Using 3-hourly reanalysis or GLDAS temperature data, can we disaggregate daily observed minimum-maximum temperature into 3-hourly or 6-hourly time steps using statistical approach?
Here, the daily maximum and minimum temperature observations are made at various stations and interpolated into uniform grids. What I need is to mimic the daily temperature cycle at 3-hourly or 6-hourly timestep based on the available model simulated temperature data.
Dear Deepak,
You are asking an ill-posed question (see https://en.wikipedia.org/wiki/Well-posed_problem), in the sense that there are multiple possible answers and no clear, unique way to determine whether one solution is better than another. To proceed, you need to to constrain your problem further, until a solution can be identified. I'd like to suggest a two-pronged approach:
1. You need to understand fully how your daily temperatures were obtained, or what they represent. For instance,
- Are these instantaneous values acquired at a particular time of the day? If so, and say the readings were made at noon local time, then you know that most of your other artificial temperature values will be lower than this value, since you expect that the daily cycle of temperature peaks in early afternoon, at least during a nominal day, in the absence of strong weather perturbations within those 24 hours.
- Or are those daily mean temperatures? In that case, multiple records were probably originally collected during the day and then averaged into the value you now have. In that case, you know that whatever set of values you will "make up" should also average to that value, otherwise you will not respect the measurements.
2. You should then use a weather generator (see http://unfccc.int/adaptation/nairobi_work_programme/knowledge_resources_and_publications/items/5510.php) or some additional hypothesis to generate an artificial degree of variability during the day (in your case on a 3-hour time scale, or 8 data points per day, or 4 data points per day if you want to use a 6-hour interval) to construct a plausible set of values. Perhaps some climatological data for the same (or a nearby) location can impose some additional limit on your problem. In any case, depending on the nature of the data you have, your artificial time series should meet the same constraints as those used during the generation of your original data series.
In the absence of any information on the range of temperatures (maximum - minimum) actually experienced in the location of interest, or some other measure of daily variability, it will be difficult to simulate realistic values, of course. So the next critical question to address is to identify your accuracy requirement, which will depend on the ultimate application. And if you are mostly interested in statistical properties, you might consider generating a large class of scenarios of daily temperature changes, subject to some set of constraints, and study whether there are some similarities amongst those that may be helpful for your purpose.
If you could describe what you want to ultimately achieve and the highest level of uncertainty you can tolerate, it may be possible to provide further advice.
Best regards, Michel.
Dear Deepak,
You are asking an ill-posed question (see https://en.wikipedia.org/wiki/Well-posed_problem), in the sense that there are multiple possible answers and no clear, unique way to determine whether one solution is better than another. To proceed, you need to to constrain your problem further, until a solution can be identified. I'd like to suggest a two-pronged approach:
1. You need to understand fully how your daily temperatures were obtained, or what they represent. For instance,
- Are these instantaneous values acquired at a particular time of the day? If so, and say the readings were made at noon local time, then you know that most of your other artificial temperature values will be lower than this value, since you expect that the daily cycle of temperature peaks in early afternoon, at least during a nominal day, in the absence of strong weather perturbations within those 24 hours.
- Or are those daily mean temperatures? In that case, multiple records were probably originally collected during the day and then averaged into the value you now have. In that case, you know that whatever set of values you will "make up" should also average to that value, otherwise you will not respect the measurements.
2. You should then use a weather generator (see http://unfccc.int/adaptation/nairobi_work_programme/knowledge_resources_and_publications/items/5510.php) or some additional hypothesis to generate an artificial degree of variability during the day (in your case on a 3-hour time scale, or 8 data points per day, or 4 data points per day if you want to use a 6-hour interval) to construct a plausible set of values. Perhaps some climatological data for the same (or a nearby) location can impose some additional limit on your problem. In any case, depending on the nature of the data you have, your artificial time series should meet the same constraints as those used during the generation of your original data series.
In the absence of any information on the range of temperatures (maximum - minimum) actually experienced in the location of interest, or some other measure of daily variability, it will be difficult to simulate realistic values, of course. So the next critical question to address is to identify your accuracy requirement, which will depend on the ultimate application. And if you are mostly interested in statistical properties, you might consider generating a large class of scenarios of daily temperature changes, subject to some set of constraints, and study whether there are some similarities amongst those that may be helpful for your purpose.
If you could describe what you want to ultimately achieve and the highest level of uncertainty you can tolerate, it may be possible to provide further advice.
Best regards, Michel.
Dear Prof. Michel,
Thankyou for your elaborate reply. I would look into the weather generator.
I have daily gridded maximum and minimum temperature data. This gridded data is derived from temperature observations made at weather observatories. I mainly need this data at hourly time-step to run a land surface model mainly to simulate daily evapotranspiration and soil moisture. As I am not interested in hourly model output, I think a reasonable cycle of 3-hourly temperature would help more than a single daily value.
Here, since I don't have 3-hourly or 6-hourly temperature, I am using 3-hourly data from GLDAS to mimic the temperature cycle.
Thankyou
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis