectively. In the Francis Marion National Forest SC, USA, I used the forecast land use changes based on historic change, forest harvesting, impervious surfaces of roads and within urban areas, etc. within GIS format and making assumptions to estimate water yield changes due to various forest management scenarios. As I remember, our location did not forecast substantial shifts in rainfall from climate change, or shifts in vegetation types due to temperature shifts. The regular harvest and replanting forest did show increases in flow, as there are periods where ET is greatly reduced. Forest stands managed as savanna or woodlands with periodic burning had low density forest which also augmented flow through reduced ET. Forecasting involves many assumptions to be documented with estimates.
If poor land use management were to result in severe erosion and gullies, flow (runoff) is increased and ET because of less soil moisture and water table decline would be decreased. On the other hand, gullied or barren lands stabilized, reforested or revegetated would decrease runoff and increase ET.
LULC changes could give some general idea of potential ET trends, but lack any ability associated in forecasting the changes that could occur in rainfall and flow timing, or predict periods of drought or excess rainfall, flooding, etc. But I would also consider the existing data, how was it determined, as estimating ET for several decades is difficult at almost any scale.
You can employ machine learning algorithms such as support vector machines or neural networks for predictions. In deep non linear frameworks, the partial least squares regression model can make better predictions too...
I think, evapotranspiration (ET) is directly related to water body area, temperature, and wind speed. LULC can directly help to caculate water body area. LULC is one component, ET is one vector of many components. Thus, predicting ET with LULC is one issue of weak or partial correlation in statistics. You need to enhance the cost function of machine learning optimization or the set of statistics estimation criteria.
ET is composed of two main factors evaporation( water bodies and soil mositure) and transpiration (leave surface), so what i suggest to find some empircal or some formula to consider the effect of those two factors. for eg. you can relate NDVI for land coverage Aarti Soni
I think, evapotranspiration (ET) is significance process for defining the energy and mass relationship between other factors, such as soil, atmosphere and crops. ET data is based on the daily, monthly and annual climatic data. Several factors like climate and LULC change again affect the hydrological cycle. First of all you should prepare whole data then apply machine learning or deep learning.
Dear Aarti Soni,
I think you should predict using the NDWI, NDVI, and Soil Moisture and the LULC map as well as in situ data using SVM algorithms or neural networks.
Bests,
Ronak
Based on different empirical functions and methods, there is no where shows a direct linkage between ET and LULC but they can have significant or insignificant correlations depending on geomorphological and climatic characteristics of a research area. However, ET is usual a function of Temperature dynamics and variations in solar radiations. Please refer to different methods like Hargreaves and Samani or other methods and algorithms. But for prediction, I suggest you to try combination of climate data and LULC under machine learning algorithms as other researchers suggested.
Good luck